size of train loader is: 90 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.4070, 0.8938, 0.4183, 0.3538, 0.4567, 0.6175, 0.5400], [0.6197, 0.3986, 0.8800, 0.4617, 0.4188, 0.4783, 0.5687, 0.5550], [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672], [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5988, 0.5283], [0.6346, 0.4165, 0.9138, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471], [0.6325, 0.4165, 0.9000, 0.4617, 0.3813, 0.4900, 0.7485, 0.5447], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[-0.0584, 0.2115, -0.0577, -0.0264, 0.6124, -0.3884, 0.2870, 0.4973], [-0.0186, 0.2038, -0.0694, -0.0212, 0.5656, -0.3732, 0.2786, 0.4981], [-0.0844, 0.2453, -0.0447, -0.0736, 0.6318, -0.3736, 0.2809, 0.5259], [-0.0908, 0.2592, -0.0879, -0.0383, 0.5870, -0.3843, 0.3103, 0.4775], [-0.0545, 0.2420, -0.0500, -0.0258, 0.6162, -0.3611, 0.2940, 0.5109], [-0.0442, 0.2109, -0.0649, -0.0483, 0.5915, -0.3324, 0.2890, 0.5157], [-0.0514, 0.2363, -0.0713, -0.0051, 0.6292, -0.3114, 0.2788, 0.4927], [-0.0822, 0.2308, -0.0656, -0.0090, 0.6174, -0.3896, 0.2823, 0.5259]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6200, 0.4070, 0.8938, 0.4183, 0.3537, 0.4567, 0.6175, 0.5400], [0.6197, 0.3986, 0.8800, 0.4617, 0.4187, 0.4783, 0.5688, 0.5550], [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672], [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5987, 0.5283], [0.6346, 0.4165, 0.9137, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471], [0.6325, 0.4165, 0.9000, 0.4617, 0.3812, 0.4900, 0.7485, 0.5447], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383]]], device='cuda:0') loss_train_step before backward: tensor(0.2998, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.2998, device='cuda:0', grad_fn=) loss_train: 0.2997697591781616 step: 1 running loss: 0.2997697591781616 Train Steps: 1/90 Loss: 0.2998 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483], [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050], [0.6200, 0.3993, 0.8519, 0.4923, 0.3962, 0.4717, 0.6013, 0.5433], [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.1958, 0.3088, 0.2890, 0.1878, 0.4821, -0.0499, 0.4648, 0.5882], [ 0.2268, 0.3071, 0.2562, 0.1235, 0.4270, -0.0543, 0.4471, 0.5042], [ 0.2150, 0.3346, 0.2953, 0.1354, 0.4465, -0.0666, 0.4839, 0.5662], [ 0.2461, 0.2619, 0.2592, 0.1226, 0.4609, -0.0660, 0.4270, 0.4831], [ 0.1998, 0.3029, 0.2570, 0.1213, 0.4600, -0.0628, 0.4580, 0.5487], [ 0.2079, 0.2926, 0.2190, 0.1779, 0.4371, -0.0531, 0.4744, 0.5373], [ 0.2588, 0.2846, 0.2885, 0.1383, 0.4450, -0.0381, 0.4251, 0.5413], [ 0.1834, 0.2973, 0.3002, 0.1576, 0.4539, -0.0788, 0.4260, 0.5624]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483], [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050], [0.6200, 0.3993, 0.8519, 0.4923, 0.3963, 0.4717, 0.6012, 0.5433], [0.6165, 0.4106, 0.7575, 0.1733, 0.3837, 0.2650, 0.5680, 0.5116]]], device='cuda:0') loss_train_step before backward: tensor(0.1120, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.1120, device='cuda:0', grad_fn=) loss_train: 0.4117617905139923 step: 2 running loss: 0.20588089525699615 Train Steps: 2/90 Loss: 0.2059 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913], [0.6059, 0.4002, 0.7562, 0.2767, 0.3538, 0.3033, 0.5529, 0.5455], [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4055, 0.3739, 0.5440, 0.2942, 0.3808, 0.2197, 0.5719, 0.5390], [0.4301, 0.4093, 0.5301, 0.2674, 0.3338, 0.2265, 0.5864, 0.5637], [0.3896, 0.3546, 0.4938, 0.2991, 0.3735, 0.1760, 0.5180, 0.5417], [0.3896, 0.3833, 0.5534, 0.3136, 0.3577, 0.1936, 0.5783, 0.5552], [0.4145, 0.4122, 0.5246, 0.2921, 0.3729, 0.2434, 0.5558, 0.5490], [0.3705, 0.3855, 0.5324, 0.3042, 0.3832, 0.1809, 0.5555, 0.5525], [0.3739, 0.3511, 0.5211, 0.2681, 0.3678, 0.1772, 0.5384, 0.5119], [0.3903, 0.3680, 0.5187, 0.2795, 0.3428, 0.1927, 0.5294, 0.5423]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6133, 0.4094, 0.8495, 0.4028, 0.3587, 0.3200, 0.5003, 0.5407], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6286, 0.4078, 0.8062, 0.2267, 0.4787, 0.1533, 0.5953, 0.4913], [0.6059, 0.4002, 0.7563, 0.2767, 0.3537, 0.3033, 0.5529, 0.5455], [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5638, 0.5633], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6229, 0.4107, 0.8138, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.0000, 0.0000, 0.7613, 0.3250, 0.4038, 0.2533, 0.5437, 0.5767]]], device='cuda:0') loss_train_step before backward: tensor(0.0267, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0267, device='cuda:0', grad_fn=) loss_train: 0.43846603110432625 step: 3 running loss: 0.1461553437014421 Train Steps: 3/90 Loss: 0.1462 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433], [0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6177, 0.4086, 0.8738, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700], [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450], [0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700], [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6063, 0.5617]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5931, 0.4201, 0.7860, 0.3801, 0.3459, 0.3958, 0.5810, 0.5520], [0.5350, 0.4044, 0.8002, 0.3776, 0.3409, 0.3038, 0.6013, 0.5565], [0.5821, 0.4141, 0.8026, 0.3910, 0.3213, 0.3766, 0.6106, 0.5469], [0.5523, 0.3895, 0.7907, 0.4043, 0.3298, 0.3388, 0.6085, 0.5748], [0.5915, 0.4425, 0.8049, 0.4194, 0.3449, 0.4237, 0.6408, 0.5606], [0.6064, 0.4165, 0.7978, 0.3757, 0.3294, 0.4055, 0.6155, 0.5291], [0.5497, 0.3888, 0.7902, 0.3943, 0.3422, 0.3726, 0.6113, 0.5332], [0.5783, 0.4244, 0.8106, 0.3968, 0.2945, 0.3802, 0.6230, 0.5543]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433], [0.6230, 0.4152, 0.7588, 0.2283, 0.4013, 0.2883, 0.6200, 0.5767], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6075, 0.4000, 0.8512, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6177, 0.4085, 0.8737, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700], [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450], [0.6212, 0.4171, 0.7875, 0.3633, 0.3812, 0.2933, 0.5675, 0.5700], [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6062, 0.5617]]], device='cuda:0') loss_train_step before backward: tensor(0.0052, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0052, device='cuda:0', grad_fn=) loss_train: 0.4436452961526811 step: 4 running loss: 0.11091132403817028 Train Steps: 4/90 Loss: 0.1109 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6115, 0.3998, 0.7063, 0.2383, 0.4037, 0.1950, 0.5320, 0.4993], [0.6160, 0.4093, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609], [0.6353, 0.4128, 0.9138, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6595, 0.4149, 0.9635, 0.4405, 0.3608, 0.5043, 0.6468, 0.5533], [0.6573, 0.4416, 0.9378, 0.4602, 0.3851, 0.4643, 0.6446, 0.5442], [0.6465, 0.3929, 0.9284, 0.3990, 0.3773, 0.3864, 0.6122, 0.5066], [0.6654, 0.4107, 0.9472, 0.4188, 0.3765, 0.4518, 0.6117, 0.5490], [0.6871, 0.4540, 0.9970, 0.4859, 0.3264, 0.5708, 0.6297, 0.5511], [0.6877, 0.4320, 0.9677, 0.5030, 0.3692, 0.5202, 0.6485, 0.5374], [0.6752, 0.4174, 1.0047, 0.5007, 0.3579, 0.5584, 0.6603, 0.5398], [0.6805, 0.4097, 0.9533, 0.4593, 0.3685, 0.4398, 0.6234, 0.5440]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6229, 0.4107, 0.8138, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6115, 0.3998, 0.7063, 0.2383, 0.4038, 0.1950, 0.5320, 0.4993], [0.6160, 0.4092, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6064, 0.4019, 0.8650, 0.4517, 0.4038, 0.5367, 0.5703, 0.5609], [0.6353, 0.4128, 0.9137, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991]]], device='cuda:0') loss_train_step before backward: tensor(0.0123, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0123, device='cuda:0', grad_fn=) loss_train: 0.455948734190315 step: 5 running loss: 0.091189746838063 Train Steps: 5/90 Loss: 0.0912 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183], [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617], [0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300], [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7639, 0.4666, 1.0737, 0.5197, 0.4200, 0.6167, 0.6571, 0.5344], [0.7446, 0.4824, 1.0979, 0.4950, 0.4316, 0.6142, 0.6310, 0.5616], [0.7546, 0.4313, 1.0392, 0.4391, 0.4339, 0.5028, 0.6431, 0.5494], [0.7280, 0.4014, 1.0350, 0.3954, 0.4080, 0.4104, 0.6010, 0.5530], [0.7477, 0.4200, 1.0253, 0.4783, 0.4805, 0.4765, 0.5870, 0.5082], [0.7686, 0.4448, 1.0690, 0.4399, 0.4344, 0.5180, 0.6374, 0.5681], [0.7443, 0.4067, 1.0424, 0.4304, 0.4043, 0.5326, 0.6042, 0.5293], [0.7320, 0.4131, 1.0113, 0.4355, 0.3841, 0.4534, 0.6299, 0.5128]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6111, 0.3995, 0.8788, 0.4567, 0.3812, 0.4833, 0.5450, 0.5700], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.6204, 0.4049, 0.7975, 0.2700, 0.3938, 0.2567, 0.5700, 0.5183], [0.6154, 0.4112, 0.7038, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617], [0.6207, 0.4081, 0.7663, 0.2067, 0.3963, 0.3200, 0.6313, 0.5300], [0.6200, 0.4071, 0.7337, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517]]], device='cuda:0') loss_train_step before backward: tensor(0.0178, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0178, device='cuda:0', grad_fn=) loss_train: 0.4737188364379108 step: 6 running loss: 0.07895313940631847 Train Steps: 6/90 Loss: 0.0790 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6350, 0.4043, 0.8738, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950], [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869], [0.6131, 0.4037, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680], [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124], [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6138, 0.5883], [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133], [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7992, 0.4712, 1.0351, 0.5252, 0.4813, 0.6162, 0.6746, 0.5770], [0.7076, 0.4036, 0.9749, 0.3386, 0.4551, 0.3386, 0.5951, 0.5499], [0.7864, 0.4010, 1.0021, 0.3990, 0.4437, 0.4824, 0.5861, 0.5256], [0.6992, 0.3757, 0.9816, 0.3626, 0.4514, 0.2998, 0.5664, 0.5662], [0.8054, 0.4612, 1.0290, 0.4557, 0.4737, 0.5697, 0.6536, 0.5601], [0.7407, 0.4508, 0.9957, 0.4088, 0.4786, 0.4903, 0.6357, 0.5672], [0.7307, 0.4101, 1.0154, 0.4157, 0.4688, 0.3993, 0.5921, 0.5474], [0.7518, 0.3969, 1.0725, 0.4365, 0.4336, 0.5652, 0.6162, 0.5677]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6350, 0.4043, 0.8737, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950], [0.6276, 0.4095, 0.8238, 0.2250, 0.4663, 0.1783, 0.6171, 0.4869], [0.6131, 0.4036, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680], [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124], [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6137, 0.5883], [0.6266, 0.4065, 0.8313, 0.2467, 0.4787, 0.1733, 0.6313, 0.5133], [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575]]], device='cuda:0') loss_train_step before backward: tensor(0.0135, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0135, device='cuda:0', grad_fn=) loss_train: 0.48724996810778975 step: 7 running loss: 0.06960713830111283 Train Steps: 7/90 Loss: 0.0696 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667], [0.6246, 0.4028, 0.8738, 0.4867, 0.4088, 0.5667, 0.6362, 0.5200], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7125, 0.4141, 0.9276, 0.3407, 0.4545, 0.3371, 0.6088, 0.5133], [0.7724, 0.4321, 0.9407, 0.3951, 0.4934, 0.5374, 0.6366, 0.5840], [0.7000, 0.4080, 0.9320, 0.3456, 0.4463, 0.3606, 0.5961, 0.5657], [0.7373, 0.4212, 0.9778, 0.3996, 0.4907, 0.4384, 0.6298, 0.5426], [0.6694, 0.3810, 0.9105, 0.2660, 0.4332, 0.2168, 0.5700, 0.5416], [0.7486, 0.4425, 0.9655, 0.4365, 0.4873, 0.4980, 0.6413, 0.5653], [0.6991, 0.4299, 0.9115, 0.3282, 0.5000, 0.3444, 0.6179, 0.5303], [0.7601, 0.4259, 0.9607, 0.4066, 0.4690, 0.5100, 0.6555, 0.5402]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6134, 0.4090, 0.6926, 0.2819, 0.3537, 0.3233, 0.5562, 0.5667], [0.6246, 0.4028, 0.8737, 0.4867, 0.4087, 0.5667, 0.6363, 0.5200], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [0.0000, 0.0000, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100]]], device='cuda:0') loss_train_step before backward: tensor(0.0159, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0159, device='cuda:0', grad_fn=) loss_train: 0.503159002866596 step: 8 running loss: 0.0628948753583245 Train Steps: 8/90 Loss: 0.0629 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166], [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926], [0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494], [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600], [0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748], [0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6111, 0.4033, 0.8300, 0.3267, 0.3588, 0.3333, 0.5444, 0.5637]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6310, 0.4255, 0.8500, 0.3970, 0.4594, 0.4231, 0.6480, 0.5211], [0.5883, 0.3990, 0.8303, 0.3005, 0.4513, 0.3195, 0.6211, 0.5154], [0.6312, 0.3964, 0.8382, 0.3922, 0.4255, 0.4006, 0.6227, 0.4975], [0.6345, 0.4232, 0.8789, 0.3901, 0.4579, 0.4217, 0.6088, 0.5509], [0.5915, 0.3512, 0.8164, 0.3049, 0.4660, 0.2282, 0.5638, 0.5399], [0.6027, 0.3788, 0.8326, 0.2586, 0.4733, 0.2217, 0.6128, 0.5120], [0.6407, 0.4505, 0.8766, 0.4017, 0.4244, 0.4355, 0.6682, 0.5646], [0.6337, 0.3537, 0.8414, 0.3109, 0.4235, 0.2654, 0.6164, 0.5033]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166], [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926], [0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494], [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600], [0.6329, 0.4196, 0.9237, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748], [0.6230, 0.4152, 0.7588, 0.2283, 0.4013, 0.2883, 0.6200, 0.5767], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6111, 0.4033, 0.8300, 0.3267, 0.3587, 0.3333, 0.5444, 0.5637]]], device='cuda:0') loss_train_step before backward: tensor(0.0041, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0041, device='cuda:0', grad_fn=) loss_train: 0.5072252913378179 step: 9 running loss: 0.05635836570420199 Train Steps: 9/90 Loss: 0.0564 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183], [0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397], [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [ nan, nan, 0.8850, 0.3000, 0.5363, 0.2250, 0.7343, 0.5771]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5854, 0.4194, 0.7982, 0.4484, 0.3842, 0.4339, 0.6975, 0.4881], [0.5636, 0.3870, 0.7519, 0.3707, 0.4280, 0.3838, 0.6772, 0.5203], [0.5463, 0.3677, 0.7634, 0.3431, 0.3938, 0.2582, 0.6202, 0.5056], [0.5107, 0.3751, 0.7688, 0.3725, 0.3725, 0.3147, 0.6341, 0.5228], [0.5468, 0.3651, 0.7660, 0.3440, 0.3912, 0.2587, 0.6207, 0.5004], [0.5300, 0.3686, 0.7474, 0.3380, 0.3827, 0.3088, 0.6002, 0.5400], [0.5577, 0.3624, 0.7734, 0.3745, 0.3982, 0.3396, 0.6820, 0.4983], [0.5084, 0.3917, 0.7298, 0.2798, 0.4158, 0.2044, 0.6265, 0.4828]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.6251, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6087, 0.5183], [0.6127, 0.4084, 0.8737, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397], [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.0000, 0.0000, 0.8850, 0.3000, 0.5362, 0.2250, 0.7343, 0.5771]]], device='cuda:0') loss_train_step before backward: tensor(0.0133, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0133, device='cuda:0', grad_fn=) loss_train: 0.5205279276706278 step: 10 running loss: 0.05205279276706278 Train Steps: 10/90 Loss: 0.0521 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6193, 0.4108, 0.7438, 0.2700, 0.3650, 0.3683, 0.6238, 0.5717], [0.6216, 0.4099, 0.7225, 0.2033, 0.4188, 0.2217, 0.5975, 0.5283], [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733], [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006], [0.6260, 0.4120, 0.8013, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895], [0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4604, 0.3238, 0.7617, 0.3943, 0.3457, 0.3056, 0.6460, 0.4995], [0.4523, 0.3381, 0.6931, 0.3316, 0.3268, 0.2285, 0.6105, 0.5138], [0.4482, 0.3509, 0.7564, 0.4134, 0.3853, 0.3224, 0.5696, 0.5136], [0.5103, 0.3802, 0.7495, 0.4832, 0.3225, 0.3799, 0.6293, 0.5123], [0.5484, 0.3973, 0.7444, 0.5327, 0.3770, 0.5066, 0.6837, 0.5201], [0.4322, 0.3602, 0.7118, 0.3229, 0.3856, 0.1895, 0.5927, 0.5381], [0.4152, 0.3308, 0.6963, 0.2808, 0.3751, 0.1732, 0.5842, 0.4984], [0.4961, 0.3820, 0.7515, 0.4521, 0.3516, 0.3834, 0.6456, 0.5185]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6193, 0.4108, 0.7437, 0.2700, 0.3650, 0.3683, 0.6237, 0.5717], [0.6216, 0.4099, 0.7225, 0.2033, 0.4187, 0.2217, 0.5975, 0.5283], [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733], [0.6236, 0.3965, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6310, 0.4017, 0.8562, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006], [0.6259, 0.4120, 0.8012, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895], [0.6202, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297]]], device='cuda:0') loss_train_step before backward: tensor(0.0071, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0071, device='cuda:0', grad_fn=) loss_train: 0.5276685687713325 step: 11 running loss: 0.04796986988830296 Train Steps: 11/90 Loss: 0.0480 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800], [0.6361, 0.4076, 0.8862, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297], [0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650], [0.6109, 0.4015, 0.7668, 0.3639, 0.3513, 0.3667, 0.5200, 0.5641], [0.6201, 0.4064, 0.8688, 0.5050, 0.4225, 0.5100, 0.6138, 0.5500], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283], [ nan, nan, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4242, 0.3276, 0.7231, 0.3277, 0.3530, 0.2147, 0.5563, 0.5227], [0.5232, 0.4021, 0.7501, 0.5769, 0.3382, 0.5162, 0.6495, 0.5368], [0.4466, 0.3568, 0.7542, 0.4222, 0.3525, 0.2569, 0.5686, 0.5441], [0.4663, 0.3350, 0.7140, 0.4204, 0.3233, 0.3100, 0.5935, 0.5369], [0.5148, 0.3751, 0.7542, 0.5413, 0.3297, 0.4700, 0.6317, 0.5378], [0.4329, 0.3714, 0.7269, 0.3483, 0.3543, 0.2071, 0.5617, 0.5350], [0.4669, 0.3490, 0.7035, 0.3924, 0.3518, 0.2807, 0.5944, 0.5369], [0.4827, 0.3401, 0.7154, 0.3492, 0.3307, 0.2902, 0.5846, 0.5344]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5462, 0.5800], [0.6361, 0.4076, 0.8863, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297], [0.6203, 0.4096, 0.8863, 0.4267, 0.3537, 0.4117, 0.6025, 0.5650], [0.6109, 0.4015, 0.7668, 0.3639, 0.3512, 0.3667, 0.5200, 0.5641], [0.6201, 0.4064, 0.8687, 0.5050, 0.4225, 0.5100, 0.6137, 0.5500], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6187, 0.5283], [0.0000, 0.0000, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533]]], device='cuda:0') loss_train_step before backward: tensor(0.0119, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0119, device='cuda:0', grad_fn=) loss_train: 0.5395649061538279 step: 12 running loss: 0.044963742179485656 Train Steps: 12/90 Loss: 0.0450 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650], [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496], [0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433], [0.6350, 0.4043, 0.8738, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950], [0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6040, 0.4002, 0.7338, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4855, 0.3495, 0.7835, 0.4110, 0.3501, 0.2785, 0.5376, 0.5742], [0.4499, 0.3128, 0.7569, 0.3151, 0.3372, 0.1992, 0.5607, 0.5833], [0.5349, 0.3750, 0.7566, 0.4635, 0.3470, 0.4020, 0.6041, 0.5436], [0.5596, 0.3703, 0.7619, 0.5797, 0.3526, 0.4759, 0.5958, 0.5680], [0.5413, 0.3735, 0.7778, 0.4743, 0.3496, 0.4225, 0.5655, 0.5585], [0.4563, 0.3362, 0.7333, 0.3678, 0.3327, 0.2827, 0.5468, 0.5548], [0.4906, 0.3384, 0.7265, 0.3375, 0.3288, 0.2491, 0.5441, 0.5544], [0.5201, 0.3942, 0.7543, 0.4075, 0.3626, 0.3394, 0.5407, 0.5594]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6203, 0.4096, 0.8863, 0.4267, 0.3537, 0.4117, 0.6025, 0.5650], [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496], [0.6250, 0.4054, 0.8770, 0.4723, 0.4663, 0.5367, 0.6162, 0.5433], [0.6350, 0.4043, 0.8737, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950], [0.6277, 0.4036, 0.8687, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6082, 0.4024, 0.8737, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6040, 0.4002, 0.7337, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334]]], device='cuda:0') loss_train_step before backward: tensor(0.0066, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0066, device='cuda:0', grad_fn=) loss_train: 0.5461384528316557 step: 13 running loss: 0.042010650217819676 Train Steps: 13/90 Loss: 0.0420 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6339, 0.4149, 0.8800, 0.5000, 0.3900, 0.5283, 0.7541, 0.5424], [0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133], [0.6160, 0.4093, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808], [0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467], [0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609], [0.6284, 0.4093, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6138, 0.4089, 0.8334, 0.4689, 0.3602, 0.4501, 0.5896, 0.5509], [0.5220, 0.3308, 0.7897, 0.2446, 0.3819, 0.1887, 0.5110, 0.5884], [0.5544, 0.3636, 0.8083, 0.4162, 0.3470, 0.3582, 0.5382, 0.5543], [0.5467, 0.3849, 0.8156, 0.3669, 0.3635, 0.3404, 0.5649, 0.5676], [0.5364, 0.3596, 0.8030, 0.4386, 0.3820, 0.3932, 0.5638, 0.5476], [0.5485, 0.3434, 0.7971, 0.3553, 0.3534, 0.2809, 0.5138, 0.5525], [0.5894, 0.3773, 0.8297, 0.4457, 0.3587, 0.4137, 0.5796, 0.5698], [0.5982, 0.4032, 0.8306, 0.4746, 0.3600, 0.4449, 0.5974, 0.5406]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6339, 0.4149, 0.8800, 0.5000, 0.3900, 0.5283, 0.7541, 0.5424], [0.6205, 0.4015, 0.8350, 0.2717, 0.3988, 0.2550, 0.5788, 0.5133], [0.6160, 0.4092, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808], [0.6273, 0.4143, 0.8750, 0.5700, 0.3988, 0.4717, 0.6012, 0.5467], [0.6064, 0.4019, 0.8650, 0.4517, 0.4038, 0.5367, 0.5703, 0.5609], [0.6284, 0.4092, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947]]], device='cuda:0') loss_train_step before backward: tensor(0.0045, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0045, device='cuda:0', grad_fn=) loss_train: 0.5506306039169431 step: 14 running loss: 0.03933075742263879 Train Steps: 14/90 Loss: 0.0393 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819], [ nan, nan, 0.6512, 0.1717, 0.4100, 0.1983, 0.5253, 0.5240], [0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400], [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700], [0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133], [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967], [0.6353, 0.4128, 0.8488, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5937, 0.3509, 0.8326, 0.3062, 0.3551, 0.3064, 0.5233, 0.5584], [0.5701, 0.3368, 0.8274, 0.2523, 0.3597, 0.2419, 0.5122, 0.5743], [0.5733, 0.3884, 0.8194, 0.2436, 0.4069, 0.2326, 0.5367, 0.5481], [0.6927, 0.4126, 0.8742, 0.5447, 0.3724, 0.5729, 0.6022, 0.5212], [0.6857, 0.4112, 0.8971, 0.5565, 0.3780, 0.5510, 0.5959, 0.5286], [0.6102, 0.3915, 0.8287, 0.3693, 0.3949, 0.3350, 0.5436, 0.5626], [0.5731, 0.3627, 0.8294, 0.3020, 0.4315, 0.2867, 0.5511, 0.5545], [0.6631, 0.4388, 0.8668, 0.5153, 0.3947, 0.5607, 0.6025, 0.5215]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819], [0.0000, 0.0000, 0.6513, 0.1717, 0.4100, 0.1983, 0.5253, 0.5240], [0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400], [0.6111, 0.3995, 0.8788, 0.4567, 0.3812, 0.4833, 0.5450, 0.5700], [0.6308, 0.3990, 0.8687, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133], [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967], [0.6353, 0.4128, 0.8487, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6299, 0.4008, 0.8450, 0.5350, 0.4212, 0.5000, 0.6350, 0.5100]]], device='cuda:0') loss_train_step before backward: tensor(0.0177, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0177, device='cuda:0', grad_fn=) loss_train: 0.5683316038921475 step: 15 running loss: 0.03788877359280984 Train Steps: 15/90 Loss: 0.0379 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6200, 0.3993, 0.8519, 0.4923, 0.3962, 0.4717, 0.6013, 0.5433], [ nan, nan, 0.8938, 0.2850, 0.4662, 0.3117, 0.7406, 0.5528], [0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6547, 0.4026, 0.9127, 0.4880, 0.4210, 0.5077, 0.6168, 0.5180], [0.6555, 0.3753, 0.8744, 0.4401, 0.4125, 0.4705, 0.6017, 0.5382], [0.5861, 0.3916, 0.8767, 0.2952, 0.4540, 0.3381, 0.5853, 0.5346], [0.6865, 0.4021, 0.9131, 0.5099, 0.3946, 0.5068, 0.6111, 0.5421], [0.6070, 0.3931, 0.8990, 0.3118, 0.4520, 0.3536, 0.5994, 0.5259], [0.6044, 0.3817, 0.8638, 0.2735, 0.4256, 0.2824, 0.5731, 0.5519], [0.6471, 0.3808, 0.8958, 0.4194, 0.4224, 0.4481, 0.5902, 0.5492], [0.6469, 0.3749, 0.8770, 0.3937, 0.4066, 0.3749, 0.5731, 0.5123]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6200, 0.3993, 0.8519, 0.4923, 0.3963, 0.4717, 0.6012, 0.5433], [0.0000, 0.0000, 0.8938, 0.2850, 0.4663, 0.3117, 0.7406, 0.5528], [0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285]]], device='cuda:0') loss_train_step before backward: tensor(0.0110, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0110, device='cuda:0', grad_fn=) loss_train: 0.5793465515598655 step: 16 running loss: 0.03620915947249159 Train Steps: 16/90 Loss: 0.0362 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134], [ nan, nan, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819], [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899], [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650], [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6272, 0.4045, 0.8538, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6627, 0.4136, 0.9392, 0.5165, 0.4523, 0.5500, 0.6682, 0.4953], [0.5376, 0.3240, 0.8913, 0.2640, 0.4416, 0.3313, 0.5895, 0.5246], [0.6395, 0.3805, 0.9361, 0.4163, 0.4485, 0.4339, 0.6147, 0.5254], [0.5680, 0.3594, 0.9200, 0.2960, 0.4363, 0.3497, 0.5919, 0.5198], [0.5743, 0.3581, 0.8822, 0.2888, 0.4754, 0.3589, 0.6170, 0.5224], [0.6266, 0.3792, 0.9241, 0.3970, 0.4373, 0.4108, 0.6244, 0.4996], [0.6290, 0.3848, 0.9422, 0.4540, 0.4268, 0.4926, 0.6146, 0.5069], [0.6318, 0.4059, 0.9420, 0.4500, 0.4560, 0.4884, 0.6553, 0.5268]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134], [0.0000, 0.0000, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819], [0.6346, 0.4144, 0.9087, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899], [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650], [0.6072, 0.4029, 0.7038, 0.2150, 0.3913, 0.2267, 0.5516, 0.5507], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6271, 0.4045, 0.8537, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6273, 0.4143, 0.8750, 0.5700, 0.3988, 0.4717, 0.6012, 0.5467]]], device='cuda:0') loss_train_step before backward: tensor(0.0105, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0105, device='cuda:0', grad_fn=) loss_train: 0.5897985212504864 step: 17 running loss: 0.034694030661793315 Train Steps: 17/90 Loss: 0.0347 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6139, 0.4019, 0.7137, 0.2150, 0.4375, 0.1533, 0.5293, 0.5006], [0.6147, 0.4107, 0.8137, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412], [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833], [0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123], [0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683], [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4989, 0.3019, 0.8501, 0.2146, 0.4736, 0.2612, 0.5998, 0.5084], [0.5352, 0.3228, 0.8733, 0.2890, 0.4420, 0.2892, 0.5881, 0.5260], [0.6540, 0.3960, 0.9527, 0.5137, 0.4338, 0.5415, 0.6720, 0.5053], [0.5784, 0.3489, 0.9258, 0.4035, 0.4336, 0.4318, 0.6304, 0.5092], [0.5422, 0.3244, 0.9022, 0.2565, 0.4607, 0.3007, 0.6357, 0.5141], [0.5923, 0.3676, 0.9305, 0.4436, 0.4626, 0.4719, 0.6405, 0.5138], [0.6616, 0.3972, 0.9572, 0.5468, 0.4315, 0.5406, 0.6886, 0.4973], [0.6289, 0.3807, 0.9102, 0.4968, 0.4496, 0.4921, 0.6804, 0.4749]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6139, 0.4019, 0.7138, 0.2150, 0.4375, 0.1533, 0.5293, 0.5006], [0.6147, 0.4107, 0.8138, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412], [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6037, 0.4833], [0.6213, 0.4131, 0.8438, 0.3550, 0.3512, 0.4400, 0.5716, 0.5123], [0.6187, 0.4103, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683], [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617]]], device='cuda:0') loss_train_step before backward: tensor(0.0042, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0042, device='cuda:0', grad_fn=) loss_train: 0.59402234852314 step: 18 running loss: 0.03300124158461889 Train Steps: 18/90 Loss: 0.0330 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921], [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383], [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012], [0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901], [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5489, 0.3691, 0.8602, 0.3525, 0.4167, 0.3724, 0.6237, 0.5266], [0.5204, 0.3422, 0.8550, 0.2708, 0.4328, 0.2782, 0.6180, 0.5341], [0.4912, 0.3260, 0.8521, 0.2680, 0.4374, 0.2647, 0.6071, 0.5039], [0.6073, 0.3701, 0.8993, 0.4969, 0.4102, 0.4751, 0.6399, 0.5054], [0.5643, 0.3735, 0.8997, 0.4671, 0.4200, 0.4458, 0.6410, 0.5132], [0.6377, 0.3858, 0.9165, 0.5669, 0.4071, 0.5251, 0.6724, 0.4992], [0.5725, 0.3525, 0.8803, 0.3685, 0.4274, 0.3686, 0.6311, 0.5081], [0.6164, 0.3833, 0.9266, 0.5551, 0.4389, 0.5435, 0.6636, 0.5159]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6257, 0.4034, 0.8288, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921], [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383], [0.6314, 0.4107, 0.8750, 0.5100, 0.3787, 0.4900, 0.7121, 0.5864], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012], [0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901], [0.6339, 0.4123, 0.8637, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436]]], device='cuda:0') loss_train_step before backward: tensor(0.0027, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0027, device='cuda:0', grad_fn=) loss_train: 0.5967375976033509 step: 19 running loss: 0.03140724197912373 Train Steps: 19/90 Loss: 0.0314 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6200, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6307, 0.4029, 0.8650, 0.5200, 0.3763, 0.4017, 0.7311, 0.5366], [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5002, 0.3322, 0.7828, 0.2725, 0.3797, 0.2560, 0.5804, 0.5275], [0.6001, 0.3807, 0.8608, 0.4400, 0.3863, 0.4352, 0.6208, 0.5364], [0.5115, 0.3182, 0.8096, 0.2736, 0.3886, 0.2493, 0.6107, 0.5172], [0.5968, 0.4046, 0.8681, 0.5250, 0.4023, 0.4724, 0.6819, 0.5299], [0.6252, 0.3984, 0.8836, 0.5122, 0.4340, 0.4717, 0.6479, 0.5141], [0.6535, 0.4172, 0.8849, 0.5469, 0.3984, 0.5496, 0.6890, 0.4915], [0.5809, 0.3773, 0.8423, 0.3697, 0.3968, 0.3501, 0.6085, 0.5313], [0.5965, 0.3755, 0.8642, 0.4635, 0.3780, 0.3989, 0.6192, 0.5332]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6149, 0.4054, 0.6712, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5362, 0.5550], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6199, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6307, 0.4029, 0.8650, 0.5200, 0.3762, 0.4017, 0.7311, 0.5366], [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297]]], device='cuda:0') loss_train_step before backward: tensor(0.0033, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0033, device='cuda:0', grad_fn=) loss_train: 0.6000654962845147 step: 20 running loss: 0.030003274814225732 Train Steps: 20/90 Loss: 0.0300 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6273, 0.4105, 0.8988, 0.4517, 0.3912, 0.2550, 0.5894, 0.4811], [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633], [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750], [0.6084, 0.3981, 0.8588, 0.5233, 0.4600, 0.5367, 0.5680, 0.5006], [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413], [ nan, nan, 0.6512, 0.1717, 0.4100, 0.1983, 0.5253, 0.5240], [0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5916, 0.3844, 0.7633, 0.3877, 0.3732, 0.3227, 0.6046, 0.5243], [0.6335, 0.4340, 0.7941, 0.4725, 0.3441, 0.4270, 0.6307, 0.5186], [0.6787, 0.4430, 0.8263, 0.5454, 0.3485, 0.4909, 0.6363, 0.4990], [0.5771, 0.3782, 0.7535, 0.3500, 0.3570, 0.3527, 0.6062, 0.5452], [0.6281, 0.4197, 0.8233, 0.5323, 0.3730, 0.4784, 0.6468, 0.5321], [0.5682, 0.4123, 0.8094, 0.3721, 0.4299, 0.3495, 0.6410, 0.5398], [0.4449, 0.3098, 0.7400, 0.2139, 0.3699, 0.2040, 0.5648, 0.5597], [0.6228, 0.4093, 0.8242, 0.4646, 0.3890, 0.4224, 0.6368, 0.5439]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6273, 0.4105, 0.8988, 0.4517, 0.3913, 0.2550, 0.5894, 0.4811], [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5638, 0.5633], [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750], [0.6084, 0.3981, 0.8587, 0.5233, 0.4600, 0.5367, 0.5680, 0.5006], [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413], [0.0000, 0.0000, 0.6513, 0.1717, 0.4100, 0.1983, 0.5253, 0.5240], [0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090]]], device='cuda:0') loss_train_step before backward: tensor(0.0075, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0075, device='cuda:0', grad_fn=) loss_train: 0.6075229728594422 step: 21 running loss: 0.028929665374259155 Train Steps: 21/90 Loss: 0.0289 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397], [0.6272, 0.4045, 0.8538, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267], [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6100, 0.4071, 0.7601, 0.3444, 0.3400, 0.4117, 0.5625, 0.5617], [0.6277, 0.4103, 0.8087, 0.5717, 0.4188, 0.4750, 0.5663, 0.6083], [0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5361, 0.3473, 0.7224, 0.2576, 0.3896, 0.2200, 0.5687, 0.5548], [0.6341, 0.4344, 0.8141, 0.4878, 0.3341, 0.4040, 0.5881, 0.5413], [0.6328, 0.4104, 0.7829, 0.5221, 0.3337, 0.4532, 0.6095, 0.5360], [0.6438, 0.4458, 0.8056, 0.4583, 0.3897, 0.4038, 0.6523, 0.5541], [0.6684, 0.4341, 0.8065, 0.5662, 0.3743, 0.4831, 0.6147, 0.5662], [0.5448, 0.4118, 0.7564, 0.3263, 0.3703, 0.3139, 0.5909, 0.5578], [0.6425, 0.4230, 0.8266, 0.5468, 0.3626, 0.4794, 0.6113, 0.5741], [0.5344, 0.3745, 0.7279, 0.2517, 0.3772, 0.2505, 0.5578, 0.5618]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6159, 0.4085, 0.6900, 0.2283, 0.4087, 0.1950, 0.5123, 0.5397], [0.6271, 0.4045, 0.8537, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267], [0.6257, 0.4167, 0.8775, 0.3433, 0.3562, 0.4133, 0.6200, 0.5667], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6100, 0.4071, 0.7601, 0.3444, 0.3400, 0.4117, 0.5625, 0.5617], [0.6277, 0.4103, 0.8087, 0.5717, 0.4187, 0.4750, 0.5663, 0.6083], [0.6123, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485]]], device='cuda:0') loss_train_step before backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train: 0.6093606404028833 step: 22 running loss: 0.027698210927403787 Train Steps: 22/90 Loss: 0.0277 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6332, 0.4118, 0.9238, 0.4267, 0.4012, 0.4733, 0.7525, 0.5436], [0.6128, 0.4118, 0.8638, 0.5333, 0.4625, 0.5267, 0.5193, 0.5475], [0.6245, 0.4115, 0.8700, 0.4883, 0.4625, 0.5517, 0.6100, 0.5217], [0.6276, 0.4120, 0.8738, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667], [0.6064, 0.3953, 0.8738, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6037, 0.4042, 0.7861, 0.4525, 0.3799, 0.4002, 0.5853, 0.5947], [0.6158, 0.3983, 0.7456, 0.4208, 0.3749, 0.3494, 0.5616, 0.5747], [0.5826, 0.3949, 0.7472, 0.4460, 0.3831, 0.4107, 0.5832, 0.5636], [0.5835, 0.4133, 0.7313, 0.3065, 0.4241, 0.2709, 0.5836, 0.5425], [0.6116, 0.4081, 0.7711, 0.4174, 0.3359, 0.3233, 0.5526, 0.5699], [0.6512, 0.4110, 0.7677, 0.5061, 0.3607, 0.4228, 0.5736, 0.5367], [0.6322, 0.4209, 0.7510, 0.4096, 0.3741, 0.3559, 0.6041, 0.5573], [0.6771, 0.3982, 0.7817, 0.4814, 0.3504, 0.4184, 0.5547, 0.5450]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6332, 0.4118, 0.9237, 0.4267, 0.4013, 0.4733, 0.7525, 0.5436], [0.6128, 0.4118, 0.8637, 0.5333, 0.4625, 0.5267, 0.5193, 0.5475], [0.6245, 0.4115, 0.8700, 0.4883, 0.4625, 0.5517, 0.6100, 0.5217], [0.6276, 0.4120, 0.8737, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6257, 0.4167, 0.8775, 0.3433, 0.3562, 0.4133, 0.6200, 0.5667], [0.6064, 0.3952, 0.8737, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416]]], device='cuda:0') loss_train_step before backward: tensor(0.0047, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0047, device='cuda:0', grad_fn=) loss_train: 0.6140206111595035 step: 23 running loss: 0.02669654831128276 Train Steps: 23/90 Loss: 0.0267 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431], [ nan, nan, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733], [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598], [0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967], [0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799], [ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6534, 0.4442, 0.7870, 0.5273, 0.3688, 0.4122, 0.5606, 0.5491], [0.4899, 0.3300, 0.6995, 0.1907, 0.4047, 0.1847, 0.5049, 0.5413], [0.5462, 0.3605, 0.7138, 0.3059, 0.3777, 0.2765, 0.5254, 0.5580], [0.6595, 0.4666, 0.8279, 0.5287, 0.3715, 0.4721, 0.6088, 0.5316], [0.6401, 0.4335, 0.7881, 0.5154, 0.3938, 0.4018, 0.5882, 0.5655], [0.7212, 0.4738, 0.8581, 0.6698, 0.3993, 0.5994, 0.6318, 0.5457], [0.5917, 0.3890, 0.7234, 0.3395, 0.3801, 0.2883, 0.5364, 0.5730], [0.5050, 0.3630, 0.7010, 0.2187, 0.3996, 0.2131, 0.5094, 0.5673]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431], [0.0000, 0.0000, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733], [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598], [0.6199, 0.4102, 0.8950, 0.4417, 0.4013, 0.5367, 0.6112, 0.5967], [0.6200, 0.4118, 0.8288, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799], [0.0000, 0.0000, 0.7513, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617]]], device='cuda:0') loss_train_step before backward: tensor(0.0145, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0145, device='cuda:0', grad_fn=) loss_train: 0.628472588956356 step: 24 running loss: 0.026186357873181503 Train Steps: 24/90 Loss: 0.0262 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6286, 0.4055, 0.9000, 0.4717, 0.3763, 0.4683, 0.7018, 0.5494], [ nan, nan, 0.7335, 0.2569, 0.3788, 0.2667, 0.5066, 0.5578], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [ nan, nan, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650], [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378], [0.6153, 0.4117, 0.8688, 0.5167, 0.4895, 0.5647, 0.5524, 0.5136], [0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6083, 0.4021, 0.8504, 0.4506, 0.4062, 0.4482, 0.5526, 0.5495], [0.6550, 0.4405, 0.8656, 0.6100, 0.3859, 0.5233, 0.5851, 0.5481], [0.4780, 0.3301, 0.7199, 0.2762, 0.4161, 0.2419, 0.4989, 0.5769], [0.5234, 0.3830, 0.8144, 0.3121, 0.4428, 0.3014, 0.5504, 0.5798], [0.4204, 0.3060, 0.7255, 0.2233, 0.4225, 0.1847, 0.5139, 0.5718], [0.6194, 0.4022, 0.8339, 0.4905, 0.4203, 0.3756, 0.5566, 0.5277], [0.5894, 0.4061, 0.8469, 0.5381, 0.4337, 0.5094, 0.5568, 0.5652], [0.4978, 0.3319, 0.7075, 0.3023, 0.4139, 0.2716, 0.4750, 0.5493]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6286, 0.4055, 0.9000, 0.4717, 0.3762, 0.4683, 0.7018, 0.5494], [0.0000, 0.0000, 0.7335, 0.2569, 0.3787, 0.2667, 0.5066, 0.5578], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.0000, 0.0000, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650], [0.6307, 0.4029, 0.8988, 0.4817, 0.3938, 0.3500, 0.7311, 0.5378], [0.6154, 0.4117, 0.8687, 0.5167, 0.4895, 0.5647, 0.5524, 0.5136], [0.6149, 0.4054, 0.6712, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695]]], device='cuda:0') loss_train_step before backward: tensor(0.0125, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0125, device='cuda:0', grad_fn=) loss_train: 0.6410121768712997 step: 25 running loss: 0.02564048707485199 Train Steps: 25/90 Loss: 0.0256 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183], [ nan, nan, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702], [0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913], [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719], [0.6199, 0.4071, 0.7600, 0.2117, 0.4037, 0.2767, 0.6138, 0.5550], [0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667], [0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967], [0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5108, 0.3604, 0.8390, 0.3908, 0.4406, 0.3910, 0.5886, 0.5432], [0.4525, 0.3217, 0.8464, 0.3252, 0.4609, 0.3114, 0.5518, 0.5576], [0.3797, 0.2495, 0.7300, 0.1931, 0.4586, 0.1687, 0.5035, 0.5400], [0.5461, 0.3603, 0.8976, 0.4546, 0.4051, 0.4570, 0.5562, 0.5573], [0.4287, 0.3064, 0.7561, 0.2479, 0.4488, 0.2206, 0.5375, 0.5389], [0.4644, 0.2900, 0.7424, 0.2811, 0.3841, 0.3158, 0.4938, 0.5445], [0.6190, 0.3835, 0.9274, 0.5857, 0.4378, 0.6095, 0.5924, 0.5434], [0.5422, 0.3625, 0.8487, 0.4538, 0.4157, 0.4287, 0.5251, 0.5353]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183], [0.0000, 0.0000, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702], [0.6286, 0.4078, 0.8062, 0.2267, 0.4787, 0.1533, 0.5953, 0.4913], [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719], [0.6199, 0.4071, 0.7600, 0.2117, 0.4038, 0.2767, 0.6137, 0.5550], [0.6134, 0.4090, 0.6926, 0.2819, 0.3537, 0.3233, 0.5562, 0.5667], [0.6199, 0.4102, 0.8950, 0.4417, 0.4013, 0.5367, 0.6112, 0.5967], [0.6278, 0.4253, 0.8875, 0.5017, 0.4112, 0.2750, 0.5413, 0.6196]]], device='cuda:0') loss_train_step before backward: tensor(0.0109, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0109, device='cuda:0', grad_fn=) loss_train: 0.6518932655453682 step: 26 running loss: 0.025072817905591085 Train Steps: 26/90 Loss: 0.0251 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6362, 0.5200], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343], [ nan, nan, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999], [0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5346, 0.3361, 0.9201, 0.4234, 0.4331, 0.4263, 0.5722, 0.5859], [0.5568, 0.3413, 0.9471, 0.4525, 0.4453, 0.4915, 0.5743, 0.5571], [0.3841, 0.2736, 0.8017, 0.2267, 0.4508, 0.2552, 0.5554, 0.5411], [0.4868, 0.3168, 0.9052, 0.4018, 0.4096, 0.3981, 0.5696, 0.5464], [0.3279, 0.2467, 0.8089, 0.1724, 0.4432, 0.1667, 0.5400, 0.5439], [0.3911, 0.2765, 0.7820, 0.1967, 0.4317, 0.2572, 0.5715, 0.5536], [0.4877, 0.3180, 0.8559, 0.3810, 0.4573, 0.3899, 0.5876, 0.5489], [0.5775, 0.3616, 0.9427, 0.5111, 0.4411, 0.5220, 0.6233, 0.5373]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6363, 0.5200], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6311, 0.3998, 0.7975, 0.5767, 0.3837, 0.4850, 0.7327, 0.5343], [0.0000, 0.0000, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999], [0.6213, 0.4001, 0.7713, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6189, 0.4049, 0.8888, 0.4417, 0.4212, 0.5200, 0.5987, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0092, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0092, device='cuda:0', grad_fn=) loss_train: 0.6610878705978394 step: 27 running loss: 0.024484735948068125 Train Steps: 27/90 Loss: 0.0245 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600], [0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6200, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6272, 0.4045, 0.8538, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4143, 0.2747, 0.8007, 0.2348, 0.4075, 0.2985, 0.5870, 0.5697], [0.4828, 0.2987, 0.9111, 0.4216, 0.4440, 0.4238, 0.6115, 0.5584], [0.3344, 0.2688, 0.8159, 0.2095, 0.4573, 0.1910, 0.6030, 0.5582], [0.5427, 0.3206, 0.9698, 0.4371, 0.4166, 0.4546, 0.6427, 0.5454], [0.5437, 0.3577, 0.9476, 0.4925, 0.3931, 0.4746, 0.6047, 0.5349], [0.3526, 0.2414, 0.7833, 0.1797, 0.4681, 0.2100, 0.6073, 0.5477], [0.4671, 0.3069, 0.9184, 0.3883, 0.4627, 0.3664, 0.6575, 0.5600], [0.4275, 0.2836, 0.8408, 0.2699, 0.4258, 0.3069, 0.5922, 0.5463]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6132, 0.4066, 0.7259, 0.2402, 0.3587, 0.3300, 0.6000, 0.5600], [0.6084, 0.4008, 0.8587, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6201, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6271, 0.4045, 0.8537, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646]]], device='cuda:0') loss_train_step before backward: tensor(0.0080, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0080, device='cuda:0', grad_fn=) loss_train: 0.6691014468669891 step: 28 running loss: 0.023896480245249613 Train Steps: 28/90 Loss: 0.0239 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510], [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.6203, 0.4078, 0.8800, 0.5083, 0.3900, 0.5000, 0.6100, 0.5583], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5390, 0.3826, 0.9603, 0.4170, 0.3944, 0.4088, 0.6443, 0.5350], [0.4486, 0.2729, 0.8097, 0.2505, 0.4108, 0.2737, 0.6175, 0.5547], [0.4647, 0.3007, 0.8495, 0.2775, 0.4124, 0.2690, 0.6329, 0.5355], [0.4015, 0.2568, 0.7904, 0.2029, 0.3834, 0.2554, 0.5896, 0.5526], [0.5965, 0.3690, 0.9675, 0.5167, 0.4347, 0.5499, 0.7048, 0.5469], [0.6405, 0.3905, 0.9769, 0.5324, 0.3739, 0.5301, 0.6639, 0.5552], [0.4013, 0.2692, 0.8289, 0.1844, 0.4672, 0.1578, 0.6531, 0.5266], [0.4481, 0.2940, 0.8121, 0.2447, 0.4270, 0.2471, 0.6292, 0.5344]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6108, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510], [0.6241, 0.4142, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.6203, 0.4078, 0.8800, 0.5083, 0.3900, 0.5000, 0.6100, 0.5583], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726]]], device='cuda:0') loss_train_step before backward: tensor(0.0062, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0062, device='cuda:0', grad_fn=) loss_train: 0.6753336312249303 step: 29 running loss: 0.02328736659396311 Train Steps: 29/90 Loss: 0.0233 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019], [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6111, 0.4033, 0.8300, 0.3267, 0.3588, 0.3333, 0.5444, 0.5637], [0.6250, 0.4106, 0.8700, 0.3717, 0.3588, 0.4967, 0.6038, 0.5167], [0.6293, 0.3982, 0.8700, 0.5300, 0.3763, 0.4717, 0.7050, 0.5297], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283], [0.6250, 0.4103, 0.8950, 0.4400, 0.3912, 0.5650, 0.6050, 0.5133]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6369, 0.4089, 0.8763, 0.3559, 0.3884, 0.3804, 0.6565, 0.5152], [0.5575, 0.3820, 0.8672, 0.3558, 0.4400, 0.3587, 0.6304, 0.5472], [0.4623, 0.2721, 0.7361, 0.2071, 0.3519, 0.1966, 0.6082, 0.5153], [0.5630, 0.3509, 0.8418, 0.3172, 0.3521, 0.2707, 0.6689, 0.4976], [0.5934, 0.3768, 0.8642, 0.3091, 0.4111, 0.3737, 0.6373, 0.5525], [0.6304, 0.3884, 0.8941, 0.4007, 0.3824, 0.3954, 0.6342, 0.5111], [0.5640, 0.3495, 0.8785, 0.3389, 0.4095, 0.2641, 0.7161, 0.5165], [0.6403, 0.4074, 0.9030, 0.3953, 0.4031, 0.4607, 0.6779, 0.5264]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6161, 0.4099, 0.8737, 0.4383, 0.3787, 0.5483, 0.5605, 0.5019], [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6111, 0.4033, 0.8300, 0.3267, 0.3587, 0.3333, 0.5444, 0.5637], [0.6250, 0.4105, 0.8700, 0.3717, 0.3587, 0.4967, 0.6037, 0.5167], [0.6293, 0.3982, 0.8700, 0.5300, 0.3762, 0.4717, 0.7050, 0.5297], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6187, 0.5283], [0.6250, 0.4103, 0.8950, 0.4400, 0.3913, 0.5650, 0.6050, 0.5133]]], device='cuda:0') loss_train_step before backward: tensor(0.0046, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0046, device='cuda:0', grad_fn=) loss_train: 0.6799087165854871 step: 30 running loss: 0.022663623886182906 Train Steps: 30/90 Loss: 0.0227 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6223, 0.4028, 0.8988, 0.4200, 0.3763, 0.5733, 0.6375, 0.5167], [ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817], [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283], [0.6188, 0.4099, 0.7400, 0.2433, 0.3962, 0.2750, 0.6162, 0.5467], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.8508, 0.5153, 0.9209, 0.4967, 0.3942, 0.5981, 0.6709, 0.5179], [0.4904, 0.3145, 0.7087, 0.2052, 0.3900, 0.1955, 0.5960, 0.5065], [0.4885, 0.3282, 0.7663, 0.2554, 0.4071, 0.1947, 0.6204, 0.5186], [0.6714, 0.4204, 0.7941, 0.3738, 0.3507, 0.3638, 0.6271, 0.5192], [0.5495, 0.3321, 0.7264, 0.2246, 0.4070, 0.2041, 0.6100, 0.5103], [0.6896, 0.4610, 0.8982, 0.4124, 0.3639, 0.4241, 0.6851, 0.5136], [0.5575, 0.3861, 0.7116, 0.2491, 0.3762, 0.2665, 0.6054, 0.5234], [0.7678, 0.4905, 0.9072, 0.5293, 0.3453, 0.4882, 0.6262, 0.4931]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6223, 0.4028, 0.8988, 0.4200, 0.3762, 0.5733, 0.6375, 0.5167], [0.0000, 0.0000, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817], [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283], [0.6188, 0.4099, 0.7400, 0.2433, 0.3963, 0.2750, 0.6162, 0.5467], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398]]], device='cuda:0') loss_train_step before backward: tensor(0.0089, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0089, device='cuda:0', grad_fn=) loss_train: 0.6888156817294657 step: 31 running loss: 0.022219860700950507 Train Steps: 31/90 Loss: 0.0222 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6250, 0.4146, 0.8838, 0.3933, 0.3588, 0.4283, 0.6162, 0.5367], [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6183, 0.4076, 0.8838, 0.4517, 0.3813, 0.4483, 0.5775, 0.5633], [0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050], [0.6104, 0.4029, 0.8738, 0.4900, 0.4088, 0.4533, 0.5070, 0.5510], [0.6157, 0.4102, 0.8513, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6710, 0.4203, 0.7945, 0.3961, 0.3727, 0.3562, 0.6165, 0.4802], [0.5841, 0.3361, 0.7245, 0.2290, 0.3879, 0.2388, 0.5897, 0.5041], [0.6982, 0.4584, 0.7993, 0.3879, 0.3324, 0.3598, 0.5931, 0.5174], [0.7280, 0.4381, 0.8203, 0.4481, 0.3542, 0.4475, 0.6107, 0.5190], [0.7076, 0.4717, 0.8278, 0.4338, 0.3845, 0.4404, 0.5895, 0.5367], [0.7239, 0.4483, 0.8322, 0.4365, 0.3643, 0.4370, 0.5907, 0.5236], [0.6941, 0.4412, 0.7852, 0.3663, 0.3613, 0.3664, 0.6066, 0.5027], [0.7049, 0.4463, 0.7884, 0.3625, 0.3525, 0.3731, 0.6007, 0.5497]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6250, 0.4146, 0.8838, 0.3933, 0.3587, 0.4283, 0.6162, 0.5367], [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6183, 0.4076, 0.8838, 0.4517, 0.3812, 0.4483, 0.5775, 0.5633], [0.6097, 0.3988, 0.8650, 0.5250, 0.4212, 0.5200, 0.5675, 0.5050], [0.6104, 0.4029, 0.8737, 0.4900, 0.4087, 0.4533, 0.5070, 0.5510], [0.6157, 0.4102, 0.8512, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583]]], device='cuda:0') loss_train_step before backward: tensor(0.0029, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0029, device='cuda:0', grad_fn=) loss_train: 0.6917505832388997 step: 32 running loss: 0.021617205726215616 Train Steps: 32/90 Loss: 0.0216 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6246, 0.4028, 0.8738, 0.4867, 0.4088, 0.5667, 0.6362, 0.5200], [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6289, 0.4024, 0.9088, 0.4567, 0.3937, 0.5633, 0.7058, 0.5609], [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7848, 0.4827, 0.8367, 0.4659, 0.3955, 0.5450, 0.6055, 0.5506], [0.6511, 0.3996, 0.7105, 0.3123, 0.3613, 0.2571, 0.5342, 0.5192], [0.7224, 0.4844, 0.8538, 0.4692, 0.3898, 0.5190, 0.5744, 0.5551], [0.7083, 0.4858, 0.8719, 0.4633, 0.4123, 0.4938, 0.5730, 0.5529], [0.6830, 0.4370, 0.7286, 0.3365, 0.3475, 0.3186, 0.5642, 0.5397], [0.6332, 0.4395, 0.7310, 0.3280, 0.3316, 0.2940, 0.5389, 0.5436], [0.6753, 0.4355, 0.7696, 0.3689, 0.3314, 0.3119, 0.5551, 0.5516], [0.7349, 0.4735, 0.8271, 0.4398, 0.3587, 0.3736, 0.5524, 0.5296]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6246, 0.4028, 0.8737, 0.4867, 0.4087, 0.5667, 0.6363, 0.5200], [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6289, 0.4024, 0.9087, 0.4567, 0.3938, 0.5633, 0.7058, 0.5609], [0.6155, 0.4124, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6076, 0.3952, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051]]], device='cuda:0') loss_train_step before backward: tensor(0.0031, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0031, device='cuda:0', grad_fn=) loss_train: 0.694860020885244 step: 33 running loss: 0.021056364269249818 Train Steps: 33/90 Loss: 0.0211 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433], [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483], [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6188, 0.5433], [0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6149, 0.4340, 0.7060, 0.3099, 0.3430, 0.3445, 0.5397, 0.5623], [0.7183, 0.4977, 0.8231, 0.5166, 0.3665, 0.5264, 0.5438, 0.5380], [0.7315, 0.4994, 0.8343, 0.4605, 0.3698, 0.5051, 0.5327, 0.5626], [0.7524, 0.4987, 0.8748, 0.5025, 0.3540, 0.5581, 0.5317, 0.5548], [0.5284, 0.3690, 0.6747, 0.3188, 0.3901, 0.2372, 0.5341, 0.5517], [0.6771, 0.4421, 0.7946, 0.4589, 0.3322, 0.4516, 0.5312, 0.5619], [0.6197, 0.4022, 0.7689, 0.3958, 0.3993, 0.2837, 0.6107, 0.5413], [0.7074, 0.4393, 0.8183, 0.4463, 0.3742, 0.3971, 0.5818, 0.5478]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6175, 0.4091, 0.7862, 0.2800, 0.3638, 0.3583, 0.6187, 0.5433], [0.6299, 0.4008, 0.8450, 0.5350, 0.4212, 0.5000, 0.6350, 0.5100], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483], [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6187, 0.5433], [0.6229, 0.4198, 0.7663, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6275, 0.4111, 0.8462, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117]]], device='cuda:0') loss_train_step before backward: tensor(0.0036, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0036, device='cuda:0', grad_fn=) loss_train: 0.6984764796216041 step: 34 running loss: 0.02054342587122365 Train Steps: 34/90 Loss: 0.0205 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823], [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690], [0.6151, 0.4058, 0.7068, 0.2680, 0.3400, 0.4083, 0.5775, 0.5733], [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960], [ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800], [0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927], [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5280, 0.3390, 0.7156, 0.3190, 0.3913, 0.2567, 0.5322, 0.5371], [0.7018, 0.5026, 0.8817, 0.5279, 0.3719, 0.5954, 0.5577, 0.5688], [0.6409, 0.4319, 0.7399, 0.3588, 0.3451, 0.4342, 0.5118, 0.5571], [0.6910, 0.4941, 0.8952, 0.5139, 0.4234, 0.5669, 0.5235, 0.5712], [0.5076, 0.3336, 0.7361, 0.3109, 0.3979, 0.2886, 0.5175, 0.5497], [0.5446, 0.3609, 0.7134, 0.3174, 0.3978, 0.2716, 0.5157, 0.5486], [0.7687, 0.5217, 0.9188, 0.5964, 0.3390, 0.5564, 0.5295, 0.5449], [0.6559, 0.4589, 0.8268, 0.4680, 0.3728, 0.3424, 0.5794, 0.5420]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6055, 0.4015, 0.7425, 0.2033, 0.4112, 0.1883, 0.5217, 0.4823], [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690], [0.6151, 0.4058, 0.7068, 0.2680, 0.3400, 0.4083, 0.5775, 0.5733], [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960], [0.0000, 0.0000, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6218, 0.4185, 0.7337, 0.2650, 0.4625, 0.1950, 0.5688, 0.5800], [0.6321, 0.4048, 0.8737, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927], [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367]]], device='cuda:0') loss_train_step before backward: tensor(0.0101, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0101, device='cuda:0', grad_fn=) loss_train: 0.7085485982242972 step: 35 running loss: 0.020244245663551347 Train Steps: 35/90 Loss: 0.0202 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200], [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.6357, 0.4097, 0.9038, 0.3883, 0.4213, 0.2950, 0.6686, 0.5390], [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6796, 0.4346, 0.9105, 0.5463, 0.3525, 0.5400, 0.5134, 0.5412], [0.6602, 0.4432, 0.9022, 0.4741, 0.3960, 0.5250, 0.5403, 0.5289], [0.6180, 0.4283, 0.8339, 0.4508, 0.4131, 0.3817, 0.5499, 0.5397], [0.5023, 0.3241, 0.7431, 0.3296, 0.4211, 0.2222, 0.5520, 0.5293], [0.5949, 0.3923, 0.8258, 0.4020, 0.4123, 0.3946, 0.5833, 0.5566], [0.6230, 0.4350, 0.8654, 0.3879, 0.3692, 0.5040, 0.5558, 0.5466], [0.5663, 0.3615, 0.7780, 0.3761, 0.4226, 0.2939, 0.5498, 0.5536], [0.6266, 0.4297, 0.8683, 0.4994, 0.3981, 0.5438, 0.5580, 0.5618]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200], [0.6068, 0.3963, 0.8650, 0.4317, 0.4038, 0.5083, 0.5253, 0.4999], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.6357, 0.4097, 0.9038, 0.3883, 0.4212, 0.2950, 0.6686, 0.5390], [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [0.6193, 0.4078, 0.7287, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690]]], device='cuda:0') loss_train_step before backward: tensor(0.0033, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0033, device='cuda:0', grad_fn=) loss_train: 0.7118957417551428 step: 36 running loss: 0.019774881715420634 Train Steps: 36/90 Loss: 0.0198 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433], [0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.6353, 0.4128, 0.9138, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991], [0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5292, 0.3872, 0.7569, 0.2700, 0.3864, 0.3128, 0.5665, 0.5585], [0.6096, 0.4050, 0.8747, 0.4772, 0.4006, 0.4952, 0.5530, 0.5093], [0.5875, 0.3963, 0.8670, 0.4399, 0.4564, 0.4967, 0.5858, 0.5465], [0.5625, 0.3597, 0.8807, 0.4257, 0.4473, 0.4756, 0.5719, 0.5578], [0.5999, 0.4131, 0.8598, 0.4714, 0.3961, 0.4890, 0.5484, 0.5499], [0.5851, 0.3990, 0.8530, 0.4214, 0.3858, 0.4066, 0.5353, 0.5361], [0.5423, 0.3711, 0.8811, 0.3812, 0.4638, 0.3094, 0.6062, 0.5717], [0.5934, 0.3772, 0.9045, 0.4281, 0.3759, 0.4242, 0.5442, 0.5451]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6175, 0.4091, 0.7862, 0.2800, 0.3638, 0.3583, 0.6187, 0.5433], [0.6308, 0.3990, 0.8687, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6251, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6087, 0.5183], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.6353, 0.4128, 0.9137, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991], [0.6212, 0.4033, 0.8938, 0.4167, 0.3812, 0.4267, 0.5612, 0.5583]]], device='cuda:0') loss_train_step before backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train: 0.7136210562894121 step: 37 running loss: 0.019287055575389515 Train Steps: 37/90 Loss: 0.0193 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366], [0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693], [0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433], [0.6275, 0.4050, 0.9038, 0.3767, 0.3838, 0.3533, 0.7074, 0.5575], [0.6221, 0.4107, 0.7788, 0.3033, 0.3950, 0.2817, 0.6075, 0.5517]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5222, 0.3515, 0.8506, 0.3014, 0.4592, 0.3052, 0.6338, 0.5594], [0.5480, 0.3766, 0.8690, 0.4320, 0.4403, 0.4208, 0.6058, 0.5543], [0.5488, 0.3863, 0.8739, 0.4569, 0.4179, 0.4869, 0.5889, 0.5529], [0.5554, 0.3790, 0.8614, 0.3570, 0.4100, 0.4137, 0.6209, 0.5165], [0.5769, 0.3792, 0.8886, 0.4361, 0.3942, 0.4907, 0.5784, 0.5046], [0.5248, 0.3738, 0.7878, 0.2725, 0.3957, 0.3849, 0.6102, 0.5456], [0.5949, 0.3894, 0.9011, 0.4044, 0.4239, 0.3835, 0.6411, 0.5413], [0.5363, 0.3850, 0.8409, 0.3620, 0.4490, 0.2931, 0.6067, 0.5606]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6226, 0.4098, 0.8913, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366], [0.6277, 0.4036, 0.8687, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6038, 0.3946, 0.8413, 0.4883, 0.3562, 0.4550, 0.5266, 0.4693], [0.6175, 0.4091, 0.7862, 0.2800, 0.3638, 0.3583, 0.6187, 0.5433], [0.6275, 0.4050, 0.9038, 0.3767, 0.3837, 0.3533, 0.7074, 0.5575], [0.6221, 0.4107, 0.7788, 0.3033, 0.3950, 0.2817, 0.6075, 0.5517]]], device='cuda:0') loss_train_step before backward: tensor(0.0029, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0029, device='cuda:0', grad_fn=) loss_train: 0.7165557543048635 step: 38 running loss: 0.018856730376443778 Train Steps: 38/90 Loss: 0.0189 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317], [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285], [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092], [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833], [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896], [0.6125, 0.3974, 0.7725, 0.2517, 0.3538, 0.3317, 0.5887, 0.5500], [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4853, 0.3183, 0.7892, 0.2729, 0.4579, 0.1858, 0.6309, 0.5517], [0.5639, 0.3632, 0.9041, 0.3791, 0.4606, 0.3558, 0.6677, 0.5591], [0.5767, 0.3924, 0.9650, 0.4746, 0.4343, 0.5645, 0.6595, 0.5016], [0.5177, 0.3271, 0.7849, 0.2934, 0.3936, 0.3067, 0.6365, 0.5375], [0.5677, 0.3806, 0.8894, 0.3982, 0.4131, 0.4094, 0.6496, 0.5093], [0.5819, 0.3677, 0.8950, 0.3920, 0.4091, 0.3712, 0.6272, 0.5226], [0.5597, 0.3846, 0.8170, 0.2857, 0.4165, 0.3674, 0.6354, 0.5120], [0.5893, 0.3831, 0.9125, 0.4968, 0.3887, 0.5205, 0.6712, 0.5350]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317], [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285], [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092], [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833], [0.6082, 0.4024, 0.8737, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896], [0.6125, 0.3974, 0.7725, 0.2517, 0.3537, 0.3317, 0.5888, 0.5500], [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373]]], device='cuda:0') loss_train_step before backward: tensor(0.0032, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0032, device='cuda:0', grad_fn=) loss_train: 0.7197442847536877 step: 39 running loss: 0.018454981660350967 Train Steps: 39/90 Loss: 0.0185 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285], [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750], [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5669, 0.3682, 0.8350, 0.2945, 0.4606, 0.2369, 0.6672, 0.5693], [0.5937, 0.3768, 0.8715, 0.4076, 0.3744, 0.4059, 0.6638, 0.4973], [0.5738, 0.3461, 0.9497, 0.3841, 0.3975, 0.4007, 0.6788, 0.5481], [0.5936, 0.3624, 0.8194, 0.3669, 0.3962, 0.3466, 0.6540, 0.5133], [0.5639, 0.3757, 0.8873, 0.4276, 0.4282, 0.4881, 0.6677, 0.5097], [0.6198, 0.4045, 0.8702, 0.4273, 0.4185, 0.3589, 0.6628, 0.5233], [0.5214, 0.3388, 0.7650, 0.2842, 0.4229, 0.2296, 0.6664, 0.5563], [0.6124, 0.3885, 0.8598, 0.3752, 0.3879, 0.3847, 0.6683, 0.5288]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6199, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6203, 0.4096, 0.8863, 0.4267, 0.3537, 0.4117, 0.6025, 0.5650], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285], [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750], [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882]]], device='cuda:0') loss_train_step before backward: tensor(0.0029, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0029, device='cuda:0', grad_fn=) loss_train: 0.7226116416277364 step: 40 running loss: 0.01806529104069341 Train Steps: 40/90 Loss: 0.0181 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6207, 0.4110, 0.8738, 0.5000, 0.4800, 0.5633, 0.6300, 0.5433], [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6138, 0.5450], [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783], [0.6122, 0.3993, 0.8738, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927], [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6200, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6079, 0.3684, 0.8767, 0.4069, 0.4390, 0.4066, 0.6747, 0.5424], [0.5971, 0.3870, 0.9012, 0.4236, 0.3968, 0.4451, 0.6842, 0.5146], [0.6840, 0.4018, 0.8202, 0.3566, 0.3806, 0.2903, 0.6120, 0.5582], [0.6218, 0.3943, 0.8632, 0.3988, 0.4154, 0.3771, 0.6226, 0.5125], [0.6515, 0.3961, 0.8786, 0.3728, 0.3809, 0.2942, 0.6653, 0.5429], [0.6099, 0.3804, 0.8950, 0.4180, 0.4207, 0.4285, 0.6519, 0.5324], [0.5717, 0.3496, 0.7514, 0.2635, 0.4039, 0.1859, 0.6216, 0.5170], [0.5850, 0.3603, 0.8961, 0.3972, 0.3829, 0.3713, 0.6643, 0.5443]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6207, 0.4110, 0.8737, 0.5000, 0.4800, 0.5633, 0.6300, 0.5433], [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6137, 0.5450], [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783], [0.6122, 0.3993, 0.8737, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927], [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6199, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334]]], device='cuda:0') loss_train_step before backward: tensor(0.0025, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0025, device='cuda:0', grad_fn=) loss_train: 0.7251194597920403 step: 41 running loss: 0.01768584048273269 Train Steps: 41/90 Loss: 0.0177 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748], [0.6268, 0.4061, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300], [0.6200, 0.4049, 0.8638, 0.5617, 0.4125, 0.5100, 0.6013, 0.5317], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6058, 0.3986, 0.8324, 0.4626, 0.3838, 0.4983, 0.5147, 0.5466], [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431], [0.6203, 0.4021, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6478, 0.3852, 0.8580, 0.4518, 0.3584, 0.3141, 0.5976, 0.5479], [0.6389, 0.3808, 0.8447, 0.3671, 0.3886, 0.2617, 0.6178, 0.5360], [0.6121, 0.3714, 0.8291, 0.2746, 0.4025, 0.2104, 0.6777, 0.5752], [0.6169, 0.3722, 0.8361, 0.4941, 0.3821, 0.4507, 0.6206, 0.5253], [0.6501, 0.3987, 0.8182, 0.3100, 0.4028, 0.2576, 0.6209, 0.5517], [0.5819, 0.3600, 0.8697, 0.4064, 0.3736, 0.3797, 0.6058, 0.5259], [0.6105, 0.3602, 0.8543, 0.4191, 0.4216, 0.3745, 0.6147, 0.5200], [0.6841, 0.4008, 0.8740, 0.4576, 0.3594, 0.3333, 0.6435, 0.5299]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6336, 0.4191, 0.8938, 0.5167, 0.3938, 0.3517, 0.7343, 0.5748], [0.6268, 0.4060, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300], [0.6199, 0.4049, 0.8637, 0.5617, 0.4125, 0.5100, 0.6012, 0.5317], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6058, 0.3986, 0.8324, 0.4626, 0.3837, 0.4983, 0.5147, 0.5466], [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431], [0.6203, 0.4020, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405]]], device='cuda:0') loss_train_step before backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train: 0.7277624051785097 step: 42 running loss: 0.017327676313774038 Train Steps: 42/90 Loss: 0.0173 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6214, 0.4112, 0.7838, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083], [0.6125, 0.4076, 0.8488, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505], [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667], [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5837, 0.5317], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817], [0.6275, 0.4157, 0.8337, 0.5800, 0.3763, 0.4200, 0.5547, 0.6125]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6859, 0.4183, 0.8201, 0.3667, 0.3846, 0.2885, 0.6160, 0.5481], [0.6493, 0.3851, 0.8564, 0.4168, 0.3520, 0.3721, 0.5779, 0.5315], [0.6808, 0.4184, 0.8806, 0.4314, 0.3619, 0.3917, 0.6199, 0.5227], [0.6688, 0.3939, 0.8313, 0.3791, 0.3818, 0.2768, 0.5978, 0.5413], [0.6302, 0.3876, 0.8173, 0.2933, 0.4438, 0.1975, 0.6074, 0.5589], [0.6045, 0.3703, 0.7571, 0.3427, 0.3783, 0.2547, 0.5605, 0.5320], [0.6031, 0.4094, 0.8788, 0.5251, 0.4059, 0.4233, 0.5849, 0.5355], [0.6389, 0.4025, 0.8686, 0.5786, 0.3450, 0.4725, 0.6304, 0.4850]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6214, 0.4112, 0.7837, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083], [0.6125, 0.4076, 0.8487, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505], [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667], [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5838, 0.5317], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817], [0.6275, 0.4157, 0.8338, 0.5800, 0.3762, 0.4200, 0.5547, 0.6125]]], device='cuda:0') loss_train_step before backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train: 0.7299826672533527 step: 43 running loss: 0.01697634109891518 Train Steps: 43/90 Loss: 0.0170 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494], [0.6250, 0.4103, 0.8950, 0.4400, 0.3912, 0.5650, 0.6050, 0.5133], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810], [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510], [0.6200, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6318, 0.4200, 0.8345, 0.3732, 0.4057, 0.3183, 0.5886, 0.5354], [0.6617, 0.4213, 0.8959, 0.4752, 0.4035, 0.5062, 0.5819, 0.5257], [0.6774, 0.3981, 0.8371, 0.3690, 0.3556, 0.3575, 0.5416, 0.5478], [0.6886, 0.4337, 0.8427, 0.3384, 0.4122, 0.2879, 0.5650, 0.5443], [0.6776, 0.4242, 0.8516, 0.5299, 0.3947, 0.4612, 0.6092, 0.5451], [0.7017, 0.4519, 0.8956, 0.5708, 0.3654, 0.4416, 0.5562, 0.5274], [0.6478, 0.4055, 0.7656, 0.3337, 0.3684, 0.2613, 0.5162, 0.5638], [0.6874, 0.4252, 0.9035, 0.5301, 0.4162, 0.4708, 0.5566, 0.5378]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494], [0.6250, 0.4103, 0.8950, 0.4400, 0.3913, 0.5650, 0.6050, 0.5133], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810], [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510], [0.6199, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913]]], device='cuda:0') loss_train_step before backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train: 0.7321775729069486 step: 44 running loss: 0.01664039938424883 Train Steps: 44/90 Loss: 0.0166 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533], [0.6246, 0.4028, 0.8738, 0.4867, 0.4088, 0.5667, 0.6362, 0.5200], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426], [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667], [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116], [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5502, 0.3357, 0.6939, 0.2983, 0.3869, 0.2363, 0.4876, 0.5458], [0.6379, 0.4245, 0.8473, 0.5687, 0.3954, 0.6164, 0.5743, 0.4970], [0.6755, 0.4439, 0.7752, 0.2956, 0.4185, 0.2184, 0.5482, 0.5289], [0.6546, 0.4260, 0.8546, 0.4950, 0.3709, 0.4749, 0.5239, 0.5447], [0.6950, 0.4629, 0.8671, 0.4698, 0.3533, 0.4084, 0.5627, 0.5403], [0.7052, 0.4755, 0.8520, 0.4392, 0.3554, 0.3897, 0.5525, 0.5459], [0.6257, 0.4419, 0.8624, 0.4957, 0.3983, 0.5694, 0.5611, 0.5239], [0.6736, 0.4606, 0.7992, 0.3360, 0.4000, 0.2437, 0.5425, 0.5643]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533], [0.6246, 0.4028, 0.8737, 0.4867, 0.4087, 0.5667, 0.6363, 0.5200], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6185, 0.4067, 0.8838, 0.4450, 0.4038, 0.4733, 0.5213, 0.5142], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426], [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5987, 0.5667], [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116], [0.6243, 0.4128, 0.7763, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417]]], device='cuda:0') loss_train_step before backward: tensor(0.0085, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0085, device='cuda:0', grad_fn=) loss_train: 0.7407123452285305 step: 45 running loss: 0.01646027433841179 Train Steps: 45/90 Loss: 0.0165 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133], [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6260, 0.4120, 0.8013, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6199, 0.4093, 0.7913, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700], [0.6064, 0.3953, 0.8738, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416], [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333], [0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6275, 0.4261, 0.8645, 0.5512, 0.3968, 0.5612, 0.5912, 0.4915], [0.5661, 0.3730, 0.7460, 0.2466, 0.4075, 0.2641, 0.5022, 0.5740], [0.6225, 0.4330, 0.7935, 0.2694, 0.4527, 0.2330, 0.5499, 0.5673], [0.6655, 0.4337, 0.7820, 0.3664, 0.3560, 0.4036, 0.5483, 0.5219], [0.6356, 0.4358, 0.8026, 0.3195, 0.4113, 0.3123, 0.5415, 0.5637], [0.6347, 0.4366, 0.8743, 0.5053, 0.3707, 0.5440, 0.5326, 0.5076], [0.6398, 0.4466, 0.8624, 0.5047, 0.3825, 0.5448, 0.5553, 0.5348], [0.6460, 0.4478, 0.9060, 0.5175, 0.3901, 0.5797, 0.5691, 0.5219]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6308, 0.3990, 0.8687, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133], [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6259, 0.4120, 0.8012, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6198, 0.4093, 0.7912, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700], [0.6064, 0.3952, 0.8737, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416], [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6137, 0.5333], [0.6200, 0.4112, 0.8863, 0.4100, 0.3638, 0.4917, 0.6087, 0.6050]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.7426065595354885 step: 46 running loss: 0.01614362085946714 Train Steps: 46/90 Loss: 0.0161 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631], [0.6267, 0.4080, 0.8438, 0.2633, 0.4763, 0.1800, 0.6259, 0.5240], [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267], [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485], [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5601, 0.4024, 0.8080, 0.4328, 0.3650, 0.5411, 0.5782, 0.5040], [0.6328, 0.4254, 0.8367, 0.2879, 0.4258, 0.3150, 0.5593, 0.5436], [0.5882, 0.4032, 0.8208, 0.5737, 0.3746, 0.5384, 0.5297, 0.5168], [0.5995, 0.4132, 0.7947, 0.2517, 0.4077, 0.3107, 0.5249, 0.5769], [0.5679, 0.4346, 0.8613, 0.4552, 0.4261, 0.5152, 0.5542, 0.5333], [0.5830, 0.4047, 0.7621, 0.3065, 0.3942, 0.3711, 0.5540, 0.5240], [0.5902, 0.4203, 0.7136, 0.2819, 0.3937, 0.3589, 0.5246, 0.5339], [0.6296, 0.4507, 0.8701, 0.4186, 0.4044, 0.4399, 0.5818, 0.5159]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631], [0.6267, 0.4080, 0.8438, 0.2633, 0.4762, 0.1800, 0.6259, 0.5240], [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267], [0.0000, 0.0000, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6123, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485], [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6263, 0.5167]]], device='cuda:0') loss_train_step before backward: tensor(0.0098, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0098, device='cuda:0', grad_fn=) loss_train: 0.7524001447018236 step: 47 running loss: 0.016008513717060077 Train Steps: 47/90 Loss: 0.0160 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567], [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389], [0.6286, 0.4060, 0.9188, 0.4333, 0.3675, 0.4167, 0.7034, 0.5528], [0.6193, 0.3930, 0.8949, 0.4437, 0.3852, 0.5435, 0.6263, 0.5263], [0.6293, 0.3982, 0.8700, 0.5300, 0.3763, 0.4717, 0.7050, 0.5297], [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5769, 0.4040, 0.7802, 0.1991, 0.4471, 0.2531, 0.5674, 0.5500], [0.5461, 0.3844, 0.7615, 0.2243, 0.4086, 0.2583, 0.5673, 0.5589], [0.5276, 0.3678, 0.8232, 0.4573, 0.4060, 0.5099, 0.5400, 0.5218], [0.5709, 0.4187, 0.7707, 0.2387, 0.3961, 0.2889, 0.5726, 0.5709], [0.5552, 0.3937, 0.8630, 0.3850, 0.4013, 0.4375, 0.5807, 0.5623], [0.5481, 0.3639, 0.8349, 0.4752, 0.3836, 0.5852, 0.5817, 0.4908], [0.5516, 0.3764, 0.8259, 0.5291, 0.3828, 0.5496, 0.5861, 0.5209], [0.5723, 0.4223, 0.8104, 0.3714, 0.3790, 0.3846, 0.5357, 0.5459]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5688, 0.5567], [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389], [0.6286, 0.4060, 0.9187, 0.4333, 0.3675, 0.4167, 0.7034, 0.5528], [0.6193, 0.3930, 0.8949, 0.4437, 0.3852, 0.5435, 0.6263, 0.5263], [0.6293, 0.3982, 0.8700, 0.5300, 0.3762, 0.4717, 0.7050, 0.5297], [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650]]], device='cuda:0') loss_train_step before backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train: 0.7547844997607172 step: 48 running loss: 0.015724677078348275 Train Steps: 48/90 Loss: 0.0157 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426], [0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921], [0.6203, 0.4078, 0.8800, 0.5083, 0.3900, 0.5000, 0.6100, 0.5583], [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116], [0.6087, 0.3976, 0.8337, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300], [0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4047, 0.2704, 0.7542, 0.2308, 0.4533, 0.2697, 0.5537, 0.5551], [0.5581, 0.3529, 0.7849, 0.2889, 0.4571, 0.3005, 0.5780, 0.5572], [0.6089, 0.4270, 0.8583, 0.2296, 0.4496, 0.3126, 0.6338, 0.5481], [0.5649, 0.3624, 0.9076, 0.5557, 0.3788, 0.6005, 0.5847, 0.5724], [0.5639, 0.3667, 0.7899, 0.3049, 0.3963, 0.3105, 0.6038, 0.5304], [0.5850, 0.4091, 0.8458, 0.3845, 0.3912, 0.4504, 0.5918, 0.5393], [0.5680, 0.3801, 0.8733, 0.4001, 0.4094, 0.3984, 0.6094, 0.5167], [0.5417, 0.3740, 0.7820, 0.2999, 0.4156, 0.3376, 0.5991, 0.5239]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426], [0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921], [0.6203, 0.4078, 0.8800, 0.5083, 0.3900, 0.5000, 0.6100, 0.5583], [0.6165, 0.4106, 0.7575, 0.1733, 0.3837, 0.2650, 0.5680, 0.5116], [0.6087, 0.3976, 0.8338, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300], [0.6277, 0.4036, 0.8687, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212]]], device='cuda:0') loss_train_step before backward: tensor(0.0062, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0062, device='cuda:0', grad_fn=) loss_train: 0.760986584238708 step: 49 running loss: 0.015530338453851184 Train Steps: 49/90 Loss: 0.0155 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133], [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250], [0.6125, 0.3974, 0.7725, 0.2517, 0.3538, 0.3317, 0.5887, 0.5500], [0.6275, 0.4050, 0.9038, 0.3767, 0.3838, 0.3533, 0.7074, 0.5575], [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5269, 0.3406, 0.8398, 0.3721, 0.4167, 0.4091, 0.5909, 0.5420], [0.5514, 0.3257, 0.8563, 0.4422, 0.3974, 0.4406, 0.6472, 0.5195], [0.5518, 0.3459, 0.7828, 0.2487, 0.3847, 0.2868, 0.6127, 0.5394], [0.5655, 0.3462, 0.8768, 0.3768, 0.4197, 0.4672, 0.6450, 0.5594], [0.5349, 0.3501, 0.7877, 0.2137, 0.3759, 0.2686, 0.5964, 0.5461], [0.5685, 0.3744, 0.9011, 0.2612, 0.4193, 0.2459, 0.6702, 0.5458], [0.5460, 0.3551, 0.8509, 0.4093, 0.3993, 0.4086, 0.6418, 0.5202], [0.5323, 0.3448, 0.8462, 0.3966, 0.4335, 0.3736, 0.5903, 0.5409]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6308, 0.3990, 0.8687, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133], [0.6102, 0.4001, 0.7738, 0.3583, 0.3462, 0.3800, 0.5524, 0.5689], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250], [0.6125, 0.3974, 0.7725, 0.2517, 0.3537, 0.3317, 0.5888, 0.5500], [0.6275, 0.4050, 0.9038, 0.3767, 0.3837, 0.3533, 0.7074, 0.5575], [0.6299, 0.4008, 0.8450, 0.5350, 0.4212, 0.5000, 0.6350, 0.5100], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817]]], device='cuda:0') loss_train_step before backward: tensor(0.0044, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0044, device='cuda:0', grad_fn=) loss_train: 0.765395499765873 step: 50 running loss: 0.015307909995317458 Train Steps: 50/90 Loss: 0.0153 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954], [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6138, 0.5450], [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397], [0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461], [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350], [0.6115, 0.4081, 0.6725, 0.2433, 0.4088, 0.1933, 0.5167, 0.5544], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5544, 0.3628, 0.8616, 0.2909, 0.4169, 0.2274, 0.6133, 0.5377], [0.5905, 0.3875, 0.9099, 0.4365, 0.3978, 0.4963, 0.6773, 0.5268], [0.4808, 0.2976, 0.7408, 0.1998, 0.4108, 0.1856, 0.6188, 0.5373], [0.4906, 0.3093, 0.7173, 0.2025, 0.3813, 0.2083, 0.5842, 0.5237], [0.5996, 0.3735, 0.8806, 0.4638, 0.4183, 0.4209, 0.6550, 0.5542], [0.5829, 0.3653, 0.9094, 0.4636, 0.4024, 0.4556, 0.6436, 0.5377], [0.4791, 0.3142, 0.7376, 0.1940, 0.3909, 0.1909, 0.5938, 0.5389], [0.5878, 0.3725, 0.8417, 0.4316, 0.4024, 0.4445, 0.6350, 0.5331]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6137, 0.4038, 0.8562, 0.4050, 0.3812, 0.2550, 0.5106, 0.4954], [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6137, 0.5450], [0.6159, 0.4085, 0.6900, 0.2283, 0.4087, 0.1950, 0.5123, 0.5397], [0.6149, 0.4054, 0.6712, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461], [0.6126, 0.3954, 0.8537, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350], [0.6115, 0.4081, 0.6725, 0.2433, 0.4087, 0.1933, 0.5167, 0.5544], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617]]], device='cuda:0') loss_train_step before backward: tensor(0.0035, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0035, device='cuda:0', grad_fn=) loss_train: 0.7689229550305754 step: 51 running loss: 0.015076920686874027 Train Steps: 51/90 Loss: 0.0151 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6188, 0.5433], [0.6263, 0.4030, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771], [0.6371, 0.4092, 0.8337, 0.5850, 0.3950, 0.5117, 0.6559, 0.5262], [0.6100, 0.4071, 0.7601, 0.3444, 0.3400, 0.4117, 0.5625, 0.5617], [0.6286, 0.4055, 0.9000, 0.4717, 0.3763, 0.4683, 0.7018, 0.5494], [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250], [0.6183, 0.4076, 0.8838, 0.4517, 0.3813, 0.4483, 0.5775, 0.5633], [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5841, 0.3481, 0.8627, 0.4467, 0.4061, 0.4344, 0.6089, 0.5218], [0.5990, 0.3450, 0.8709, 0.4506, 0.3757, 0.4238, 0.5918, 0.5000], [0.5780, 0.3840, 0.8173, 0.4550, 0.3951, 0.3989, 0.6279, 0.5487], [0.5716, 0.3701, 0.7407, 0.2549, 0.3653, 0.2957, 0.6200, 0.5060], [0.6063, 0.3638, 0.8475, 0.3999, 0.3978, 0.3789, 0.6600, 0.5402], [0.6086, 0.3748, 0.8693, 0.3564, 0.3924, 0.2627, 0.6464, 0.5142], [0.5666, 0.3266, 0.8481, 0.4245, 0.3842, 0.3965, 0.6189, 0.5329], [0.5406, 0.3473, 0.7894, 0.1998, 0.4322, 0.1229, 0.6078, 0.5353]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6187, 0.5433], [0.6263, 0.4029, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771], [0.6371, 0.4092, 0.8338, 0.5850, 0.3950, 0.5117, 0.6559, 0.5262], [0.6100, 0.4071, 0.7601, 0.3444, 0.3400, 0.4117, 0.5625, 0.5617], [0.6286, 0.4055, 0.9000, 0.4717, 0.3762, 0.4683, 0.7018, 0.5494], [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250], [0.6183, 0.4076, 0.8838, 0.4517, 0.3812, 0.4483, 0.5775, 0.5633], [0.6229, 0.4107, 0.8138, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400]]], device='cuda:0') loss_train_step before backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train: 0.7715486923698336 step: 52 running loss: 0.01483747485326603 Train Steps: 52/90 Loss: 0.0148 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6277, 0.4083, 0.8350, 0.2717, 0.4562, 0.1800, 0.5918, 0.4878], [0.6275, 0.4050, 0.9038, 0.3767, 0.3838, 0.3533, 0.7074, 0.5575], [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6262, 0.5367], [0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133], [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170], [ nan, nan, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5546, 0.3509, 0.7689, 0.2416, 0.4326, 0.1638, 0.5948, 0.5102], [0.6363, 0.4136, 0.8775, 0.4243, 0.3832, 0.2892, 0.6573, 0.5144], [0.6187, 0.3739, 0.8605, 0.6347, 0.3554, 0.5029, 0.5993, 0.5492], [0.5874, 0.3769, 0.7346, 0.3428, 0.3623, 0.3021, 0.6076, 0.5091], [0.6045, 0.3746, 0.8653, 0.5789, 0.3946, 0.5484, 0.6011, 0.5000], [0.6121, 0.3957, 0.7958, 0.3137, 0.3871, 0.2547, 0.6147, 0.5380], [0.5961, 0.3638, 0.7663, 0.2782, 0.4365, 0.2224, 0.6491, 0.5104], [0.5224, 0.3245, 0.8160, 0.2858, 0.4545, 0.2050, 0.6430, 0.5117]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6277, 0.4083, 0.8350, 0.2717, 0.4563, 0.1800, 0.5918, 0.4878], [0.6275, 0.4050, 0.9038, 0.3767, 0.3837, 0.3533, 0.7074, 0.5575], [0.6222, 0.4171, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6263, 0.5367], [0.6205, 0.4015, 0.8350, 0.2717, 0.3988, 0.2550, 0.5788, 0.5133], [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170], [0.0000, 0.0000, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737]]], device='cuda:0') loss_train_step before backward: tensor(0.0079, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0079, device='cuda:0', grad_fn=) loss_train: 0.7794073515105993 step: 53 running loss: 0.014705799085105647 Train Steps: 53/90 Loss: 0.0147 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960], [0.6115, 0.4081, 0.6725, 0.2433, 0.4088, 0.1933, 0.5167, 0.5544], [ nan, nan, 0.6793, 0.2110, 0.4012, 0.2167, 0.5112, 0.5583], [0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200], [0.6099, 0.4030, 0.8638, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6415, 0.4262, 0.8866, 0.2842, 0.4599, 0.2407, 0.6511, 0.5367], [0.5397, 0.3495, 0.7260, 0.2746, 0.3897, 0.2141, 0.5626, 0.5319], [0.4686, 0.3070, 0.6932, 0.2379, 0.4155, 0.1864, 0.5605, 0.5481], [0.6670, 0.4144, 0.8815, 0.4926, 0.3525, 0.4148, 0.5968, 0.5435], [0.6459, 0.4183, 0.8974, 0.6042, 0.3895, 0.4822, 0.6102, 0.5310], [0.6237, 0.3969, 0.8895, 0.5081, 0.4466, 0.4090, 0.6234, 0.5345], [0.6468, 0.4130, 0.8978, 0.5317, 0.4216, 0.4910, 0.6532, 0.5334], [0.6487, 0.4075, 0.7803, 0.3463, 0.3606, 0.3783, 0.6201, 0.5232]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6259, 0.4156, 0.8813, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960], [0.6115, 0.4081, 0.6725, 0.2433, 0.4087, 0.1933, 0.5167, 0.5544], [0.0000, 0.0000, 0.6793, 0.2110, 0.4013, 0.2167, 0.5113, 0.5583], [0.6151, 0.4125, 0.8737, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200], [0.6098, 0.4030, 0.8637, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6197, 0.4051, 0.7812, 0.2650, 0.3512, 0.4050, 0.6112, 0.5500]]], device='cuda:0') loss_train_step before backward: tensor(0.0063, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0063, device='cuda:0', grad_fn=) loss_train: 0.7857216421980411 step: 54 running loss: 0.014550400781445205 Train Steps: 54/90 Loss: 0.0146 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6362, 0.5000], [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378], [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864], [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250], [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012], [0.6273, 0.4100, 0.7137, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6358, 0.4032, 0.8800, 0.5148, 0.4160, 0.5419, 0.5985, 0.5544], [0.6271, 0.4309, 0.8779, 0.4180, 0.4278, 0.2989, 0.6445, 0.5209], [0.6354, 0.4176, 0.8796, 0.4386, 0.3965, 0.3395, 0.5999, 0.5642], [0.5834, 0.3736, 0.8202, 0.4424, 0.3896, 0.4175, 0.5881, 0.5508], [0.6109, 0.4081, 0.8717, 0.4947, 0.4118, 0.4414, 0.6274, 0.5529], [0.6070, 0.4307, 0.8827, 0.4406, 0.4242, 0.4110, 0.6265, 0.5525], [0.6422, 0.4216, 0.8492, 0.5282, 0.4020, 0.4422, 0.6189, 0.5158], [0.5524, 0.3625, 0.7314, 0.2355, 0.4092, 0.1922, 0.5738, 0.5495]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6363, 0.5000], [0.6307, 0.4029, 0.8988, 0.4817, 0.3938, 0.3500, 0.7311, 0.5378], [0.6252, 0.4162, 0.8662, 0.4467, 0.3625, 0.3567, 0.6037, 0.5533], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6314, 0.4107, 0.8750, 0.5100, 0.3787, 0.4900, 0.7121, 0.5864], [0.6263, 0.4065, 0.9038, 0.4317, 0.3587, 0.4550, 0.6325, 0.5250], [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012], [0.6273, 0.4099, 0.7138, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.7869276243727654 step: 55 running loss: 0.014307774988595735 Train Steps: 55/90 Loss: 0.0143 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833], [0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585], [0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913], [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799], [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400], [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5301, 0.3866, 0.7485, 0.3359, 0.3693, 0.2995, 0.5671, 0.5649], [0.6037, 0.4085, 0.8429, 0.4616, 0.4316, 0.4904, 0.6046, 0.5223], [0.5098, 0.3556, 0.8043, 0.2597, 0.4651, 0.1535, 0.5834, 0.5238], [0.6246, 0.4276, 0.9140, 0.5233, 0.4322, 0.5490, 0.6235, 0.5671], [0.5730, 0.4027, 0.8279, 0.3740, 0.3846, 0.2692, 0.5577, 0.5516], [0.6058, 0.3992, 0.8399, 0.3735, 0.3873, 0.4737, 0.6092, 0.5630], [0.6692, 0.4559, 0.9170, 0.5879, 0.3788, 0.4056, 0.6201, 0.5233], [0.6449, 0.4282, 0.8855, 0.5277, 0.4079, 0.5283, 0.6165, 0.5521]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833], [0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585], [0.6286, 0.4078, 0.8062, 0.2267, 0.4787, 0.1533, 0.5953, 0.4913], [0.6241, 0.4142, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.6200, 0.4118, 0.8288, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799], [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400], [0.6272, 0.4071, 0.8737, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.7881838389439508 step: 56 running loss: 0.014074711409713407 Train Steps: 56/90 Loss: 0.0141 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6117, 0.4019, 0.8538, 0.4067, 0.3513, 0.3583, 0.5663, 0.5133], [0.6075, 0.4007, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280], [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500], [0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494], [0.6202, 0.4079, 0.8025, 0.2500, 0.3763, 0.3217, 0.6125, 0.5533], [0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600], [0.6179, 0.4082, 0.6688, 0.2667, 0.3588, 0.3317, 0.5750, 0.5783]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6256, 0.4479, 0.9064, 0.4920, 0.3685, 0.3892, 0.6073, 0.5520], [0.6175, 0.4390, 0.8795, 0.5274, 0.4236, 0.5354, 0.5843, 0.5512], [0.6787, 0.4470, 0.9259, 0.5724, 0.4245, 0.5948, 0.5997, 0.5572], [0.6151, 0.4132, 0.8721, 0.4313, 0.4208, 0.4392, 0.6534, 0.5397], [0.6369, 0.4252, 0.8739, 0.3726, 0.4443, 0.3527, 0.6142, 0.5384], [0.5044, 0.3321, 0.7214, 0.3021, 0.4216, 0.3138, 0.5538, 0.5532], [0.5753, 0.4049, 0.7476, 0.3612, 0.3829, 0.3842, 0.6129, 0.5504], [0.5940, 0.4013, 0.7566, 0.3702, 0.3730, 0.3629, 0.5568, 0.5448]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6116, 0.4019, 0.8537, 0.4067, 0.3512, 0.3583, 0.5663, 0.5133], [0.6075, 0.4006, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280], [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5838, 0.5500], [0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494], [0.6202, 0.4079, 0.8025, 0.2500, 0.3762, 0.3217, 0.6125, 0.5533], [0.6270, 0.4266, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [0.6132, 0.4066, 0.7259, 0.2402, 0.3587, 0.3300, 0.6000, 0.5600], [0.6179, 0.4082, 0.6687, 0.2667, 0.3587, 0.3317, 0.5750, 0.5783]]], device='cuda:0') loss_train_step before backward: tensor(0.0031, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0031, device='cuda:0', grad_fn=) loss_train: 0.7912858905037865 step: 57 running loss: 0.013882208605329589 Train Steps: 57/90 Loss: 0.0139 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123], [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117], [0.6179, 0.4008, 0.8600, 0.4015, 0.3932, 0.2515, 0.5711, 0.5438], [ nan, nan, 0.7725, 0.2611, 0.3675, 0.2733, 0.5413, 0.5167], [0.6259, 0.4133, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6118, 0.4052, 0.8463, 0.3917, 0.3538, 0.3450, 0.5053, 0.5593]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5422, 0.3630, 0.8784, 0.3457, 0.4616, 0.3802, 0.6294, 0.5449], [0.5507, 0.3724, 0.7493, 0.3501, 0.3832, 0.3287, 0.5473, 0.5549], [0.6227, 0.4195, 0.7417, 0.3306, 0.3928, 0.3698, 0.5464, 0.5404], [0.7243, 0.4714, 0.9390, 0.5854, 0.3803, 0.6338, 0.6083, 0.5526], [0.5849, 0.4123, 0.8004, 0.3820, 0.4038, 0.3513, 0.5566, 0.5422], [0.5595, 0.3575, 0.7537, 0.3040, 0.3828, 0.3033, 0.5416, 0.5295], [0.5844, 0.3820, 0.8241, 0.2918, 0.4535, 0.3019, 0.6126, 0.5243], [0.6423, 0.4446, 0.8735, 0.4902, 0.3554, 0.5000, 0.5660, 0.5697]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6162, 0.4134, 0.6700, 0.2467, 0.3963, 0.2533, 0.5738, 0.5467], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123], [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117], [0.6179, 0.4008, 0.8600, 0.4015, 0.3932, 0.2515, 0.5711, 0.5438], [0.0000, 0.0000, 0.7725, 0.2611, 0.3675, 0.2733, 0.5412, 0.5167], [0.6259, 0.4132, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6118, 0.4052, 0.8462, 0.3917, 0.3537, 0.3450, 0.5053, 0.5593]]], device='cuda:0') loss_train_step before backward: tensor(0.0170, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0170, device='cuda:0', grad_fn=) loss_train: 0.8083162078401074 step: 58 running loss: 0.013936486342070818 Train Steps: 58/90 Loss: 0.0139 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6115, 0.4005, 0.8838, 0.3867, 0.3763, 0.4700, 0.5800, 0.5550], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407], [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246], [0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667], [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350], [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822], [0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6941, 0.4152, 0.8931, 0.4134, 0.3910, 0.5324, 0.5766, 0.5431], [0.5465, 0.3455, 0.7323, 0.2617, 0.4144, 0.2806, 0.5498, 0.5433], [0.5984, 0.3990, 0.8667, 0.3813, 0.4116, 0.3690, 0.5511, 0.5340], [0.4598, 0.3009, 0.7508, 0.1999, 0.4238, 0.2400, 0.5251, 0.5418], [0.6028, 0.3708, 0.7092, 0.2750, 0.3748, 0.3487, 0.5502, 0.5231], [0.6448, 0.4068, 0.8794, 0.5131, 0.4047, 0.5059, 0.5803, 0.5123], [0.6070, 0.3968, 0.8930, 0.4225, 0.3829, 0.4504, 0.5513, 0.5751], [0.6185, 0.3836, 0.8760, 0.4788, 0.4527, 0.4987, 0.5625, 0.5476]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6115, 0.4005, 0.8838, 0.3867, 0.3762, 0.4700, 0.5800, 0.5550], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6133, 0.4094, 0.8495, 0.4028, 0.3587, 0.3200, 0.5003, 0.5407], [0.0000, 0.0000, 0.6900, 0.1917, 0.3938, 0.2367, 0.5240, 0.5246], [0.6134, 0.4090, 0.6926, 0.2819, 0.3537, 0.3233, 0.5562, 0.5667], [0.6143, 0.3982, 0.8650, 0.4883, 0.3913, 0.4317, 0.5315, 0.5350], [0.6182, 0.4058, 0.8737, 0.4350, 0.3562, 0.3400, 0.5290, 0.5822], [0.6229, 0.4066, 0.8512, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350]]], device='cuda:0') loss_train_step before backward: tensor(0.0060, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0060, device='cuda:0', grad_fn=) loss_train: 0.8143562333425507 step: 59 running loss: 0.013802648022755097 Train Steps: 59/90 Loss: 0.0138 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540], [0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183], [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567], [0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999], [0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693], [0.6132, 0.4037, 0.6963, 0.2217, 0.4100, 0.1950, 0.5395, 0.5175]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6058, 0.3606, 0.8605, 0.3885, 0.4073, 0.3083, 0.5800, 0.5317], [0.5016, 0.3565, 0.7446, 0.2448, 0.4139, 0.2543, 0.5300, 0.5583], [0.6240, 0.3779, 0.8711, 0.4626, 0.3618, 0.4597, 0.5244, 0.5394], [0.5254, 0.3303, 0.8582, 0.3840, 0.4548, 0.4209, 0.5505, 0.5425], [0.5991, 0.3819, 0.8540, 0.4017, 0.3868, 0.4029, 0.5255, 0.5292], [0.6212, 0.3578, 0.8351, 0.3708, 0.3966, 0.4988, 0.5554, 0.5446], [0.6145, 0.3990, 0.8441, 0.4090, 0.3736, 0.4491, 0.5455, 0.5312], [0.4360, 0.2643, 0.6835, 0.1788, 0.4104, 0.1723, 0.5302, 0.5247]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540], [0.6204, 0.4049, 0.7975, 0.2700, 0.3938, 0.2567, 0.5700, 0.5183], [0.6236, 0.3965, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6112, 0.4029, 0.8637, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567], [0.6108, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6068, 0.3963, 0.8650, 0.4317, 0.4038, 0.5083, 0.5253, 0.4999], [0.6038, 0.3946, 0.8413, 0.4883, 0.3562, 0.4550, 0.5266, 0.4693], [0.6132, 0.4037, 0.6963, 0.2217, 0.4100, 0.1950, 0.5395, 0.5175]]], device='cuda:0') loss_train_step before backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train: 0.8169467503903434 step: 60 running loss: 0.01361577917317239 Train Steps: 60/90 Loss: 0.0136 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6234, 0.4179, 0.7825, 0.3450, 0.3813, 0.2867, 0.5675, 0.5617], [0.6090, 0.4010, 0.7838, 0.3483, 0.3538, 0.3783, 0.5462, 0.5077], [0.6109, 0.4015, 0.7668, 0.3639, 0.3513, 0.3667, 0.5200, 0.5641], [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167], [0.6200, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249], [0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4491, 0.2910, 0.7879, 0.2603, 0.4166, 0.1751, 0.5204, 0.5405], [0.5714, 0.3382, 0.7695, 0.2836, 0.3397, 0.3514, 0.5478, 0.5101], [0.5415, 0.3494, 0.7559, 0.2859, 0.3479, 0.3300, 0.5355, 0.5410], [0.5705, 0.3587, 0.8571, 0.4829, 0.3522, 0.4165, 0.5035, 0.5268], [0.5447, 0.3654, 0.8386, 0.2982, 0.4155, 0.3437, 0.5604, 0.5473], [0.5558, 0.3338, 0.8648, 0.4201, 0.3983, 0.3573, 0.5427, 0.5397], [0.6101, 0.3714, 0.8259, 0.4802, 0.3784, 0.4516, 0.5814, 0.5289], [0.5437, 0.3351, 0.8664, 0.3020, 0.3978, 0.2938, 0.5660, 0.5327]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6234, 0.4179, 0.7825, 0.3450, 0.3812, 0.2867, 0.5675, 0.5617], [0.6090, 0.4010, 0.7837, 0.3483, 0.3537, 0.3783, 0.5462, 0.5077], [0.6109, 0.4015, 0.7668, 0.3639, 0.3512, 0.3667, 0.5200, 0.5641], [0.6236, 0.3965, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6037, 0.6167], [0.6201, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249], [0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901]]], device='cuda:0') loss_train_step before backward: tensor(0.0032, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0032, device='cuda:0', grad_fn=) loss_train: 0.8201047728070989 step: 61 running loss: 0.013444340537821294 Train Steps: 61/90 Loss: 0.0134 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167], [ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383], [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283], [0.6218, 0.4098, 0.7238, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5500, 0.3635, 0.7813, 0.2577, 0.3925, 0.3167, 0.5751, 0.5632], [0.3665, 0.2547, 0.7070, 0.2034, 0.3935, 0.1840, 0.5407, 0.5374], [0.6265, 0.3912, 0.8983, 0.4880, 0.3373, 0.4178, 0.5388, 0.5015], [0.5959, 0.3870, 0.8860, 0.4874, 0.3806, 0.4561, 0.5588, 0.5384], [0.5975, 0.3805, 0.8572, 0.4717, 0.3760, 0.3946, 0.5776, 0.5321], [0.6113, 0.3528, 0.8609, 0.4085, 0.3538, 0.4293, 0.5591, 0.5227], [0.5645, 0.3679, 0.8989, 0.3874, 0.3639, 0.3078, 0.5889, 0.5240], [0.4978, 0.3402, 0.7412, 0.1953, 0.4422, 0.2193, 0.5693, 0.5261]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167], [0.0000, 0.0000, 0.7268, 0.2333, 0.4125, 0.1933, 0.5113, 0.5383], [0.6236, 0.3965, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6163, 0.4001, 0.8788, 0.5033, 0.4013, 0.4633, 0.5337, 0.5767], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283], [0.6218, 0.4098, 0.7237, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350]]], device='cuda:0') loss_train_step before backward: tensor(0.0045, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0045, device='cuda:0', grad_fn=) loss_train: 0.824571170960553 step: 62 running loss: 0.01329953501549279 Train Steps: 62/90 Loss: 0.0133 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6128, 0.4118, 0.8638, 0.5333, 0.4625, 0.5267, 0.5193, 0.5475], [0.6197, 0.4118, 0.8688, 0.5517, 0.4037, 0.5233, 0.5875, 0.5600], [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117], [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167], [0.6336, 0.4086, 0.8900, 0.3950, 0.3900, 0.2950, 0.6504, 0.5066], [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915], [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5856, 0.3832, 0.8667, 0.5038, 0.3929, 0.4306, 0.5876, 0.5351], [0.6264, 0.3945, 0.8378, 0.5135, 0.3571, 0.5204, 0.6011, 0.5409], [0.4624, 0.3228, 0.7371, 0.2446, 0.3710, 0.2221, 0.5621, 0.5435], [0.4359, 0.2948, 0.7194, 0.2421, 0.3845, 0.2105, 0.5463, 0.5387], [0.5081, 0.3626, 0.7719, 0.2733, 0.3700, 0.3125, 0.5887, 0.5739], [0.5906, 0.3802, 0.8621, 0.3707, 0.3631, 0.3314, 0.6340, 0.4989], [0.5955, 0.3969, 0.8825, 0.4975, 0.3163, 0.2913, 0.5767, 0.5016], [0.4488, 0.3219, 0.8165, 0.2282, 0.4266, 0.1937, 0.6143, 0.5187]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6128, 0.4118, 0.8637, 0.5333, 0.4625, 0.5267, 0.5193, 0.5475], [0.6197, 0.4118, 0.8687, 0.5517, 0.4038, 0.5233, 0.5875, 0.5600], [0.6108, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5788, 0.5117], [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167], [0.6336, 0.4086, 0.8900, 0.3950, 0.3900, 0.2950, 0.6504, 0.5066], [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915], [0.6276, 0.4095, 0.8238, 0.2250, 0.4663, 0.1783, 0.6171, 0.4869]]], device='cuda:0') loss_train_step before backward: tensor(0.0030, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0030, device='cuda:0', grad_fn=) loss_train: 0.8275592027930543 step: 63 running loss: 0.013135860361794512 Train Steps: 63/90 Loss: 0.0131 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496], [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6223, 0.4028, 0.8988, 0.4200, 0.3763, 0.5733, 0.6375, 0.5167], [0.6224, 0.4061, 0.8988, 0.4300, 0.3838, 0.4750, 0.6112, 0.5483], [0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6083, 0.3957, 0.8638, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4981, 0.3343, 0.8217, 0.2960, 0.3723, 0.1908, 0.6148, 0.5354], [0.4779, 0.3498, 0.6806, 0.3148, 0.3164, 0.2243, 0.5684, 0.5736], [0.5703, 0.4017, 0.8371, 0.4836, 0.3770, 0.4784, 0.6296, 0.5814], [0.4624, 0.3337, 0.7966, 0.2725, 0.4403, 0.1343, 0.6080, 0.5010], [0.6624, 0.4184, 0.8434, 0.4282, 0.3695, 0.5177, 0.6353, 0.5406], [0.6169, 0.4147, 0.8715, 0.4719, 0.3306, 0.4530, 0.6164, 0.5171], [0.5956, 0.3959, 0.8104, 0.3552, 0.3451, 0.2360, 0.6289, 0.5024], [0.5393, 0.3685, 0.8544, 0.4550, 0.3720, 0.3635, 0.5987, 0.5363]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496], [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6223, 0.4028, 0.8988, 0.4200, 0.3762, 0.5733, 0.6375, 0.5167], [0.6224, 0.4061, 0.8988, 0.4300, 0.3837, 0.4750, 0.6112, 0.5483], [0.6277, 0.4036, 0.8687, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6083, 0.3957, 0.8637, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980]]], device='cuda:0') loss_train_step before backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train: 0.830406641936861 step: 64 running loss: 0.012975103780263453 Train Steps: 64/90 Loss: 0.0130 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350], [0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6218, 0.4098, 0.7238, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350], [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426], [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050], [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397], [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6362, 0.5200]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5603, 0.4120, 0.7911, 0.3092, 0.4101, 0.2705, 0.6473, 0.5467], [0.5694, 0.3968, 0.8457, 0.4254, 0.3745, 0.2732, 0.5703, 0.5492], [0.6512, 0.4446, 0.9230, 0.5599, 0.3967, 0.4851, 0.6733, 0.5388], [0.5368, 0.4146, 0.7461, 0.2814, 0.4314, 0.2485, 0.6166, 0.5398], [0.5517, 0.3704, 0.7644, 0.3018, 0.4405, 0.2130, 0.5926, 0.5472], [0.6188, 0.4373, 0.8200, 0.3571, 0.3411, 0.3280, 0.6295, 0.5389], [0.4397, 0.3159, 0.7076, 0.2952, 0.4046, 0.2003, 0.5565, 0.5434], [0.6508, 0.4446, 0.8944, 0.4353, 0.3638, 0.5242, 0.6494, 0.5329]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6229, 0.4066, 0.7613, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350], [0.6186, 0.4080, 0.8625, 0.3483, 0.3787, 0.2650, 0.5320, 0.5272], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6218, 0.4098, 0.7237, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350], [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426], [0.6135, 0.3994, 0.7912, 0.3050, 0.3625, 0.3050, 0.5838, 0.5050], [0.6159, 0.4085, 0.6900, 0.2283, 0.4087, 0.1950, 0.5123, 0.5397], [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6363, 0.5200]]], device='cuda:0') loss_train_step before backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train: 0.832502335193567 step: 65 running loss: 0.012807728233747184 Train Steps: 65/90 Loss: 0.0128 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6286, 0.4060, 0.9188, 0.4333, 0.3675, 0.4167, 0.7034, 0.5528], [0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702], [ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767], [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395], [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6186, 0.4025, 0.9083, 0.4060, 0.3845, 0.3154, 0.6452, 0.5331], [0.7243, 0.4844, 0.8551, 0.5121, 0.3872, 0.4598, 0.6464, 0.5225], [0.6315, 0.4299, 0.7698, 0.2274, 0.3831, 0.2722, 0.6215, 0.5394], [0.6787, 0.4502, 0.8737, 0.4614, 0.4429, 0.4977, 0.6371, 0.5566], [0.6768, 0.4684, 0.8210, 0.4514, 0.4072, 0.4349, 0.6284, 0.5537], [0.5218, 0.3466, 0.7645, 0.2660, 0.4297, 0.2230, 0.5633, 0.5546], [0.5632, 0.3770, 0.7442, 0.2615, 0.4154, 0.1907, 0.6021, 0.5459], [0.6338, 0.4305, 0.8497, 0.4663, 0.4445, 0.4249, 0.6135, 0.5289]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6286, 0.4060, 0.9187, 0.4333, 0.3675, 0.4167, 0.7034, 0.5528], [0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6346, 0.4086, 0.7937, 0.5500, 0.3963, 0.4867, 0.7343, 0.5702], [0.0000, 0.0000, 0.7613, 0.3250, 0.4038, 0.2533, 0.5437, 0.5767], [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395], [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050]]], device='cuda:0') loss_train_step before backward: tensor(0.0078, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0078, device='cuda:0', grad_fn=) loss_train: 0.8403434584615752 step: 66 running loss: 0.0127324766433572 Train Steps: 66/90 Loss: 0.0127 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6144, 0.4032, 0.8563, 0.3283, 0.3525, 0.4200, 0.5775, 0.5583], [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058], [0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767], [0.6248, 0.4032, 0.7738, 0.1900, 0.4813, 0.1400, 0.5941, 0.4904], [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133], [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6932, 0.4450, 0.8531, 0.3368, 0.3961, 0.3903, 0.6321, 0.5604], [0.7086, 0.4471, 0.8215, 0.3704, 0.4030, 0.4471, 0.6338, 0.5231], [0.6424, 0.4138, 0.8391, 0.3307, 0.4225, 0.2711, 0.6367, 0.5248], [0.6202, 0.4266, 0.8456, 0.4140, 0.4342, 0.4385, 0.6412, 0.5650], [0.6707, 0.4512, 0.8489, 0.4675, 0.4233, 0.4563, 0.6271, 0.5644], [0.5194, 0.3450, 0.6983, 0.1942, 0.4850, 0.1519, 0.6323, 0.5231], [0.6732, 0.4427, 0.8219, 0.4614, 0.4517, 0.4596, 0.6375, 0.5391], [0.7076, 0.4632, 0.8452, 0.4561, 0.4193, 0.4764, 0.6555, 0.5310]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6144, 0.4032, 0.8562, 0.3283, 0.3525, 0.4200, 0.5775, 0.5583], [0.6084, 0.4005, 0.8400, 0.4317, 0.3762, 0.4750, 0.5476, 0.5058], [0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6163, 0.4001, 0.8788, 0.5033, 0.4013, 0.4633, 0.5337, 0.5767], [0.6248, 0.4032, 0.7738, 0.1900, 0.4812, 0.1400, 0.5941, 0.4904], [0.6202, 0.4053, 0.8637, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133], [0.6275, 0.4024, 0.8500, 0.5383, 0.3913, 0.4883, 0.6288, 0.5100]]], device='cuda:0') loss_train_step before backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train: 0.84257042047102 step: 67 running loss: 0.012575677917477912 Train Steps: 67/90 Loss: 0.0126 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133], [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689], [0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6659, 0.4144, 0.8573, 0.4683, 0.4495, 0.4916, 0.6621, 0.5355], [0.6986, 0.4502, 0.8708, 0.4760, 0.4388, 0.5419, 0.6633, 0.5493], [0.6271, 0.3910, 0.8422, 0.2759, 0.4727, 0.2775, 0.6521, 0.5047], [0.6675, 0.4245, 0.8207, 0.4953, 0.4582, 0.4860, 0.6060, 0.5264], [0.6498, 0.4277, 0.8808, 0.4407, 0.4198, 0.4427, 0.6638, 0.5178], [0.6913, 0.4296, 0.8735, 0.4553, 0.4264, 0.4849, 0.6353, 0.5129], [0.6791, 0.4282, 0.7736, 0.3109, 0.3759, 0.3660, 0.6060, 0.5501], [0.6112, 0.3996, 0.7485, 0.2576, 0.4500, 0.2823, 0.6342, 0.5452]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6275, 0.4111, 0.8462, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6202, 0.4053, 0.8637, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133], [0.6263, 0.4065, 0.9038, 0.4317, 0.3587, 0.4550, 0.6325, 0.5250], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6102, 0.4001, 0.7738, 0.3583, 0.3462, 0.3800, 0.5524, 0.5689], [0.6229, 0.4066, 0.7613, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.8437589161330834 step: 68 running loss: 0.012408219354898286 Train Steps: 68/90 Loss: 0.0124 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767], [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417], [0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.6139, 0.4019, 0.7137, 0.2150, 0.4375, 0.1533, 0.5293, 0.5006], [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400], [ nan, nan, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6574, 0.4143, 0.8031, 0.3755, 0.4134, 0.3475, 0.6050, 0.5221], [0.6055, 0.3822, 0.7905, 0.3181, 0.4092, 0.3100, 0.6078, 0.5171], [0.6002, 0.3704, 0.7617, 0.3438, 0.4273, 0.2910, 0.5583, 0.5095], [0.5412, 0.3202, 0.6806, 0.2306, 0.4217, 0.2505, 0.5640, 0.5077], [0.6804, 0.4248, 0.9077, 0.5563, 0.4179, 0.5659, 0.5997, 0.5211], [0.4818, 0.3003, 0.6873, 0.2609, 0.4478, 0.2683, 0.5393, 0.5116], [0.7298, 0.4601, 0.8969, 0.5499, 0.4313, 0.6434, 0.6491, 0.5297], [0.6359, 0.4006, 0.8891, 0.3423, 0.4818, 0.3901, 0.6773, 0.4832]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6230, 0.4152, 0.7588, 0.2283, 0.4013, 0.2883, 0.6200, 0.5767], [0.6243, 0.4128, 0.7763, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417], [0.6143, 0.4040, 0.8238, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.6139, 0.4019, 0.7138, 0.2150, 0.4375, 0.1533, 0.5293, 0.5006], [0.6201, 0.3970, 0.8413, 0.4950, 0.4412, 0.5183, 0.6087, 0.5400], [0.0000, 0.0000, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413]]], device='cuda:0') loss_train_step before backward: tensor(0.0075, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0075, device='cuda:0', grad_fn=) loss_train: 0.8512897320324555 step: 69 running loss: 0.012337532348296456 Train Steps: 69/90 Loss: 0.0123 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683], [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700], [0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136], [0.6346, 0.4165, 0.9138, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471], [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6429, 0.3846, 0.9090, 0.4461, 0.4379, 0.4578, 0.5886, 0.5220], [0.6341, 0.3974, 0.8621, 0.5092, 0.4944, 0.4995, 0.6058, 0.5295], [0.6139, 0.4076, 0.7800, 0.3218, 0.3755, 0.4063, 0.5946, 0.5400], [0.6022, 0.3759, 0.8596, 0.4588, 0.5010, 0.5012, 0.6091, 0.5032], [0.6845, 0.4206, 0.8694, 0.4623, 0.3940, 0.5087, 0.6284, 0.5446], [0.5806, 0.3420, 0.6818, 0.2290, 0.4298, 0.2716, 0.5601, 0.5289], [0.6133, 0.3850, 0.9095, 0.4131, 0.4331, 0.4100, 0.6622, 0.5053], [0.6524, 0.3739, 0.8804, 0.4687, 0.3983, 0.4143, 0.6038, 0.5301]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683], [0.6111, 0.3995, 0.8788, 0.4567, 0.3812, 0.4833, 0.5450, 0.5700], [0.6043, 0.4022, 0.6888, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136], [0.6346, 0.4165, 0.9137, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471], [0.6262, 0.4163, 0.8850, 0.5183, 0.3762, 0.4150, 0.6025, 0.5500]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.8528305467916653 step: 70 running loss: 0.012183293525595218 Train Steps: 70/90 Loss: 0.0122 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967], [0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913], [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631], [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433], [0.6286, 0.4055, 0.9000, 0.4717, 0.3763, 0.4683, 0.7018, 0.5494], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6255, 0.3748, 0.8938, 0.4418, 0.3911, 0.5406, 0.5918, 0.5268], [0.5163, 0.3047, 0.7975, 0.2625, 0.4969, 0.1968, 0.5712, 0.5116], [0.6273, 0.3768, 0.8290, 0.3421, 0.3700, 0.3842, 0.5794, 0.5481], [0.5764, 0.3479, 0.7528, 0.3221, 0.3897, 0.3325, 0.5494, 0.5493], [0.5990, 0.3626, 0.8662, 0.4235, 0.3871, 0.5108, 0.6057, 0.5491], [0.6179, 0.3711, 0.8998, 0.5116, 0.4923, 0.4869, 0.5835, 0.5161], [0.6275, 0.3973, 0.9284, 0.4910, 0.4084, 0.5210, 0.6650, 0.5278], [0.6413, 0.3900, 0.8882, 0.5357, 0.3857, 0.4287, 0.5707, 0.5402]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967], [0.6286, 0.4078, 0.8062, 0.2267, 0.4787, 0.1533, 0.5953, 0.4913], [0.6102, 0.4001, 0.7738, 0.3583, 0.3462, 0.3800, 0.5524, 0.5689], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631], [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433], [0.6286, 0.4055, 0.9000, 0.4717, 0.3762, 0.4683, 0.7018, 0.5494], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5412, 0.5683]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.8542323596775532 step: 71 running loss: 0.01203144168559934 Train Steps: 71/90 Loss: 0.0120 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6339, 0.4102, 0.8588, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089], [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261], [0.6118, 0.4052, 0.8463, 0.3917, 0.3538, 0.3450, 0.5053, 0.5593], [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267], [0.6157, 0.4102, 0.8513, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890], [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550], [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5770, 0.3702, 0.8246, 0.3224, 0.4389, 0.3208, 0.6044, 0.5446], [0.5811, 0.3755, 0.8506, 0.5164, 0.3648, 0.5003, 0.5496, 0.5302], [0.5792, 0.3675, 0.8441, 0.4449, 0.3259, 0.4114, 0.5041, 0.5739], [0.5408, 0.3342, 0.7698, 0.2914, 0.4380, 0.2589, 0.5881, 0.5285], [0.5530, 0.3469, 0.8460, 0.4321, 0.3400, 0.4086, 0.5180, 0.5602], [0.5483, 0.3793, 0.8731, 0.4913, 0.4134, 0.4735, 0.5696, 0.5521], [0.5468, 0.3407, 0.8052, 0.2672, 0.4373, 0.2658, 0.5853, 0.5293], [0.5349, 0.3487, 0.8739, 0.4018, 0.4091, 0.3381, 0.5687, 0.5299]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6339, 0.4102, 0.8587, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089], [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261], [0.6118, 0.4052, 0.8462, 0.3917, 0.3537, 0.3450, 0.5053, 0.5593], [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267], [0.6157, 0.4102, 0.8512, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890], [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550], [0.6276, 0.4095, 0.8238, 0.2250, 0.4663, 0.1783, 0.6171, 0.4869], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986]]], device='cuda:0') loss_train_step before backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train: 0.856674063950777 step: 72 running loss: 0.011898250888205238 Train Steps: 72/90 Loss: 0.0119 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6198, 0.4115, 0.7762, 0.2717, 0.3713, 0.3200, 0.5837, 0.5683], [0.6202, 0.4079, 0.8025, 0.2500, 0.3763, 0.3217, 0.6125, 0.5533], [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667], [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197], [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050], [0.6200, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400], [0.6161, 0.4055, 0.8675, 0.3867, 0.3713, 0.4033, 0.5195, 0.5162]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5721, 0.3648, 0.8202, 0.3269, 0.3815, 0.3023, 0.5253, 0.5557], [0.6015, 0.3963, 0.8259, 0.3027, 0.3946, 0.2854, 0.5421, 0.5332], [0.6075, 0.4009, 0.8630, 0.3853, 0.3760, 0.3307, 0.5422, 0.5464], [0.5815, 0.3768, 0.8523, 0.5206, 0.4387, 0.4850, 0.5653, 0.5391], [0.5945, 0.3826, 0.7995, 0.3192, 0.3579, 0.3361, 0.5497, 0.5543], [0.5935, 0.3802, 0.8935, 0.5007, 0.3762, 0.4188, 0.5464, 0.5353], [0.5296, 0.3543, 0.8798, 0.5008, 0.4285, 0.4748, 0.5198, 0.5715], [0.5407, 0.3619, 0.8395, 0.4189, 0.3694, 0.3604, 0.4980, 0.5526]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6198, 0.4115, 0.7763, 0.2717, 0.3713, 0.3200, 0.5838, 0.5683], [0.6202, 0.4079, 0.8025, 0.2500, 0.3762, 0.3217, 0.6125, 0.5533], [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5987, 0.5667], [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197], [0.6135, 0.3994, 0.7912, 0.3050, 0.3625, 0.3050, 0.5838, 0.5050], [0.6201, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6201, 0.3970, 0.8413, 0.4950, 0.4412, 0.5183, 0.6087, 0.5400], [0.6161, 0.4055, 0.8675, 0.3867, 0.3713, 0.4033, 0.5195, 0.5162]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.8580122668063268 step: 73 running loss: 0.01175359269597708 Train Steps: 73/90 Loss: 0.0118 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6277, 0.4083, 0.8350, 0.2717, 0.4562, 0.1800, 0.5918, 0.4878], [0.6199, 0.4065, 0.7598, 0.2385, 0.4317, 0.1981, 0.5933, 0.5221], [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183], [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700], [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600], [0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5556, 0.3674, 0.8377, 0.2737, 0.4335, 0.2239, 0.5279, 0.5071], [0.5495, 0.3718, 0.7725, 0.2917, 0.4186, 0.2792, 0.5280, 0.5383], [0.5975, 0.4284, 0.8890, 0.5499, 0.3918, 0.4092, 0.5249, 0.5415], [0.5203, 0.3569, 0.8910, 0.2834, 0.4844, 0.2743, 0.6489, 0.5207], [0.5784, 0.4225, 0.8261, 0.3489, 0.3764, 0.2834, 0.5185, 0.5526], [0.6175, 0.4279, 0.9348, 0.5584, 0.3562, 0.5445, 0.5273, 0.5581], [0.6138, 0.4256, 0.7771, 0.3004, 0.3516, 0.3386, 0.5488, 0.5665], [0.6027, 0.4076, 0.7594, 0.3404, 0.3540, 0.3074, 0.4966, 0.5514]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6277, 0.4083, 0.8350, 0.2717, 0.4563, 0.1800, 0.5918, 0.4878], [0.6199, 0.4065, 0.7598, 0.2385, 0.4317, 0.1981, 0.5933, 0.5221], [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.6204, 0.4049, 0.7975, 0.2700, 0.3938, 0.2567, 0.5700, 0.5183], [0.6111, 0.3995, 0.8788, 0.4567, 0.3812, 0.4833, 0.5450, 0.5700], [0.6132, 0.4066, 0.7259, 0.2402, 0.3587, 0.3300, 0.6000, 0.5600], [0.6123, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485]]], device='cuda:0') loss_train_step before backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train: 0.8606450074585155 step: 74 running loss: 0.011630337938628587 Train Steps: 74/90 Loss: 0.0116 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6357, 0.4097, 0.9038, 0.3883, 0.4213, 0.2950, 0.6686, 0.5390], [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773], [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114], [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833], [0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071], [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082], [0.6176, 0.3911, 0.8738, 0.4217, 0.3488, 0.4033, 0.6025, 0.4817], [0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6290, 0.4395, 0.8563, 0.3367, 0.4081, 0.2847, 0.6071, 0.5613], [0.6373, 0.4309, 0.8770, 0.4503, 0.4436, 0.4187, 0.5412, 0.5536], [0.6073, 0.4401, 0.8787, 0.4466, 0.4488, 0.4279, 0.5620, 0.5581], [0.6246, 0.4447, 0.8888, 0.4558, 0.3932, 0.4469, 0.5294, 0.5504], [0.6540, 0.4370, 0.7701, 0.2456, 0.3673, 0.2334, 0.5448, 0.5332], [0.6194, 0.4169, 0.8621, 0.4388, 0.4443, 0.5149, 0.5497, 0.5537], [0.6443, 0.4170, 0.8732, 0.3948, 0.3623, 0.3728, 0.5195, 0.5393], [0.6201, 0.4285, 0.8864, 0.3037, 0.4047, 0.3044, 0.6332, 0.5474]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6357, 0.4097, 0.9038, 0.3883, 0.4212, 0.2950, 0.6686, 0.5390], [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773], [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114], [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6037, 0.4833], [0.6163, 0.4114, 0.7650, 0.2017, 0.3762, 0.2867, 0.5631, 0.5071], [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082], [0.6176, 0.3911, 0.8737, 0.4217, 0.3487, 0.4033, 0.6025, 0.4817], [0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494]]], device='cuda:0') loss_train_step before backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train: 0.8626611140789464 step: 75 running loss: 0.011502148187719285 Train Steps: 75/90 Loss: 0.0115 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950], [0.6222, 0.4169, 0.8638, 0.5650, 0.4313, 0.4783, 0.5637, 0.5633], [0.6268, 0.4061, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300], [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343], [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166], [0.6296, 0.4045, 0.9138, 0.4100, 0.4232, 0.4242, 0.7422, 0.5297], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6245, 0.4360, 0.8635, 0.4273, 0.4271, 0.4352, 0.5655, 0.5258], [0.6267, 0.4584, 0.8634, 0.4847, 0.3942, 0.4329, 0.5814, 0.5642], [0.6581, 0.4601, 0.8488, 0.2100, 0.4235, 0.1943, 0.6559, 0.5183], [0.6071, 0.4238, 0.8191, 0.2969, 0.3818, 0.2718, 0.6269, 0.5375], [0.6349, 0.4543, 0.8456, 0.4415, 0.4136, 0.4470, 0.6092, 0.5495], [0.6615, 0.4462, 0.8442, 0.3378, 0.3510, 0.3641, 0.6180, 0.5369], [0.6187, 0.4142, 0.8872, 0.4145, 0.3976, 0.4471, 0.5936, 0.5266], [0.6192, 0.4561, 0.8630, 0.4248, 0.4287, 0.4223, 0.5841, 0.5305]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950], [0.6222, 0.4169, 0.8637, 0.5650, 0.4313, 0.4783, 0.5638, 0.5633], [0.6268, 0.4060, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300], [0.6275, 0.4003, 0.9100, 0.3783, 0.4387, 0.3133, 0.7058, 0.5343], [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166], [0.6296, 0.4045, 0.9137, 0.4100, 0.4232, 0.4242, 0.7422, 0.5297], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5412, 0.5717]]], device='cuda:0') loss_train_step before backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train: 0.86508232972119 step: 76 running loss: 0.011382662233173553 Train Steps: 76/90 Loss: 0.0114 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892], [0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5771, 0.3978, 0.8029, 0.2722, 0.4141, 0.3073, 0.6815, 0.5198], [0.6085, 0.4261, 0.8393, 0.4270, 0.3635, 0.4012, 0.5760, 0.5183], [0.5968, 0.4245, 0.8060, 0.4617, 0.4207, 0.4515, 0.6187, 0.5316], [0.6251, 0.4489, 0.8296, 0.3746, 0.3768, 0.3473, 0.5871, 0.5281], [0.6556, 0.4485, 0.8583, 0.2192, 0.4339, 0.2290, 0.6672, 0.5177], [0.6243, 0.4550, 0.7873, 0.1887, 0.4288, 0.1917, 0.6569, 0.5013], [0.5891, 0.4459, 0.8296, 0.4930, 0.4048, 0.5645, 0.6507, 0.5437], [0.6506, 0.4640, 0.8796, 0.4551, 0.3506, 0.4018, 0.6355, 0.4990]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892], [0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6250, 0.4131, 0.8687, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.8669920227257535 step: 77 running loss: 0.01125963665877602 Train Steps: 77/90 Loss: 0.0113 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6205, 0.4062, 0.8337, 0.2683, 0.3675, 0.4283, 0.6338, 0.5250], [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6125, 0.4035, 0.7825, 0.3100, 0.3463, 0.4900, 0.5832, 0.5637], [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550], [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6650, 0.4347, 0.8744, 0.4900, 0.3861, 0.4386, 0.6273, 0.5055], [0.6044, 0.4057, 0.8026, 0.2726, 0.3897, 0.3852, 0.6577, 0.5237], [0.7325, 0.5001, 0.8779, 0.5122, 0.4095, 0.3320, 0.6974, 0.5019], [0.6298, 0.4239, 0.7692, 0.2225, 0.4536, 0.1487, 0.6777, 0.4890], [0.6142, 0.3649, 0.8179, 0.2979, 0.4253, 0.2464, 0.6475, 0.5156], [0.5920, 0.3943, 0.7995, 0.3386, 0.3788, 0.4432, 0.6688, 0.5305], [0.6075, 0.4290, 0.8975, 0.4775, 0.4835, 0.4502, 0.6819, 0.5282], [0.6853, 0.4624, 0.8745, 0.5447, 0.3783, 0.4823, 0.6775, 0.5141]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6108, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6205, 0.4062, 0.8338, 0.2683, 0.3675, 0.4283, 0.6338, 0.5250], [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6125, 0.4035, 0.7825, 0.3100, 0.3462, 0.4900, 0.5832, 0.5637], [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550], [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.8684362914646044 step: 78 running loss: 0.01113379860852057 Train Steps: 78/90 Loss: 0.0111 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6216, 0.4099, 0.7225, 0.2033, 0.4188, 0.2217, 0.5975, 0.5283], [0.6252, 0.4158, 0.8988, 0.4083, 0.3788, 0.4783, 0.6225, 0.5633], [0.6115, 0.3998, 0.7063, 0.2383, 0.4037, 0.1950, 0.5320, 0.4993], [0.6275, 0.4157, 0.8337, 0.5800, 0.3763, 0.4200, 0.5547, 0.6125], [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533], [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166], [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800], [0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6333, 0.4204, 0.7791, 0.2819, 0.4128, 0.2189, 0.6612, 0.5254], [0.7018, 0.4661, 0.9411, 0.4837, 0.3963, 0.5429, 0.7287, 0.5176], [0.6514, 0.4204, 0.7638, 0.2597, 0.4110, 0.1882, 0.6433, 0.4838], [0.7044, 0.4533, 0.8991, 0.6170, 0.3925, 0.5476, 0.6761, 0.5157], [0.5008, 0.2885, 0.7548, 0.2519, 0.4445, 0.1971, 0.6265, 0.5172], [0.6706, 0.4102, 0.9236, 0.5346, 0.4214, 0.6095, 0.6859, 0.4901], [0.6292, 0.3884, 0.8051, 0.2587, 0.4408, 0.2057, 0.6468, 0.5499], [0.6336, 0.3999, 0.9124, 0.6172, 0.4317, 0.5350, 0.6373, 0.5216]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6216, 0.4099, 0.7225, 0.2033, 0.4187, 0.2217, 0.5975, 0.5283], [0.6252, 0.4158, 0.8988, 0.4083, 0.3787, 0.4783, 0.6225, 0.5633], [0.6115, 0.3998, 0.7063, 0.2383, 0.4038, 0.1950, 0.5320, 0.4993], [0.6275, 0.4157, 0.8338, 0.5800, 0.3762, 0.4200, 0.5547, 0.6125], [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533], [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166], [0.6218, 0.4185, 0.7337, 0.2650, 0.4625, 0.1950, 0.5688, 0.5800], [0.6222, 0.3937, 0.8350, 0.5617, 0.4137, 0.4600, 0.5800, 0.5233]]], device='cuda:0') loss_train_step before backward: tensor(0.0033, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0033, device='cuda:0', grad_fn=) loss_train: 0.8717839835444465 step: 79 running loss: 0.011035240298030968 Train Steps: 79/90 Loss: 0.0110 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064], [0.6353, 0.4128, 0.8488, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114], [0.6124, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402], [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297], [0.6040, 0.4002, 0.7338, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6030, 0.3594, 0.8412, 0.5356, 0.4528, 0.4940, 0.6149, 0.5121], [0.5933, 0.3381, 0.8520, 0.2899, 0.5123, 0.2131, 0.6733, 0.5340], [0.6143, 0.3798, 0.8628, 0.5381, 0.4447, 0.5084, 0.6310, 0.5308], [0.6493, 0.3699, 0.8399, 0.5038, 0.3668, 0.4863, 0.6120, 0.5405], [0.6789, 0.4246, 0.8697, 0.5666, 0.3557, 0.4841, 0.6516, 0.5291], [0.6382, 0.3728, 0.7099, 0.2474, 0.4178, 0.1808, 0.5965, 0.5131], [0.6297, 0.3438, 0.8159, 0.3174, 0.4052, 0.2538, 0.6249, 0.5244], [0.7083, 0.4381, 0.9010, 0.4947, 0.4144, 0.3560, 0.7052, 0.5052]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6102, 0.4005, 0.8687, 0.5100, 0.4812, 0.5400, 0.5404, 0.5064], [0.6353, 0.4128, 0.8487, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114], [0.6123, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402], [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297], [0.6040, 0.4002, 0.7337, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6307, 0.4029, 0.8988, 0.4817, 0.3938, 0.3500, 0.7311, 0.5378]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.8729040948674083 step: 80 running loss: 0.010911301185842603 Train Steps: 80/90 Loss: 0.0109 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567], [0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667], [0.6200, 0.4049, 0.8638, 0.5617, 0.4125, 0.5100, 0.6013, 0.5317], [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183], [0.6246, 0.4028, 0.8738, 0.4867, 0.4088, 0.5667, 0.6362, 0.5200], [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350], [0.6277, 0.4103, 0.8087, 0.5717, 0.4188, 0.4750, 0.5663, 0.6083], [0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5762, 0.3329, 0.7517, 0.2781, 0.4145, 0.1953, 0.5292, 0.5349], [0.6681, 0.4076, 0.7123, 0.3287, 0.3506, 0.2726, 0.5663, 0.5310], [0.6413, 0.3766, 0.8545, 0.6113, 0.4202, 0.5100, 0.5876, 0.5184], [0.5882, 0.3514, 0.9092, 0.5452, 0.4597, 0.5661, 0.6020, 0.5366], [0.6590, 0.3704, 0.8900, 0.5376, 0.4571, 0.5833, 0.6199, 0.5511], [0.6882, 0.4035, 0.7938, 0.2930, 0.4340, 0.1886, 0.6237, 0.5297], [0.6247, 0.3833, 0.8365, 0.5761, 0.4203, 0.4921, 0.5985, 0.5927], [0.6403, 0.3778, 0.8570, 0.3007, 0.5033, 0.2290, 0.6974, 0.5138]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.7515, 0.2708, 0.3988, 0.2267, 0.5163, 0.5567], [0.6134, 0.4090, 0.6926, 0.2819, 0.3537, 0.3233, 0.5562, 0.5667], [0.6199, 0.4049, 0.8637, 0.5617, 0.4125, 0.5100, 0.6012, 0.5317], [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183], [0.6246, 0.4028, 0.8737, 0.4867, 0.4087, 0.5667, 0.6363, 0.5200], [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350], [0.6277, 0.4103, 0.8087, 0.5717, 0.4187, 0.4750, 0.5663, 0.6083], [0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413]]], device='cuda:0') loss_train_step before backward: tensor(0.0079, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0079, device='cuda:0', grad_fn=) loss_train: 0.880808724090457 step: 81 running loss: 0.01087418177889453 Train Steps: 81/90 Loss: 0.0109 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.6090, 0.4010, 0.7838, 0.3483, 0.3538, 0.3783, 0.5462, 0.5077], [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.6133, 0.4066, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358], [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6216, 0.3691, 0.8633, 0.5882, 0.4457, 0.5147, 0.6355, 0.5467], [0.6850, 0.3830, 0.8209, 0.3345, 0.4442, 0.2901, 0.6099, 0.5626], [0.6290, 0.3722, 0.8811, 0.4910, 0.4291, 0.4807, 0.6123, 0.5476], [0.6498, 0.3795, 0.8206, 0.3536, 0.3999, 0.3637, 0.5786, 0.5534], [0.6327, 0.3948, 0.7703, 0.3668, 0.4033, 0.3676, 0.5708, 0.5905], [0.6211, 0.3289, 0.8803, 0.5134, 0.4360, 0.4698, 0.5691, 0.5627], [0.5953, 0.3649, 0.7168, 0.2784, 0.4429, 0.1925, 0.5494, 0.5407], [0.6489, 0.3648, 0.8822, 0.5375, 0.4523, 0.4195, 0.5986, 0.5419]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.6090, 0.4010, 0.7837, 0.3483, 0.3537, 0.3783, 0.5462, 0.5077], [0.6170, 0.4102, 0.7468, 0.3695, 0.3462, 0.3767, 0.5238, 0.5823], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.6133, 0.4065, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358], [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.882431311882101 step: 82 running loss: 0.010761357461976841 Train Steps: 82/90 Loss: 0.0108 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773], [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791], [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433], [0.6339, 0.4118, 0.7988, 0.5800, 0.3912, 0.4583, 0.7343, 0.5760], [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [ nan, nan, 0.8363, 0.3317, 0.3563, 0.3367, 0.5329, 0.5142]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6214, 0.3727, 0.7104, 0.2819, 0.4215, 0.2213, 0.5095, 0.5704], [0.6013, 0.3810, 0.8906, 0.5344, 0.4157, 0.5617, 0.6100, 0.5443], [0.6315, 0.3857, 0.8651, 0.4446, 0.3643, 0.4125, 0.5376, 0.5502], [0.6312, 0.3986, 0.8583, 0.5682, 0.5022, 0.4889, 0.5536, 0.5491], [0.5955, 0.4035, 0.8144, 0.5087, 0.3951, 0.4339, 0.5913, 0.5845], [0.5218, 0.3318, 0.7971, 0.2683, 0.4273, 0.2634, 0.5459, 0.5848], [0.5299, 0.3395, 0.7230, 0.2444, 0.4156, 0.1974, 0.5371, 0.5694], [0.5515, 0.3686, 0.8005, 0.3105, 0.3675, 0.3014, 0.5491, 0.5271]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6147, 0.4026, 0.6600, 0.2467, 0.4087, 0.2150, 0.5489, 0.5773], [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791], [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433], [0.6339, 0.4118, 0.7987, 0.5800, 0.3913, 0.4583, 0.7343, 0.5760], [0.0000, 0.0000, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.0000, 0.0000, 0.8363, 0.3317, 0.3562, 0.3367, 0.5329, 0.5142]]], device='cuda:0') loss_train_step before backward: tensor(0.0140, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0140, device='cuda:0', grad_fn=) loss_train: 0.8964379815151915 step: 83 running loss: 0.010800457608616766 Train Steps: 83/90 Loss: 0.0108 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852], [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667], [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5963, 0.6217], [0.6293, 0.3982, 0.8700, 0.5300, 0.3763, 0.4717, 0.7050, 0.5297], [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204], [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050], [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5326, 0.3346, 0.8373, 0.5127, 0.4292, 0.4920, 0.5383, 0.5489], [0.5526, 0.3637, 0.8333, 0.2463, 0.4774, 0.2220, 0.6235, 0.5523], [0.5676, 0.3812, 0.8502, 0.3600, 0.3812, 0.3840, 0.5246, 0.5630], [0.5029, 0.3379, 0.7094, 0.2572, 0.4360, 0.2681, 0.5220, 0.5629], [0.5869, 0.3627, 0.8255, 0.5369, 0.3665, 0.4870, 0.5639, 0.5548], [0.5658, 0.3800, 0.8173, 0.5128, 0.4016, 0.4779, 0.5786, 0.5503], [0.5854, 0.3794, 0.7980, 0.2851, 0.3610, 0.3052, 0.5562, 0.5387], [0.5243, 0.3401, 0.8351, 0.3306, 0.3540, 0.4555, 0.5691, 0.5675]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6140, 0.4070, 0.8700, 0.5000, 0.4613, 0.4900, 0.5260, 0.5852], [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667], [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5962, 0.6217], [0.6293, 0.3982, 0.8700, 0.5300, 0.3762, 0.4717, 0.7050, 0.5297], [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204], [0.6135, 0.3994, 0.7912, 0.3050, 0.3625, 0.3050, 0.5838, 0.5050], [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400]]], device='cuda:0') loss_train_step before backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train: 0.8985337674384937 step: 84 running loss: 0.010696830564743973 Train Steps: 84/90 Loss: 0.0107 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058], [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5963, 0.6217], [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882], [0.6214, 0.4112, 0.7838, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6293, 0.3982, 0.8700, 0.5300, 0.3763, 0.4717, 0.7050, 0.5297], [0.6277, 0.4103, 0.8087, 0.5717, 0.4188, 0.4750, 0.5663, 0.6083]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5225, 0.3469, 0.8365, 0.3800, 0.3507, 0.4460, 0.5555, 0.5378], [0.4763, 0.3448, 0.7153, 0.2141, 0.4209, 0.2689, 0.5513, 0.5684], [0.5475, 0.3803, 0.8438, 0.5036, 0.3565, 0.4401, 0.5760, 0.5350], [0.4865, 0.3568, 0.8518, 0.3545, 0.3320, 0.3578, 0.5361, 0.5598], [0.5865, 0.3979, 0.8120, 0.2146, 0.3638, 0.2559, 0.5730, 0.5537], [0.4913, 0.3482, 0.8402, 0.4484, 0.4285, 0.4935, 0.5410, 0.5497], [0.5621, 0.3714, 0.8258, 0.4831, 0.3521, 0.4723, 0.6025, 0.5578], [0.5258, 0.3645, 0.8049, 0.4832, 0.3943, 0.4806, 0.5652, 0.6209]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6084, 0.4005, 0.8400, 0.4317, 0.3762, 0.4750, 0.5476, 0.5058], [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5962, 0.6217], [0.6334, 0.4037, 0.8637, 0.5733, 0.4013, 0.4717, 0.6369, 0.4938], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882], [0.6214, 0.4112, 0.7837, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6293, 0.3982, 0.8700, 0.5300, 0.3762, 0.4717, 0.7050, 0.5297], [0.6277, 0.4103, 0.8087, 0.5717, 0.4187, 0.4750, 0.5663, 0.6083]]], device='cuda:0') loss_train_step before backward: tensor(0.0027, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0027, device='cuda:0', grad_fn=) loss_train: 0.901197498315014 step: 85 running loss: 0.0106023235095884 Train Steps: 85/90 Loss: 0.0106 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [ nan, nan, 0.8363, 0.3317, 0.3563, 0.3367, 0.5329, 0.5142], [0.6133, 0.4066, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5604, 0.4111, 0.7505, 0.3297, 0.3333, 0.3955, 0.5836, 0.5738], [0.4741, 0.3754, 0.8639, 0.3444, 0.4385, 0.3285, 0.6015, 0.5387], [0.4714, 0.3799, 0.8131, 0.3051, 0.3737, 0.3540, 0.5557, 0.5612], [0.3484, 0.2857, 0.7763, 0.2619, 0.3515, 0.3478, 0.5347, 0.5393], [0.4544, 0.3571, 0.6789, 0.2221, 0.3817, 0.2357, 0.5271, 0.5358], [0.5848, 0.3920, 0.8319, 0.3195, 0.3314, 0.4011, 0.6398, 0.5342], [0.5762, 0.3898, 0.8072, 0.5178, 0.3728, 0.5168, 0.5860, 0.5971], [0.5460, 0.3911, 0.8495, 0.4813, 0.3454, 0.3946, 0.5857, 0.5254]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6170, 0.4102, 0.7468, 0.3695, 0.3462, 0.3767, 0.5238, 0.5823], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.0000, 0.0000, 0.8363, 0.3317, 0.3562, 0.3367, 0.5329, 0.5142], [0.6133, 0.4065, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915]]], device='cuda:0') loss_train_step before backward: tensor(0.0056, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0056, device='cuda:0', grad_fn=) loss_train: 0.9067885667318478 step: 86 running loss: 0.010544053101533115 Train Steps: 86/90 Loss: 0.0105 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364], [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461], [0.6059, 0.4002, 0.7562, 0.2767, 0.3538, 0.3033, 0.5529, 0.5455], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810], [0.6203, 0.4021, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621], [0.6129, 0.4063, 0.8738, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874], [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4804, 0.3765, 0.7629, 0.2209, 0.4160, 0.2672, 0.5957, 0.5630], [0.5317, 0.4018, 0.8057, 0.3704, 0.3130, 0.4027, 0.5884, 0.5213], [0.5202, 0.4078, 0.7146, 0.2645, 0.3400, 0.3162, 0.5732, 0.5277], [0.5567, 0.4072, 0.7966, 0.2697, 0.3687, 0.3272, 0.5992, 0.5057], [0.5670, 0.3871, 0.8749, 0.5211, 0.3079, 0.4514, 0.5793, 0.5269], [0.3928, 0.3086, 0.8130, 0.2346, 0.4734, 0.2574, 0.5827, 0.5535], [0.5530, 0.3793, 0.8436, 0.5372, 0.3630, 0.5155, 0.5582, 0.5575], [0.5016, 0.3815, 0.8580, 0.4418, 0.3176, 0.5024, 0.6462, 0.5298]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364], [0.6058, 0.3978, 0.8288, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461], [0.6059, 0.4002, 0.7563, 0.2767, 0.3537, 0.3033, 0.5529, 0.5455], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810], [0.6203, 0.4020, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [0.0000, 0.0000, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621], [0.6130, 0.4063, 0.8737, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874], [0.6263, 0.4065, 0.9038, 0.4317, 0.3587, 0.4550, 0.6325, 0.5250]]], device='cuda:0') loss_train_step before backward: tensor(0.0060, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0060, device='cuda:0', grad_fn=) loss_train: 0.9128216054523364 step: 87 running loss: 0.010492202361521107 Train Steps: 87/90 Loss: 0.0105 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467], [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947], [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650], [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567], [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183], [ nan, nan, 0.7725, 0.2611, 0.3675, 0.2733, 0.5413, 0.5167], [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6309, 0.4411, 0.8566, 0.5458, 0.3440, 0.4899, 0.6248, 0.5228], [0.6560, 0.4673, 0.8795, 0.3399, 0.3029, 0.3756, 0.6669, 0.4935], [0.5384, 0.3722, 0.7807, 0.1819, 0.4088, 0.2115, 0.6427, 0.5097], [0.4743, 0.3617, 0.8475, 0.3988, 0.3414, 0.3437, 0.5752, 0.5261], [0.4291, 0.3059, 0.7685, 0.2349, 0.4137, 0.2631, 0.5626, 0.5433], [0.5035, 0.3525, 0.8084, 0.2813, 0.3893, 0.3058, 0.5725, 0.5391], [0.3030, 0.2367, 0.7332, 0.2236, 0.3690, 0.2707, 0.5061, 0.5414], [0.5988, 0.4033, 0.8804, 0.5317, 0.3392, 0.4191, 0.6056, 0.5135]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467], [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947], [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650], [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5688, 0.5567], [0.6196, 0.4094, 0.7563, 0.2817, 0.3938, 0.3183, 0.6012, 0.6183], [0.0000, 0.0000, 0.7725, 0.2611, 0.3675, 0.2733, 0.5412, 0.5167], [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317]]], device='cuda:0') loss_train_step before backward: tensor(0.0050, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0050, device='cuda:0', grad_fn=) loss_train: 0.9177949238801375 step: 88 running loss: 0.010429487771365199 Train Steps: 88/90 Loss: 0.0104 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967], [0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165], [0.6216, 0.4099, 0.7225, 0.2033, 0.4188, 0.2217, 0.5975, 0.5283], [0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609], [0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467], [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350], [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250], [0.6203, 0.4078, 0.8800, 0.5083, 0.3900, 0.5000, 0.6100, 0.5583]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5929, 0.3731, 0.8705, 0.4031, 0.3528, 0.4151, 0.6100, 0.5097], [0.4927, 0.3507, 0.8621, 0.2883, 0.3869, 0.2458, 0.5743, 0.5135], [0.5013, 0.3601, 0.7256, 0.2112, 0.4145, 0.1674, 0.5532, 0.5421], [0.5520, 0.3886, 0.8470, 0.3874, 0.4040, 0.4425, 0.6123, 0.5298], [0.6129, 0.4044, 0.8738, 0.4380, 0.4148, 0.4499, 0.6192, 0.5410], [0.5775, 0.3733, 0.8324, 0.4880, 0.3656, 0.3719, 0.5710, 0.5103], [0.5449, 0.3729, 0.8941, 0.4075, 0.3529, 0.3875, 0.6507, 0.5150], [0.5946, 0.3648, 0.8880, 0.5087, 0.3705, 0.4399, 0.5809, 0.5327]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967], [0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165], [0.6216, 0.4099, 0.7225, 0.2033, 0.4187, 0.2217, 0.5975, 0.5283], [0.6064, 0.4019, 0.8650, 0.4517, 0.4038, 0.5367, 0.5703, 0.5609], [0.6199, 0.4059, 0.8700, 0.4900, 0.4162, 0.5000, 0.6162, 0.5467], [0.6143, 0.3982, 0.8650, 0.4883, 0.3913, 0.4317, 0.5315, 0.5350], [0.6263, 0.4065, 0.9038, 0.4317, 0.3587, 0.4550, 0.6325, 0.5250], [0.6203, 0.4078, 0.8800, 0.5083, 0.3900, 0.5000, 0.6100, 0.5583]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.9196687457151711 step: 89 running loss: 0.010333356693428889 Train Steps: 89/90 Loss: 0.0103 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123], [0.6128, 0.4115, 0.7934, 0.3778, 0.3450, 0.4033, 0.5337, 0.5456], [0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700], [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583], [ nan, nan, 0.8938, 0.2850, 0.4662, 0.3117, 0.7406, 0.5528], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5435, 0.3552, 0.8985, 0.4497, 0.3395, 0.4142, 0.5730, 0.5355], [0.5902, 0.3901, 0.8323, 0.3892, 0.3376, 0.3772, 0.5743, 0.5076], [0.6234, 0.4027, 0.7875, 0.3771, 0.3801, 0.3209, 0.5971, 0.5437], [0.5622, 0.3665, 0.7203, 0.2761, 0.3826, 0.2560, 0.5562, 0.5354], [0.5153, 0.3298, 0.9352, 0.3171, 0.4653, 0.2956, 0.6762, 0.5267], [0.5273, 0.3673, 0.7741, 0.2380, 0.4547, 0.1751, 0.5787, 0.5063], [0.6221, 0.4118, 0.8282, 0.2833, 0.4419, 0.2312, 0.6158, 0.5239], [0.6492, 0.3897, 0.8705, 0.5609, 0.4380, 0.5152, 0.5819, 0.5223]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6213, 0.4131, 0.8438, 0.3550, 0.3512, 0.4400, 0.5716, 0.5123], [0.6128, 0.4115, 0.7934, 0.3778, 0.3450, 0.4033, 0.5337, 0.5456], [0.6212, 0.4171, 0.7875, 0.3633, 0.3812, 0.2933, 0.5675, 0.5700], [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6087, 0.5583], [0.0000, 0.0000, 0.8938, 0.2850, 0.4663, 0.3117, 0.7406, 0.5528], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6277, 0.4056, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033]]], device='cuda:0') loss_train_step before backward: tensor(0.0071, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0071, device='cuda:0', grad_fn=) loss_train: 0.9267268031835556 step: 90 running loss: 0.010296964479817285 Valid Steps: 10/10 Loss: nan 9.9663 -------------------------------------------------- Epoch: 1 Train Loss: 0.0103 Valid Loss: nan -------------------------------------------------- size of train loader is: 90 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804], [0.6090, 0.4010, 0.7838, 0.3483, 0.3538, 0.3783, 0.5462, 0.5077], [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391], [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6151, 0.4058, 0.7068, 0.2680, 0.3400, 0.4083, 0.5775, 0.5733], [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167], [0.6350, 0.4043, 0.8738, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6872, 0.4337, 0.8972, 0.4201, 0.3770, 0.3324, 0.6289, 0.5134], [0.6172, 0.3803, 0.8485, 0.3648, 0.3583, 0.3490, 0.5906, 0.5130], [0.6515, 0.3853, 0.8185, 0.2864, 0.4441, 0.2439, 0.6147, 0.5116], [0.4792, 0.2720, 0.8376, 0.3685, 0.4626, 0.3081, 0.5410, 0.5550], [0.5997, 0.3969, 0.7568, 0.2729, 0.3546, 0.3386, 0.5979, 0.5413], [0.5900, 0.3652, 0.8555, 0.3659, 0.4145, 0.3966, 0.5892, 0.5485], [0.6974, 0.4176, 0.8878, 0.5917, 0.4094, 0.4353, 0.6341, 0.5390], [0.4887, 0.3089, 0.8156, 0.2632, 0.4195, 0.2369, 0.5856, 0.5529]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804], [0.6090, 0.4010, 0.7837, 0.3483, 0.3537, 0.3783, 0.5462, 0.5077], [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391], [0.6299, 0.4303, 0.7962, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6151, 0.4058, 0.7068, 0.2680, 0.3400, 0.4083, 0.5775, 0.5733], [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6037, 0.6167], [0.6350, 0.4043, 0.8737, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389]]], device='cuda:0') loss_train_step before backward: tensor(0.0023, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0023, device='cuda:0', grad_fn=) loss_train: 0.002330109942704439 step: 1 running loss: 0.002330109942704439 Train Steps: 1/90 Loss: 0.0023 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656], [0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153], [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517], [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6245, 0.4115, 0.8700, 0.4883, 0.4625, 0.5517, 0.6100, 0.5217], [0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183], [0.6148, 0.3996, 0.8488, 0.3867, 0.3488, 0.4067, 0.5863, 0.5000]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6145, 0.3738, 0.8607, 0.4208, 0.3780, 0.3628, 0.5549, 0.5305], [0.6738, 0.3928, 0.8229, 0.4044, 0.3819, 0.3683, 0.5741, 0.5606], [0.6270, 0.3677, 0.8551, 0.4638, 0.4560, 0.4233, 0.5913, 0.5175], [0.5668, 0.3573, 0.7547, 0.2084, 0.4347, 0.1876, 0.6145, 0.5623], [0.7260, 0.4346, 0.8786, 0.5015, 0.4410, 0.4891, 0.6635, 0.5420], [0.6284, 0.3754, 0.8509, 0.4435, 0.4574, 0.4805, 0.6267, 0.5466], [0.5074, 0.3131, 0.7886, 0.2444, 0.4170, 0.2448, 0.5512, 0.5679], [0.6932, 0.4020, 0.8751, 0.3573, 0.3753, 0.3645, 0.6225, 0.5110]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6108, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6091, 0.3997, 0.8314, 0.4334, 0.3787, 0.4550, 0.5213, 0.5656], [0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153], [0.6200, 0.4071, 0.7337, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517], [0.6339, 0.4123, 0.8637, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6245, 0.4115, 0.8700, 0.4883, 0.4625, 0.5517, 0.6100, 0.5217], [0.6204, 0.4049, 0.7975, 0.2700, 0.3938, 0.2567, 0.5700, 0.5183], [0.6148, 0.3996, 0.8487, 0.3867, 0.3487, 0.4067, 0.5863, 0.5000]]], device='cuda:0') loss_train_step before backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train: 0.004169229883700609 step: 2 running loss: 0.0020846149418503046 Train Steps: 2/90 Loss: 0.0021 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507], [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683], [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166], [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881], [0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609], [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6210, 0.3733, 0.7070, 0.1875, 0.4298, 0.2392, 0.6003, 0.5556], [0.5637, 0.3689, 0.8830, 0.3828, 0.3889, 0.4043, 0.5980, 0.5772], [0.6553, 0.3891, 0.8563, 0.4888, 0.3935, 0.3760, 0.5745, 0.5657], [0.7391, 0.4401, 0.8648, 0.4355, 0.4207, 0.4890, 0.6368, 0.5273], [0.7080, 0.4109, 0.8534, 0.5227, 0.4033, 0.4343, 0.5789, 0.5379], [0.6257, 0.4029, 0.8454, 0.4019, 0.4475, 0.5099, 0.6265, 0.5559], [0.5795, 0.3841, 0.8735, 0.4313, 0.3932, 0.4179, 0.6239, 0.5749], [0.7210, 0.4312, 0.7837, 0.2215, 0.4313, 0.2648, 0.6441, 0.5431]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6072, 0.4029, 0.7038, 0.2150, 0.3913, 0.2267, 0.5516, 0.5507], [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5412, 0.5683], [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166], [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881], [0.6064, 0.4019, 0.8650, 0.4517, 0.4038, 0.5367, 0.5703, 0.5609], [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6204, 0.4110, 0.7912, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.005690339137800038 step: 3 running loss: 0.0018967797126000125 Train Steps: 3/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183], [0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334], [0.6095, 0.3970, 0.8688, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940], [0.6339, 0.4118, 0.7988, 0.5800, 0.3912, 0.4583, 0.7343, 0.5760], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123], [0.6346, 0.4092, 0.7712, 0.5917, 0.4037, 0.4767, 0.7343, 0.5725], [ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6077, 0.3637, 0.8153, 0.3016, 0.4283, 0.3415, 0.6094, 0.5728], [0.7280, 0.4462, 0.8907, 0.4716, 0.3838, 0.4796, 0.6378, 0.5437], [0.6187, 0.3866, 0.8951, 0.4905, 0.4692, 0.4792, 0.5636, 0.5441], [0.6859, 0.4542, 0.8208, 0.4771, 0.3889, 0.4240, 0.6505, 0.5725], [0.6441, 0.4278, 0.7551, 0.2768, 0.4004, 0.3053, 0.5613, 0.5257], [0.6740, 0.4174, 0.8163, 0.4472, 0.3975, 0.4796, 0.6239, 0.5814], [0.4253, 0.2775, 0.7268, 0.2199, 0.4191, 0.2303, 0.5546, 0.5585], [0.7390, 0.4758, 0.8779, 0.4258, 0.3620, 0.4260, 0.6183, 0.5347]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6196, 0.4094, 0.7563, 0.2817, 0.3938, 0.3183, 0.6012, 0.6183], [0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334], [0.6095, 0.3970, 0.8687, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940], [0.6339, 0.4118, 0.7987, 0.5800, 0.3913, 0.4583, 0.7343, 0.5760], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123], [0.6346, 0.4092, 0.7713, 0.5917, 0.4038, 0.4767, 0.7343, 0.5725], [0.0000, 0.0000, 0.7268, 0.2333, 0.4125, 0.1933, 0.5113, 0.5383], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131]]], device='cuda:0') loss_train_step before backward: tensor(0.0060, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0060, device='cuda:0', grad_fn=) loss_train: 0.011675780289806426 step: 4 running loss: 0.0029189450724516064 Train Steps: 4/90 Loss: 0.0029 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426], [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5787, 0.5550], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817], [0.6339, 0.4118, 0.7988, 0.5800, 0.3912, 0.4583, 0.7343, 0.5760], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6264, 0.3890, 0.8849, 0.4577, 0.3787, 0.3752, 0.6042, 0.5390], [0.6357, 0.4101, 0.8017, 0.3282, 0.3719, 0.4082, 0.6085, 0.5483], [0.5796, 0.3818, 0.7151, 0.2136, 0.4244, 0.2336, 0.6140, 0.5432], [0.5836, 0.3846, 0.8652, 0.4589, 0.3859, 0.4856, 0.5789, 0.5566], [0.4955, 0.3177, 0.7559, 0.3055, 0.3664, 0.3517, 0.5298, 0.5826], [0.6555, 0.4345, 0.8346, 0.5506, 0.4640, 0.4997, 0.5712, 0.5504], [0.6729, 0.4611, 0.7886, 0.4845, 0.3807, 0.4276, 0.6603, 0.5793], [0.6554, 0.4426, 0.8285, 0.4267, 0.4289, 0.4932, 0.5877, 0.5431]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426], [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5788, 0.5550], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817], [0.6339, 0.4118, 0.7987, 0.5800, 0.3913, 0.4583, 0.7343, 0.5760], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.013066976680420339 step: 5 running loss: 0.002613395336084068 Train Steps: 5/90 Loss: 0.0026 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100], [ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633], [0.6034, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084], [0.6136, 0.4085, 0.6688, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6205, 0.4062, 0.8337, 0.2683, 0.3675, 0.4283, 0.6338, 0.5250], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5977, 0.4350, 0.8050, 0.4319, 0.3415, 0.4584, 0.5779, 0.5418], [0.6764, 0.4721, 0.8811, 0.6016, 0.4422, 0.5708, 0.6735, 0.5176], [0.3770, 0.2507, 0.8024, 0.3486, 0.4141, 0.3534, 0.5272, 0.5591], [0.6396, 0.4189, 0.7438, 0.3510, 0.3652, 0.4136, 0.6224, 0.5179], [0.5993, 0.3926, 0.7307, 0.3044, 0.3918, 0.3173, 0.5537, 0.5438], [0.5945, 0.3974, 0.8848, 0.5304, 0.3780, 0.4062, 0.5683, 0.5607], [0.6597, 0.4484, 0.8276, 0.3433, 0.3842, 0.4623, 0.6476, 0.5271], [0.6816, 0.4689, 0.7731, 0.3694, 0.3935, 0.3450, 0.6229, 0.5515]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6299, 0.4008, 0.8450, 0.5350, 0.4212, 0.5000, 0.6350, 0.5100], [0.0000, 0.0000, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633], [0.6033, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084], [0.6136, 0.4085, 0.6687, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6205, 0.4062, 0.8338, 0.2683, 0.3675, 0.4283, 0.6338, 0.5250], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750]]], device='cuda:0') loss_train_step before backward: tensor(0.0054, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0054, device='cuda:0', grad_fn=) loss_train: 0.018506109598092735 step: 6 running loss: 0.0030843515996821225 Train Steps: 6/90 Loss: 0.0031 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6128, 0.4115, 0.7934, 0.3778, 0.3450, 0.4033, 0.5337, 0.5456], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250], [0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183], [0.6177, 0.4086, 0.8738, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700], [0.6040, 0.4002, 0.7338, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778], [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817], [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5708, 0.3968, 0.7693, 0.4015, 0.3199, 0.4287, 0.5562, 0.5247], [0.5546, 0.3773, 0.8702, 0.5485, 0.3906, 0.5924, 0.6215, 0.5399], [0.6222, 0.4316, 0.7642, 0.3521, 0.3556, 0.3273, 0.5974, 0.5201], [0.6468, 0.4305, 0.8375, 0.4463, 0.3929, 0.5610, 0.6378, 0.5235], [0.5255, 0.3410, 0.6833, 0.2793, 0.3942, 0.2555, 0.5385, 0.5084], [0.5840, 0.3774, 0.7915, 0.2801, 0.4503, 0.2402, 0.6608, 0.5368], [0.5709, 0.4004, 0.8241, 0.6080, 0.4465, 0.5133, 0.5526, 0.5187], [0.5726, 0.3613, 0.8635, 0.5536, 0.3483, 0.4468, 0.5813, 0.5152]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6128, 0.4115, 0.7934, 0.3778, 0.3450, 0.4033, 0.5337, 0.5456], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250], [0.6201, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183], [0.6177, 0.4085, 0.8737, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700], [0.6040, 0.4002, 0.7337, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778], [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817], [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.01996891770977527 step: 7 running loss: 0.0028527025299678955 Train Steps: 7/90 Loss: 0.0029 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133], [0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283], [0.6200, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913], [0.6273, 0.4105, 0.8988, 0.4517, 0.3912, 0.2550, 0.5894, 0.4811], [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5045, 0.3187, 0.7871, 0.2918, 0.3870, 0.3102, 0.5536, 0.5498], [0.4373, 0.3065, 0.6487, 0.2241, 0.3569, 0.2030, 0.5153, 0.5458], [0.6749, 0.4379, 0.8354, 0.4962, 0.3637, 0.5110, 0.5942, 0.4977], [0.5190, 0.3392, 0.8178, 0.3587, 0.4088, 0.2443, 0.5731, 0.5093], [0.5491, 0.3799, 0.8266, 0.4311, 0.3344, 0.3934, 0.5794, 0.5571], [0.6324, 0.3970, 0.8339, 0.5412, 0.4179, 0.5375, 0.5340, 0.5230], [0.5478, 0.3553, 0.8102, 0.4492, 0.3866, 0.3094, 0.5413, 0.5078], [0.6430, 0.4296, 0.8219, 0.5368, 0.3574, 0.5627, 0.6212, 0.5529]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6205, 0.4015, 0.8350, 0.2717, 0.3988, 0.2550, 0.5788, 0.5133], [0.6055, 0.4015, 0.7425, 0.2033, 0.4112, 0.1883, 0.5217, 0.4823], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6273, 0.4110, 0.8900, 0.3817, 0.4187, 0.2167, 0.5858, 0.4835], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283], [0.6199, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913], [0.6273, 0.4105, 0.8988, 0.4517, 0.3913, 0.2550, 0.5894, 0.4811], [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795]]], device='cuda:0') loss_train_step before backward: tensor(0.0027, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0027, device='cuda:0', grad_fn=) loss_train: 0.022683476679958403 step: 8 running loss: 0.0028354345849948004 Train Steps: 8/90 Loss: 0.0028 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767], [0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207], [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6063, 0.5567], [0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995], [0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165], [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317], [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373], [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6031, 0.4043, 0.8679, 0.4927, 0.4254, 0.4905, 0.5511, 0.5095], [0.5522, 0.3762, 0.7357, 0.2780, 0.3510, 0.3617, 0.5283, 0.5241], [0.5653, 0.3769, 0.8385, 0.4342, 0.4193, 0.4412, 0.5588, 0.5099], [0.5708, 0.3708, 0.7378, 0.2137, 0.3824, 0.2108, 0.5804, 0.4831], [0.5330, 0.3495, 0.8532, 0.3130, 0.3553, 0.2720, 0.5710, 0.4873], [0.6152, 0.3664, 0.8556, 0.5128, 0.3753, 0.3823, 0.5417, 0.4911], [0.6507, 0.4053, 0.8211, 0.5239, 0.3740, 0.4510, 0.5287, 0.5169], [0.5049, 0.3249, 0.8977, 0.4204, 0.3946, 0.3056, 0.6048, 0.5078]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6148, 0.4076, 0.8666, 0.4820, 0.4137, 0.5067, 0.5250, 0.5767], [0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207], [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6062, 0.5567], [0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995], [0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165], [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317], [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373], [0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540]]], device='cuda:0') loss_train_step before backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train: 0.02442885807249695 step: 9 running loss: 0.002714317563610772 Train Steps: 9/90 Loss: 0.0027 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433], [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383], [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350], [0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6262, 0.5367], [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [0.6117, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6161, 0.3976, 0.8913, 0.5049, 0.4627, 0.4538, 0.5329, 0.5048], [0.4613, 0.3151, 0.7858, 0.2366, 0.3750, 0.2504, 0.5672, 0.5301], [0.6097, 0.4158, 0.8656, 0.5438, 0.3685, 0.4619, 0.5735, 0.4866], [0.5733, 0.3802, 0.7500, 0.2107, 0.4054, 0.1448, 0.5669, 0.5125], [0.5220, 0.3303, 0.7986, 0.2464, 0.3953, 0.2086, 0.5738, 0.5130], [0.5901, 0.4047, 0.9199, 0.4660, 0.4396, 0.5506, 0.5753, 0.4923], [0.6348, 0.4031, 0.9029, 0.5513, 0.3419, 0.3958, 0.5533, 0.5214], [0.5432, 0.3565, 0.7056, 0.2154, 0.3556, 0.2192, 0.5092, 0.5026]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433], [0.0000, 0.0000, 0.7525, 0.2291, 0.3837, 0.3017, 0.6050, 0.5667], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383], [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350], [0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6263, 0.5367], [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [0.6116, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103]]], device='cuda:0') loss_train_step before backward: tensor(0.0062, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0062, device='cuda:0', grad_fn=) loss_train: 0.030593233765102923 step: 10 running loss: 0.003059323376510292 Train Steps: 10/90 Loss: 0.0031 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133], [0.6248, 0.4032, 0.7738, 0.1900, 0.4813, 0.1400, 0.5941, 0.4904], [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350], [0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650], [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633], [0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567], [0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6524, 0.4194, 0.8302, 0.2453, 0.4642, 0.2088, 0.6127, 0.5133], [0.5333, 0.3385, 0.7581, 0.2202, 0.4625, 0.1170, 0.5673, 0.4804], [0.4795, 0.3257, 0.8051, 0.2669, 0.3902, 0.2321, 0.5726, 0.5239], [0.5878, 0.4107, 0.8545, 0.5105, 0.3563, 0.4354, 0.5768, 0.5078], [0.4419, 0.3261, 0.7948, 0.2944, 0.3893, 0.2562, 0.5276, 0.5387], [0.6046, 0.4057, 0.9291, 0.5008, 0.4122, 0.5362, 0.5740, 0.5424], [0.5551, 0.3804, 0.8104, 0.3170, 0.3766, 0.2811, 0.5634, 0.5270], [0.6001, 0.3880, 0.7420, 0.3093, 0.3546, 0.2835, 0.4877, 0.5316]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6266, 0.4065, 0.8313, 0.2467, 0.4787, 0.1733, 0.6313, 0.5133], [0.6248, 0.4032, 0.7738, 0.1900, 0.4812, 0.1400, 0.5941, 0.4904], [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350], [0.6311, 0.3998, 0.7975, 0.5767, 0.3837, 0.4850, 0.7327, 0.5343], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650], [0.6189, 0.4049, 0.8888, 0.4417, 0.4212, 0.5200, 0.5987, 0.5633], [0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567], [0.6123, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485]]], device='cuda:0') loss_train_step before backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train: 0.033024250413291156 step: 11 running loss: 0.0030022045830264688 Train Steps: 11/90 Loss: 0.0030 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6336, 0.4154, 0.8900, 0.2767, 0.4988, 0.2867, 0.7422, 0.5540], [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841], [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494], [0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100], [0.6325, 0.4165, 0.9000, 0.4617, 0.3813, 0.4900, 0.7485, 0.5447], [0.6198, 0.4130, 0.8762, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103], [0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799], [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5612, 0.3811, 0.8602, 0.2296, 0.4916, 0.2306, 0.6434, 0.5336], [0.6425, 0.4218, 0.8356, 0.3556, 0.4207, 0.3121, 0.6238, 0.5244], [0.6146, 0.4007, 0.8371, 0.4343, 0.4082, 0.3477, 0.6174, 0.5276], [0.6623, 0.4279, 0.8380, 0.4641, 0.4136, 0.4596, 0.5665, 0.5384], [0.5593, 0.3786, 0.9084, 0.4347, 0.4017, 0.4581, 0.6508, 0.5335], [0.6136, 0.4184, 0.9224, 0.3770, 0.3960, 0.4299, 0.5676, 0.5565], [0.5552, 0.3466, 0.8568, 0.3470, 0.4307, 0.2913, 0.5624, 0.5741], [0.6698, 0.4377, 0.8489, 0.4851, 0.4061, 0.4395, 0.5585, 0.5461]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6336, 0.4154, 0.8900, 0.2767, 0.4988, 0.2867, 0.7422, 0.5540], [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841], [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494], [0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100], [0.6325, 0.4165, 0.9000, 0.4617, 0.3812, 0.4900, 0.7485, 0.5447], [0.6198, 0.4130, 0.8763, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103], [0.6250, 0.4236, 0.8637, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799], [0.6222, 0.4171, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train: 0.03542143327649683 step: 12 running loss: 0.0029517861063747355 Train Steps: 12/90 Loss: 0.0030 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6102, 0.4020, 0.8638, 0.3717, 0.3625, 0.5017, 0.6038, 0.5500], [0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300], [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385], [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617], [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600], [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501], [0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5890, 0.4046, 0.8961, 0.4043, 0.4132, 0.5221, 0.6450, 0.5752], [0.6591, 0.4443, 0.7609, 0.2467, 0.4018, 0.2730, 0.6092, 0.5504], [0.6208, 0.4260, 0.8892, 0.4191, 0.3758, 0.3963, 0.5760, 0.5460], [0.6432, 0.4296, 0.8513, 0.2810, 0.4636, 0.2191, 0.6936, 0.5478], [0.6852, 0.4710, 0.8878, 0.5743, 0.4098, 0.4560, 0.5968, 0.5572], [0.3552, 0.2292, 0.8169, 0.2640, 0.4351, 0.2463, 0.5953, 0.5789], [0.6208, 0.4168, 0.8051, 0.2661, 0.4353, 0.2119, 0.6514, 0.5384], [0.5933, 0.4019, 0.8066, 0.2821, 0.4158, 0.2314, 0.5887, 0.5460]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6102, 0.4020, 0.8637, 0.3717, 0.3625, 0.5017, 0.6037, 0.5500], [0.6207, 0.4081, 0.7663, 0.2067, 0.3963, 0.3200, 0.6313, 0.5300], [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385], [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617], [0.0000, 0.0000, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600], [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501], [0.6204, 0.4049, 0.7975, 0.2700, 0.3938, 0.2567, 0.5700, 0.5183]]], device='cuda:0') loss_train_step before backward: tensor(0.0035, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0035, device='cuda:0', grad_fn=) loss_train: 0.038919162820093334 step: 13 running loss: 0.0029937817553917947 Train Steps: 13/90 Loss: 0.0030 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148], [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598], [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833], [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082], [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6063, 0.5617], [0.6216, 0.4167, 0.8588, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667], [0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6692, 0.4524, 0.8693, 0.4409, 0.3938, 0.4273, 0.6492, 0.5632], [0.5813, 0.3906, 0.8782, 0.1791, 0.4501, 0.1834, 0.7445, 0.5954], [0.6371, 0.4335, 0.8636, 0.4553, 0.4532, 0.3713, 0.6627, 0.5859], [0.6131, 0.4125, 0.8602, 0.4216, 0.3713, 0.4308, 0.6201, 0.5651], [0.5784, 0.3985, 0.8561, 0.4410, 0.4433, 0.4864, 0.6356, 0.5596], [0.6493, 0.4302, 0.8929, 0.4087, 0.3751, 0.4351, 0.6520, 0.5835], [0.6617, 0.4436, 0.8709, 0.4943, 0.4211, 0.4950, 0.6436, 0.5883], [0.5980, 0.4068, 0.8920, 0.3774, 0.4322, 0.4469, 0.6501, 0.5687]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148], [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598], [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6037, 0.4833], [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082], [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6062, 0.5617], [0.6216, 0.4167, 0.8587, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667], [0.6250, 0.4054, 0.8770, 0.4723, 0.4663, 0.5367, 0.6162, 0.5433]]], device='cuda:0') loss_train_step before backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train: 0.04132039553951472 step: 14 running loss: 0.0029514568242510514 Train Steps: 14/90 Loss: 0.0030 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6081, 0.3950, 0.8538, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374], [0.6118, 0.4052, 0.8463, 0.3917, 0.3538, 0.3450, 0.5053, 0.5593], [0.6339, 0.4102, 0.8588, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089], [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134], [ nan, nan, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6275, 0.4050, 0.9038, 0.3767, 0.3838, 0.3533, 0.7074, 0.5575], [0.6200, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5942, 0.3891, 0.8576, 0.4474, 0.4046, 0.4978, 0.5956, 0.5803], [0.6719, 0.4472, 0.8531, 0.3704, 0.3951, 0.3693, 0.6507, 0.5522], [0.6160, 0.4064, 0.8342, 0.3875, 0.3559, 0.3626, 0.5747, 0.6071], [0.7079, 0.4728, 0.8355, 0.2891, 0.4455, 0.2665, 0.7089, 0.5726], [0.6223, 0.4199, 0.8382, 0.5415, 0.4098, 0.5251, 0.6462, 0.5646], [0.3999, 0.2640, 0.8994, 0.3008, 0.4934, 0.2666, 0.7216, 0.5950], [0.6839, 0.4563, 0.8921, 0.3541, 0.4058, 0.3401, 0.7596, 0.5681], [0.6237, 0.4080, 0.8588, 0.4664, 0.3939, 0.5808, 0.6382, 0.6014]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6081, 0.3950, 0.8537, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374], [0.6118, 0.4052, 0.8462, 0.3917, 0.3537, 0.3450, 0.5053, 0.5593], [0.6339, 0.4102, 0.8587, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089], [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134], [0.0000, 0.0000, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6275, 0.4050, 0.9038, 0.3767, 0.3837, 0.3533, 0.7074, 0.5575], [0.6199, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367]]], device='cuda:0') loss_train_step before backward: tensor(0.0049, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0049, device='cuda:0', grad_fn=) loss_train: 0.046259972150437534 step: 15 running loss: 0.003083998143362502 Train Steps: 15/90 Loss: 0.0031 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117], [0.6200, 0.3993, 0.8519, 0.4923, 0.3962, 0.4717, 0.6013, 0.5433], [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6362, 0.5000], [ nan, nan, 0.6793, 0.2110, 0.4012, 0.2167, 0.5112, 0.5583], [0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626], [ nan, nan, 0.7725, 0.2611, 0.3675, 0.2733, 0.5413, 0.5167]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6457, 0.4295, 0.8061, 0.3196, 0.3867, 0.3391, 0.6472, 0.5647], [0.6592, 0.4132, 0.8562, 0.3014, 0.4840, 0.2955, 0.7247, 0.5346], [0.6431, 0.4059, 0.9184, 0.3719, 0.3832, 0.3503, 0.6498, 0.5410], [0.6882, 0.4392, 0.8831, 0.5727, 0.3650, 0.5405, 0.6501, 0.5698], [0.6855, 0.4245, 0.9158, 0.5830, 0.3851, 0.6142, 0.6524, 0.5226], [0.3411, 0.2191, 0.7176, 0.2231, 0.4243, 0.2191, 0.5480, 0.5725], [0.6789, 0.4679, 0.8847, 0.6112, 0.3784, 0.5188, 0.6684, 0.5326], [0.2940, 0.1794, 0.7829, 0.2765, 0.3914, 0.2806, 0.5706, 0.5670]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117], [0.6200, 0.3993, 0.8519, 0.4923, 0.3963, 0.4717, 0.6012, 0.5433], [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6363, 0.5000], [0.0000, 0.0000, 0.6793, 0.2110, 0.4013, 0.2167, 0.5113, 0.5583], [0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626], [0.0000, 0.0000, 0.7725, 0.2611, 0.3675, 0.2733, 0.5412, 0.5167]]], device='cuda:0') loss_train_step before backward: tensor(0.0061, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0061, device='cuda:0', grad_fn=) loss_train: 0.05236703611444682 step: 16 running loss: 0.003272939757152926 Train Steps: 16/90 Loss: 0.0033 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071], [0.6202, 0.4064, 0.7879, 0.2179, 0.4567, 0.1725, 0.5955, 0.5478], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050], [0.6127, 0.4115, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6228, 0.4119, 0.7938, 0.2233, 0.4674, 0.1773, 0.6188, 0.5433], [0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5899, 0.3427, 0.7989, 0.3637, 0.3790, 0.3644, 0.6021, 0.5073], [0.5794, 0.3500, 0.7829, 0.3558, 0.4715, 0.2606, 0.6293, 0.5235], [0.5524, 0.3206, 0.8937, 0.4374, 0.4395, 0.4080, 0.6941, 0.5116], [0.5741, 0.3633, 0.9493, 0.5738, 0.3724, 0.6298, 0.6568, 0.5525], [0.5215, 0.3038, 0.7782, 0.3703, 0.3806, 0.3965, 0.5499, 0.5504], [0.6112, 0.3637, 0.8144, 0.3356, 0.3792, 0.4078, 0.6076, 0.5224], [0.5639, 0.3496, 0.8526, 0.3172, 0.4734, 0.2581, 0.6480, 0.5238], [0.4399, 0.2583, 0.7541, 0.2975, 0.3986, 0.2429, 0.5493, 0.5150]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6163, 0.4114, 0.7650, 0.2017, 0.3762, 0.2867, 0.5631, 0.5071], [0.6202, 0.4064, 0.7879, 0.2179, 0.4567, 0.1725, 0.5955, 0.5478], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6200, 0.4112, 0.8863, 0.4100, 0.3638, 0.4917, 0.6087, 0.6050], [0.6127, 0.4114, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6228, 0.4119, 0.7937, 0.2233, 0.4674, 0.1773, 0.6187, 0.5433], [0.6055, 0.4015, 0.7425, 0.2033, 0.4112, 0.1883, 0.5217, 0.4823]]], device='cuda:0') loss_train_step before backward: tensor(0.0050, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0050, device='cuda:0', grad_fn=) loss_train: 0.057400591555051506 step: 17 running loss: 0.003376505385591265 Train Steps: 17/90 Loss: 0.0034 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950], [0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6131, 0.4037, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680], [ nan, nan, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533], [0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933], [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167], [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5694, 0.3072, 0.9180, 0.5808, 0.4260, 0.5578, 0.5957, 0.4780], [0.4657, 0.2800, 0.7147, 0.2770, 0.4011, 0.2651, 0.5142, 0.5039], [0.5343, 0.2985, 0.7305, 0.2937, 0.3831, 0.3645, 0.5357, 0.5165], [0.3305, 0.1637, 0.7213, 0.2061, 0.4155, 0.2128, 0.5293, 0.5100], [0.5894, 0.3573, 0.9104, 0.5207, 0.3549, 0.4583, 0.5722, 0.4644], [0.6676, 0.4207, 0.8576, 0.6183, 0.4383, 0.4724, 0.6427, 0.5154], [0.5665, 0.3570, 0.8540, 0.3346, 0.3749, 0.3530, 0.5816, 0.4958], [0.5989, 0.3703, 0.8312, 0.2364, 0.4190, 0.3152, 0.6319, 0.4938]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950], [0.6149, 0.4054, 0.6712, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6131, 0.4036, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680], [0.0000, 0.0000, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533], [0.6108, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933], [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167], [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320]]], device='cuda:0') loss_train_step before backward: tensor(0.0045, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0045, device='cuda:0', grad_fn=) loss_train: 0.06192554801236838 step: 18 running loss: 0.0034403082229093546 Train Steps: 18/90 Loss: 0.0034 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284], [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954], [0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019], [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364], [0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110], [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5737, 0.5350], [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100], [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5125, 0.2871, 0.7511, 0.2094, 0.4505, 0.1997, 0.5630, 0.4911], [0.4933, 0.3047, 0.8090, 0.3301, 0.3396, 0.3870, 0.5353, 0.5002], [0.5513, 0.3362, 0.8229, 0.4283, 0.3745, 0.5247, 0.5657, 0.4955], [0.5318, 0.3230, 0.7729, 0.2162, 0.4420, 0.2273, 0.6116, 0.5084], [0.5149, 0.2947, 0.9065, 0.5164, 0.4007, 0.5175, 0.5539, 0.5059], [0.6284, 0.3912, 0.8496, 0.4788, 0.4001, 0.3236, 0.5528, 0.5050], [0.5536, 0.3453, 0.8191, 0.5324, 0.3937, 0.4933, 0.5993, 0.4796], [0.5261, 0.3023, 0.7768, 0.2523, 0.4267, 0.2545, 0.5609, 0.4950]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284], [0.6076, 0.3952, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954], [0.6161, 0.4099, 0.8737, 0.4383, 0.3787, 0.5483, 0.5605, 0.5019], [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364], [0.6186, 0.4129, 0.8900, 0.4567, 0.3938, 0.5417, 0.5734, 0.5110], [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5738, 0.5350], [0.6275, 0.4024, 0.8500, 0.5383, 0.3913, 0.4883, 0.6288, 0.5100], [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391]]], device='cuda:0') loss_train_step before backward: tensor(0.0025, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0025, device='cuda:0', grad_fn=) loss_train: 0.0644436435541138 step: 19 running loss: 0.003391770713374411 Train Steps: 19/90 Loss: 0.0034 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5988, 0.5667], [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6200, 0.3993, 0.8519, 0.4923, 0.3962, 0.4717, 0.6013, 0.5433], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550], [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986], [0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5718, 0.3611, 0.7569, 0.1923, 0.4113, 0.2334, 0.5492, 0.5280], [0.5646, 0.3674, 0.8176, 0.4881, 0.4099, 0.5417, 0.6300, 0.5248], [0.5698, 0.3466, 0.8024, 0.4662, 0.3711, 0.4423, 0.5449, 0.5125], [0.6244, 0.3827, 0.8229, 0.3401, 0.3483, 0.3685, 0.5368, 0.5160], [0.5406, 0.3454, 0.6538, 0.2001, 0.3887, 0.2130, 0.4810, 0.4885], [0.5050, 0.3236, 0.8718, 0.3897, 0.4116, 0.2701, 0.5983, 0.4890], [0.4879, 0.3313, 0.8646, 0.3113, 0.4415, 0.2104, 0.6023, 0.4683], [0.5498, 0.3600, 0.8173, 0.4844, 0.4335, 0.4855, 0.4931, 0.4710]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5987, 0.5667], [0.6339, 0.4123, 0.8637, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6200, 0.3993, 0.8519, 0.4923, 0.3963, 0.4717, 0.6012, 0.5433], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5362, 0.5550], [0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986], [0.6097, 0.3988, 0.8650, 0.5250, 0.4212, 0.5200, 0.5675, 0.5050]]], device='cuda:0') loss_train_step before backward: tensor(0.0027, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0027, device='cuda:0', grad_fn=) loss_train: 0.06718559598084539 step: 20 running loss: 0.0033592797990422696 Train Steps: 20/90 Loss: 0.0034 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261], [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400], [ nan, nan, 0.7648, 0.2722, 0.3962, 0.2183, 0.5060, 0.5422], [0.6268, 0.4029, 0.8500, 0.2683, 0.3937, 0.3500, 0.6860, 0.5297], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5591, 0.3825, 0.8259, 0.4147, 0.4141, 0.3745, 0.5544, 0.5147], [0.5839, 0.4089, 0.7758, 0.2976, 0.3758, 0.2668, 0.5041, 0.5495], [0.5310, 0.3788, 0.8315, 0.4217, 0.4207, 0.3984, 0.5459, 0.5369], [0.6807, 0.4494, 0.8215, 0.5131, 0.3843, 0.3442, 0.5462, 0.5332], [0.5852, 0.4040, 0.8322, 0.2623, 0.3493, 0.4323, 0.6110, 0.5552], [0.3095, 0.2227, 0.7204, 0.1732, 0.4180, 0.1595, 0.4772, 0.5234], [0.6958, 0.4777, 0.7973, 0.2119, 0.3881, 0.2798, 0.6438, 0.5153], [0.5824, 0.4102, 0.8456, 0.4937, 0.4169, 0.3476, 0.5649, 0.5045]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261], [0.6179, 0.4118, 0.7278, 0.4237, 0.3587, 0.3400, 0.5675, 0.5917], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400], [0.0000, 0.0000, 0.7648, 0.2722, 0.3963, 0.2183, 0.5060, 0.5422], [0.6268, 0.4029, 0.8500, 0.2683, 0.3938, 0.3500, 0.6860, 0.5297], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869]]], device='cuda:0') loss_train_step before backward: tensor(0.0048, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0048, device='cuda:0', grad_fn=) loss_train: 0.07193988130893558 step: 21 running loss: 0.0034257086337588375 Train Steps: 21/90 Loss: 0.0034 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767], [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6312, 0.5250], [0.6059, 0.4002, 0.7562, 0.2767, 0.3538, 0.3033, 0.5529, 0.5455], [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915], [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617], [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6355, 0.4614, 0.8250, 0.4840, 0.3991, 0.4634, 0.5587, 0.5688], [0.5501, 0.3784, 0.8598, 0.2704, 0.3954, 0.2741, 0.6230, 0.5681], [0.6268, 0.4654, 0.7134, 0.2568, 0.3502, 0.2695, 0.5437, 0.5430], [0.6804, 0.4713, 0.8506, 0.4430, 0.3397, 0.2755, 0.5728, 0.5126], [0.5909, 0.4225, 0.8107, 0.4818, 0.3905, 0.4480, 0.5322, 0.5546], [0.5062, 0.3929, 0.8450, 0.2082, 0.5034, 0.1870, 0.6914, 0.5587], [0.3170, 0.2427, 0.7268, 0.1864, 0.4102, 0.1592, 0.4986, 0.5714], [0.5695, 0.4248, 0.7992, 0.4864, 0.3944, 0.4476, 0.5511, 0.5406]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6148, 0.4076, 0.8666, 0.4820, 0.4137, 0.5067, 0.5250, 0.5767], [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6313, 0.5250], [0.6059, 0.4002, 0.7563, 0.2767, 0.3537, 0.3033, 0.5529, 0.5455], [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915], [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.0000, 0.0000, 0.7513, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617], [0.6098, 0.3991, 0.8637, 0.4717, 0.4263, 0.4967, 0.5213, 0.5650]]], device='cuda:0') loss_train_step before backward: tensor(0.0041, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0041, device='cuda:0', grad_fn=) loss_train: 0.07605478435289115 step: 22 running loss: 0.003457035652404143 Train Steps: 22/90 Loss: 0.0035 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118], [0.6206, 0.4001, 0.8900, 0.3933, 0.3588, 0.3567, 0.5837, 0.5083], [0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4578, 0.3519, 0.8790, 0.2221, 0.5106, 0.2299, 0.6964, 0.5784], [0.5498, 0.3720, 0.7646, 0.2784, 0.3612, 0.3090, 0.5244, 0.5695], [0.6279, 0.4609, 0.8070, 0.5108, 0.3879, 0.4680, 0.6271, 0.5730], [0.6880, 0.4932, 0.8739, 0.4629, 0.3677, 0.3500, 0.5938, 0.5400], [0.5222, 0.4169, 0.7394, 0.2919, 0.3977, 0.2462, 0.5193, 0.6034], [0.6170, 0.4211, 0.8745, 0.3801, 0.3498, 0.3347, 0.5682, 0.5492], [0.6112, 0.4401, 0.8528, 0.4759, 0.3710, 0.4550, 0.5889, 0.5859], [0.5333, 0.4033, 0.8534, 0.3997, 0.4135, 0.4587, 0.5547, 0.5614]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6300, 0.4133, 0.8537, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6346, 0.4086, 0.7937, 0.5500, 0.3963, 0.4867, 0.7343, 0.5702], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118], [0.6206, 0.4001, 0.8900, 0.3933, 0.3587, 0.3567, 0.5838, 0.5083], [0.6197, 0.4091, 0.8800, 0.4783, 0.3537, 0.4767, 0.5950, 0.5550], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.07791255612391979 step: 23 running loss: 0.0033875024401704254 Train Steps: 23/90 Loss: 0.0034 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6125, 0.4035, 0.7825, 0.3100, 0.3463, 0.4900, 0.5832, 0.5637], [0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954], [0.6261, 0.4066, 0.8325, 0.2150, 0.4763, 0.2667, 0.7002, 0.5633], [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250], [0.6267, 0.4080, 0.8438, 0.2633, 0.4763, 0.1800, 0.6259, 0.5240], [ nan, nan, 0.8850, 0.2817, 0.5112, 0.2183, 0.7184, 0.5436], [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378], [0.6069, 0.3975, 0.8625, 0.5083, 0.4388, 0.5483, 0.5650, 0.4967]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5630, 0.4143, 0.7481, 0.3400, 0.3264, 0.4892, 0.5747, 0.5941], [0.5112, 0.3690, 0.7888, 0.3976, 0.3908, 0.2515, 0.4968, 0.5846], [0.5933, 0.4247, 0.7916, 0.2289, 0.4429, 0.2337, 0.6436, 0.6058], [0.6395, 0.4488, 0.8615, 0.4161, 0.3459, 0.3268, 0.5817, 0.5449], [0.5625, 0.4243, 0.8189, 0.2745, 0.4324, 0.2153, 0.5916, 0.5724], [0.4407, 0.3259, 0.8781, 0.2678, 0.4856, 0.2394, 0.6811, 0.5764], [0.6495, 0.4705, 0.8612, 0.4888, 0.3962, 0.3317, 0.6322, 0.5577], [0.6276, 0.4688, 0.8113, 0.5343, 0.4078, 0.5237, 0.5561, 0.6000]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6125, 0.4035, 0.7825, 0.3100, 0.3462, 0.4900, 0.5832, 0.5637], [0.6137, 0.4038, 0.8562, 0.4050, 0.3812, 0.2550, 0.5106, 0.4954], [0.6261, 0.4066, 0.8325, 0.2150, 0.4762, 0.2667, 0.7002, 0.5633], [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250], [0.6267, 0.4080, 0.8438, 0.2633, 0.4762, 0.1800, 0.6259, 0.5240], [0.0000, 0.0000, 0.8850, 0.2817, 0.5113, 0.2183, 0.7184, 0.5436], [0.6307, 0.4029, 0.8988, 0.4817, 0.3938, 0.3500, 0.7311, 0.5378], [0.6069, 0.3975, 0.8625, 0.5083, 0.4387, 0.5483, 0.5650, 0.4967]]], device='cuda:0') loss_train_step before backward: tensor(0.0063, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0063, device='cuda:0', grad_fn=) loss_train: 0.08417737309355289 step: 24 running loss: 0.0035073905455647036 Train Steps: 24/90 Loss: 0.0035 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533], [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250], [0.6346, 0.4092, 0.7712, 0.5917, 0.4037, 0.4767, 0.7343, 0.5725], [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750], [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407], [0.6075, 0.4007, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280], [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.3853, 0.2412, 0.7946, 0.2664, 0.3667, 0.2621, 0.5691, 0.5883], [0.6728, 0.4472, 0.9294, 0.3902, 0.3840, 0.3261, 0.6674, 0.5337], [0.6466, 0.4458, 0.8176, 0.4990, 0.4050, 0.4815, 0.6699, 0.6091], [0.2990, 0.2104, 0.7380, 0.1906, 0.4629, 0.1769, 0.5771, 0.6129], [0.6577, 0.4726, 0.9106, 0.4686, 0.3657, 0.4160, 0.6843, 0.5803], [0.5791, 0.3975, 0.8685, 0.3931, 0.3932, 0.3091, 0.5628, 0.5776], [0.6461, 0.4384, 0.8341, 0.4709, 0.4037, 0.5106, 0.6007, 0.5653], [0.7838, 0.5244, 0.8995, 0.5693, 0.4290, 0.5269, 0.6741, 0.5730]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533], [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250], [0.6346, 0.4092, 0.7713, 0.5917, 0.4038, 0.4767, 0.7343, 0.5725], [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5512, 0.5750], [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6133, 0.4094, 0.8495, 0.4028, 0.3587, 0.3200, 0.5003, 0.5407], [0.6075, 0.4006, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280], [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600]]], device='cuda:0') loss_train_step before backward: tensor(0.0073, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0073, device='cuda:0', grad_fn=) loss_train: 0.09143890452105552 step: 25 running loss: 0.0036575561808422207 Train Steps: 25/90 Loss: 0.0037 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6145, 0.4007, 0.8775, 0.4533, 0.4562, 0.5533, 0.6088, 0.5533], [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.6339, 0.4102, 0.8588, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089], [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6265, 0.4088, 0.8025, 0.1850, 0.4163, 0.2500, 0.6290, 0.4947], [0.6118, 0.4052, 0.8463, 0.3917, 0.3538, 0.3450, 0.5053, 0.5593]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6105, 0.3962, 0.8903, 0.5136, 0.4423, 0.5345, 0.6518, 0.5574], [0.7116, 0.4549, 0.8984, 0.5937, 0.4033, 0.5888, 0.7316, 0.5672], [0.7130, 0.4386, 0.9326, 0.5597, 0.3984, 0.5887, 0.6999, 0.5537], [0.4522, 0.2795, 0.7579, 0.2800, 0.4087, 0.2454, 0.5992, 0.5730], [0.6696, 0.4300, 0.9044, 0.3524, 0.4478, 0.2530, 0.6978, 0.5384], [0.4153, 0.2761, 0.7562, 0.2718, 0.4272, 0.2038, 0.5829, 0.5567], [0.4782, 0.3126, 0.8689, 0.2631, 0.4484, 0.2345, 0.6533, 0.5612], [0.6062, 0.3808, 0.8682, 0.4297, 0.3464, 0.3637, 0.5453, 0.5646]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6145, 0.4007, 0.8775, 0.4533, 0.4563, 0.5533, 0.6087, 0.5533], [0.6339, 0.4123, 0.8637, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6254, 0.3993, 0.8988, 0.4767, 0.3988, 0.5517, 0.6955, 0.5285], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.6339, 0.4102, 0.8587, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089], [0.6154, 0.4112, 0.7038, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6265, 0.4088, 0.8025, 0.1850, 0.4162, 0.2500, 0.6290, 0.4947], [0.6118, 0.4052, 0.8462, 0.3917, 0.3537, 0.3450, 0.5053, 0.5593]]], device='cuda:0') loss_train_step before backward: tensor(0.0035, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0035, device='cuda:0', grad_fn=) loss_train: 0.09492857556324452 step: 26 running loss: 0.0036510990601247894 Train Steps: 26/90 Loss: 0.0037 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6339, 0.4149, 0.8800, 0.5000, 0.3900, 0.5283, 0.7541, 0.5424], [ nan, nan, 0.7335, 0.2569, 0.3788, 0.2667, 0.5066, 0.5578], [0.6252, 0.4158, 0.8988, 0.4083, 0.3788, 0.4783, 0.6225, 0.5633], [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167], [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183], [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611], [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397], [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5787, 0.5600]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7710, 0.4596, 0.9157, 0.5361, 0.3850, 0.5504, 0.7209, 0.5340], [0.1755, 0.0911, 0.7831, 0.2816, 0.4241, 0.2348, 0.5595, 0.5750], [0.6858, 0.4413, 0.8724, 0.4188, 0.3791, 0.4858, 0.6770, 0.5232], [0.6731, 0.4134, 0.9235, 0.4454, 0.3964, 0.3642, 0.6883, 0.5083], [0.6874, 0.3868, 0.9253, 0.4927, 0.4442, 0.5354, 0.6994, 0.5263], [0.6664, 0.3893, 0.8907, 0.4971, 0.4151, 0.4613, 0.6301, 0.5303], [0.4252, 0.2396, 0.7291, 0.2757, 0.4188, 0.1851, 0.5661, 0.5395], [0.6321, 0.3918, 0.8892, 0.5291, 0.4634, 0.4708, 0.6284, 0.5273]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6339, 0.4149, 0.8800, 0.5000, 0.3900, 0.5283, 0.7541, 0.5424], [0.0000, 0.0000, 0.7335, 0.2569, 0.3787, 0.2667, 0.5066, 0.5578], [0.6252, 0.4158, 0.8988, 0.4083, 0.3787, 0.4783, 0.6225, 0.5633], [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6263, 0.5167], [0.6251, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6087, 0.5183], [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611], [0.6159, 0.4085, 0.6900, 0.2283, 0.4087, 0.1950, 0.5123, 0.5397], [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5788, 0.5600]]], device='cuda:0') loss_train_step before backward: tensor(0.0031, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0031, device='cuda:0', grad_fn=) loss_train: 0.09805791999679059 step: 27 running loss: 0.003631774814695948 Train Steps: 27/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683], [0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343], [0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350], [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117], [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5901, 0.3425, 0.8881, 0.3529, 0.3891, 0.5200, 0.6666, 0.5400], [0.6469, 0.3587, 0.8641, 0.3298, 0.3718, 0.4317, 0.6578, 0.5125], [0.4706, 0.2787, 0.8005, 0.2850, 0.4421, 0.2119, 0.6169, 0.5088], [0.7143, 0.4069, 0.8990, 0.5666, 0.3859, 0.4117, 0.5918, 0.5235], [0.5770, 0.3445, 0.8539, 0.5257, 0.4015, 0.4867, 0.6826, 0.5029], [0.6092, 0.3633, 0.8144, 0.3106, 0.4484, 0.2981, 0.6340, 0.5114], [0.6334, 0.3657, 0.9173, 0.5061, 0.3971, 0.5020, 0.6485, 0.4940], [0.6372, 0.3549, 0.8855, 0.5803, 0.4888, 0.5151, 0.5938, 0.5249]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5412, 0.5683], [0.6311, 0.3998, 0.7975, 0.5767, 0.3837, 0.4850, 0.7327, 0.5343], [0.6229, 0.4066, 0.7613, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350], [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117], [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461]]], device='cuda:0') loss_train_step before backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train: 0.1000079937512055 step: 28 running loss: 0.0035717140625430538 Train Steps: 28/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5737, 0.5700], [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5837, 0.5583], [ nan, nan, 0.9088, 0.3783, 0.4562, 0.2617, 0.6741, 0.5575], [0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5637, 0.5083], [0.6022, 0.3994, 0.8025, 0.3350, 0.3350, 0.4400, 0.5565, 0.5025]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7127, 0.4073, 0.8357, 0.5393, 0.4057, 0.5256, 0.6161, 0.5145], [0.5640, 0.3250, 0.8582, 0.3967, 0.4815, 0.4831, 0.5866, 0.5296], [0.4414, 0.2499, 0.8859, 0.3510, 0.4725, 0.3156, 0.6743, 0.5210], [0.5667, 0.3196, 0.8532, 0.3744, 0.4023, 0.2934, 0.6232, 0.5004], [0.5504, 0.3089, 0.8489, 0.3942, 0.4237, 0.3078, 0.5773, 0.4963], [0.7316, 0.4151, 0.8370, 0.5199, 0.3864, 0.4018, 0.6152, 0.4532], [0.6351, 0.3639, 0.8403, 0.4770, 0.4038, 0.4873, 0.5663, 0.4826], [0.5848, 0.3555, 0.7833, 0.3287, 0.3813, 0.4267, 0.5954, 0.5164]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5738, 0.5700], [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5838, 0.5583], [0.0000, 0.0000, 0.9087, 0.3783, 0.4563, 0.2617, 0.6741, 0.5575], [0.6277, 0.4117, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6226, 0.4098, 0.8913, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6272, 0.4071, 0.8737, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5638, 0.5083], [0.6022, 0.3994, 0.8025, 0.3350, 0.3350, 0.4400, 0.5565, 0.5025]]], device='cuda:0') loss_train_step before backward: tensor(0.0057, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0057, device='cuda:0', grad_fn=) loss_train: 0.10573010763619095 step: 29 running loss: 0.003645865780558309 Train Steps: 29/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667], [0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881], [0.6127, 0.4115, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495], [0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283], [ nan, nan, 0.9088, 0.3783, 0.4562, 0.2617, 0.6741, 0.5575], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5710, 0.3733, 0.7081, 0.3260, 0.3558, 0.3593, 0.5647, 0.5183], [0.6567, 0.4062, 0.8642, 0.4450, 0.4036, 0.5435, 0.6240, 0.4872], [0.5467, 0.3397, 0.8757, 0.5004, 0.4346, 0.4875, 0.5935, 0.5118], [0.6675, 0.4063, 0.8734, 0.5478, 0.3892, 0.4632, 0.5665, 0.5052], [0.5614, 0.3626, 0.7455, 0.3123, 0.3859, 0.3185, 0.5494, 0.5276], [0.6181, 0.3625, 0.8679, 0.5189, 0.4946, 0.4535, 0.6002, 0.4781], [0.4248, 0.2752, 0.9296, 0.4028, 0.4848, 0.3101, 0.6864, 0.5216], [0.6556, 0.4074, 0.8741, 0.4087, 0.3746, 0.4201, 0.6057, 0.4803]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6134, 0.4090, 0.6926, 0.2819, 0.3537, 0.3233, 0.5562, 0.5667], [0.6161, 0.4099, 0.8737, 0.4383, 0.3787, 0.5483, 0.5605, 0.5019], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881], [0.6127, 0.4114, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495], [0.6084, 0.4008, 0.8587, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283], [0.0000, 0.0000, 0.9087, 0.3783, 0.4563, 0.2617, 0.6741, 0.5575], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131]]], device='cuda:0') loss_train_step before backward: tensor(0.0052, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0052, device='cuda:0', grad_fn=) loss_train: 0.11088850733358413 step: 30 running loss: 0.003696283577786138 Train Steps: 30/90 Loss: 0.0037 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383], [ nan, nan, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533], [0.6110, 0.3984, 0.8750, 0.4933, 0.4625, 0.4950, 0.5578, 0.5676], [0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5637, 0.5083], [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750], [ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5922, 0.4017, 0.8713, 0.4405, 0.3612, 0.4238, 0.5184, 0.5151], [0.3582, 0.2298, 0.7556, 0.3063, 0.4014, 0.2411, 0.5185, 0.5444], [0.2946, 0.1886, 0.7170, 0.2601, 0.4136, 0.2340, 0.5188, 0.5606], [0.7030, 0.4408, 0.9341, 0.5412, 0.4614, 0.5772, 0.5865, 0.5240], [0.7953, 0.5029, 0.9016, 0.5643, 0.3886, 0.5612, 0.5478, 0.4993], [0.5802, 0.3809, 0.7125, 0.2922, 0.4219, 0.2625, 0.5282, 0.5478], [0.3664, 0.2401, 0.7380, 0.2883, 0.4304, 0.2344, 0.5059, 0.5507], [0.6292, 0.4025, 0.9526, 0.4345, 0.4087, 0.4079, 0.6802, 0.5200]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6082, 0.4024, 0.8737, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.0000, 0.0000, 0.7268, 0.2333, 0.4125, 0.1933, 0.5113, 0.5383], [0.0000, 0.0000, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533], [0.6110, 0.3984, 0.8750, 0.4933, 0.4625, 0.4950, 0.5578, 0.5676], [0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5638, 0.5083], [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5512, 0.5750], [0.0000, 0.0000, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6332, 0.4165, 0.9100, 0.3350, 0.4187, 0.3683, 0.7438, 0.5528]]], device='cuda:0') loss_train_step before backward: tensor(0.0100, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0100, device='cuda:0', grad_fn=) loss_train: 0.12087913427967578 step: 31 running loss: 0.003899326912247606 Train Steps: 31/90 Loss: 0.0039 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378], [0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767], [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350], [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183], [0.6183, 0.4076, 0.8838, 0.4517, 0.3813, 0.4483, 0.5775, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4806, 0.2998, 0.8918, 0.4452, 0.4107, 0.5104, 0.5396, 0.5420], [0.4820, 0.3437, 0.8743, 0.5086, 0.3730, 0.4249, 0.5120, 0.5413], [0.4814, 0.3564, 0.7628, 0.3216, 0.4086, 0.2817, 0.5722, 0.5560], [0.5544, 0.3869, 0.7628, 0.2905, 0.4290, 0.2236, 0.5466, 0.5370], [0.4566, 0.3315, 0.8674, 0.4907, 0.3861, 0.4601, 0.6026, 0.5633], [0.5705, 0.4234, 0.7099, 0.3097, 0.3921, 0.2522, 0.5098, 0.5591], [0.5269, 0.3574, 0.8905, 0.4368, 0.4251, 0.5545, 0.5447, 0.5600], [0.4295, 0.2856, 0.8784, 0.4406, 0.3848, 0.4491, 0.5455, 0.5579]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378], [0.6230, 0.4152, 0.7588, 0.2283, 0.4013, 0.2883, 0.6200, 0.5767], [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350], [0.6314, 0.4107, 0.8750, 0.5100, 0.3787, 0.4900, 0.7121, 0.5864], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183], [0.6183, 0.4076, 0.8838, 0.4517, 0.3812, 0.4483, 0.5775, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0040, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0040, device='cuda:0', grad_fn=) loss_train: 0.12484196957666427 step: 32 running loss: 0.0039013115492707584 Train Steps: 32/90 Loss: 0.0039 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852], [0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773], [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411], [0.6224, 0.4097, 0.7438, 0.2267, 0.3850, 0.2850, 0.5988, 0.5250], [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6312, 0.5250], [0.6353, 0.4128, 0.8488, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461], [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5385, 0.3726, 0.8863, 0.5260, 0.4290, 0.5047, 0.5160, 0.5818], [0.6499, 0.4474, 0.6659, 0.2844, 0.3601, 0.2414, 0.4926, 0.5716], [0.4714, 0.3264, 0.7618, 0.3114, 0.3952, 0.2498, 0.5081, 0.5671], [0.6327, 0.4728, 0.7146, 0.2756, 0.3698, 0.3003, 0.5684, 0.5530], [0.5012, 0.3294, 0.8662, 0.3438, 0.3913, 0.3386, 0.5953, 0.5621], [0.1845, 0.1296, 0.8389, 0.2707, 0.5110, 0.2170, 0.5929, 0.5841], [0.6692, 0.4743, 0.8151, 0.3947, 0.3312, 0.4365, 0.5337, 0.5487], [0.0382, 0.0638, 0.8499, 0.2784, 0.4862, 0.1974, 0.5803, 0.5828]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6140, 0.4070, 0.8700, 0.5000, 0.4613, 0.4900, 0.5260, 0.5852], [0.6147, 0.4026, 0.6600, 0.2467, 0.4087, 0.2150, 0.5489, 0.5773], [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411], [0.6224, 0.4097, 0.7437, 0.2267, 0.3850, 0.2850, 0.5987, 0.5250], [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6313, 0.5250], [0.6353, 0.4128, 0.8487, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6058, 0.3978, 0.8288, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461], [0.0000, 0.0000, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621]]], device='cuda:0') loss_train_step before backward: tensor(0.0065, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0065, device='cuda:0', grad_fn=) loss_train: 0.1313714758725837 step: 33 running loss: 0.003980953814320717 Train Steps: 33/90 Loss: 0.0040 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8888, 0.3100, 0.5262, 0.2817, 0.7145, 0.6003], [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823], [0.6104, 0.4029, 0.8738, 0.4900, 0.4088, 0.4533, 0.5070, 0.5510], [0.6136, 0.4085, 0.6688, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6361, 0.4076, 0.8862, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.2869, 0.1952, 0.8825, 0.2657, 0.4910, 0.2325, 0.6424, 0.5585], [0.5847, 0.4135, 0.7391, 0.3092, 0.3318, 0.3839, 0.5174, 0.6051], [0.5349, 0.3649, 0.8738, 0.4602, 0.4067, 0.4506, 0.5075, 0.5688], [0.5407, 0.3837, 0.6720, 0.2071, 0.3599, 0.2307, 0.5037, 0.5578], [0.3352, 0.2542, 0.6870, 0.2115, 0.3747, 0.2059, 0.5195, 0.5942], [0.4570, 0.3381, 0.8863, 0.4052, 0.4083, 0.2791, 0.6182, 0.5715], [0.4635, 0.3456, 0.8060, 0.3912, 0.4406, 0.2866, 0.5146, 0.6138], [0.6070, 0.4359, 0.8601, 0.5168, 0.3663, 0.4816, 0.6266, 0.5784]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8888, 0.3100, 0.5263, 0.2817, 0.7145, 0.6003], [0.6170, 0.4102, 0.7468, 0.3695, 0.3462, 0.3767, 0.5238, 0.5823], [0.6104, 0.4029, 0.8737, 0.4900, 0.4087, 0.4533, 0.5070, 0.5510], [0.6136, 0.4085, 0.6687, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6361, 0.4076, 0.8863, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297]]], device='cuda:0') loss_train_step before backward: tensor(0.0055, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0055, device='cuda:0', grad_fn=) loss_train: 0.1368311574915424 step: 34 running loss: 0.004024445808574776 Train Steps: 34/90 Loss: 0.0040 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207], [0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983], [0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485], [0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995], [0.6131, 0.4064, 0.8638, 0.5200, 0.4788, 0.4783, 0.5258, 0.5867]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4963, 0.3862, 0.7514, 0.2513, 0.3589, 0.3394, 0.5619, 0.5993], [0.4562, 0.3419, 0.8819, 0.4533, 0.4124, 0.4145, 0.5854, 0.5761], [0.6197, 0.4237, 0.8685, 0.5307, 0.3926, 0.4467, 0.5880, 0.6016], [0.5460, 0.3521, 0.9146, 0.4189, 0.4082, 0.4825, 0.5842, 0.5735], [0.5055, 0.3693, 0.7325, 0.2684, 0.3590, 0.2630, 0.5277, 0.6086], [0.3609, 0.2583, 0.6952, 0.1869, 0.4173, 0.1355, 0.5613, 0.5715], [0.5359, 0.4054, 0.7737, 0.1995, 0.4128, 0.1637, 0.6384, 0.5586], [0.5642, 0.3956, 0.8697, 0.4497, 0.4499, 0.4201, 0.6075, 0.5898]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207], [0.6219, 0.3934, 0.8687, 0.5267, 0.4313, 0.4967, 0.5987, 0.4983], [0.6310, 0.4017, 0.8562, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6123, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485], [0.6054, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995], [0.6132, 0.4063, 0.8637, 0.5200, 0.4787, 0.4783, 0.5258, 0.5867]]], device='cuda:0') loss_train_step before backward: tensor(0.0043, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0043, device='cuda:0', grad_fn=) loss_train: 0.14112150168512017 step: 35 running loss: 0.004032042905289148 Train Steps: 35/90 Loss: 0.0040 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693], [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500], [0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [ nan, nan, 0.7192, 0.2346, 0.4037, 0.2050, 0.5138, 0.5650]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4819, 0.3458, 0.6869, 0.2910, 0.3688, 0.2382, 0.5282, 0.5730], [0.5417, 0.3932, 0.8482, 0.2817, 0.5320, 0.2175, 0.7606, 0.5362], [0.6417, 0.4459, 0.8624, 0.4964, 0.3670, 0.4702, 0.5220, 0.5442], [0.6376, 0.4358, 0.8091, 0.2879, 0.3997, 0.2430, 0.6488, 0.5202], [0.6563, 0.4553, 0.7634, 0.2758, 0.3470, 0.3813, 0.6249, 0.5541], [0.5801, 0.4157, 0.7382, 0.2946, 0.3624, 0.3834, 0.5646, 0.5530], [0.5191, 0.3747, 0.8364, 0.2815, 0.5005, 0.2101, 0.6680, 0.5510], [0.2037, 0.1414, 0.7333, 0.2334, 0.4252, 0.2144, 0.5101, 0.5678]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.6038, 0.3946, 0.8413, 0.4883, 0.3562, 0.4550, 0.5266, 0.4693], [0.6257, 0.4034, 0.8288, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6197, 0.4051, 0.7812, 0.2650, 0.3512, 0.4050, 0.6112, 0.5500], [0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.0000, 0.0000, 0.7192, 0.2346, 0.4038, 0.2050, 0.5138, 0.5650]]], device='cuda:0') loss_train_step before backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train: 0.14336391224060208 step: 36 running loss: 0.00398233089557228 Train Steps: 36/90 Loss: 0.0040 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6081, 0.3950, 0.8538, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954], [0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748], [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817], [0.6177, 0.4086, 0.8738, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7228, 0.4681, 0.7525, 0.3056, 0.3635, 0.3611, 0.5924, 0.4883], [0.6475, 0.4415, 0.7789, 0.4468, 0.3754, 0.3749, 0.5925, 0.5845], [0.5587, 0.3827, 0.8006, 0.4904, 0.4415, 0.3936, 0.5719, 0.4867], [0.4817, 0.3434, 0.8249, 0.4164, 0.4155, 0.4739, 0.6631, 0.5691], [0.5975, 0.3980, 0.8270, 0.3494, 0.3865, 0.3966, 0.5595, 0.4871], [0.5687, 0.3749, 0.8677, 0.3652, 0.4322, 0.1858, 0.6998, 0.4778], [0.4849, 0.3461, 0.7653, 0.2249, 0.4421, 0.1733, 0.6068, 0.5384], [0.5811, 0.3800, 0.8053, 0.2997, 0.3964, 0.4467, 0.6727, 0.5288]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6081, 0.3950, 0.8537, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954], [0.6329, 0.4196, 0.9237, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748], [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817], [0.6177, 0.4085, 0.8737, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700]]], device='cuda:0') loss_train_step before backward: tensor(0.0046, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0046, device='cuda:0', grad_fn=) loss_train: 0.14793575985822827 step: 37 running loss: 0.003998263779952115 Train Steps: 37/90 Loss: 0.0040 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611], [0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804], [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823], [0.6201, 0.4017, 0.8871, 0.4621, 0.3517, 0.4675, 0.5999, 0.5106], [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835], [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097], [0.6205, 0.4062, 0.8337, 0.2683, 0.3675, 0.4283, 0.6338, 0.5250]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5927, 0.3793, 0.8371, 0.4345, 0.4220, 0.4288, 0.6075, 0.5332], [0.6252, 0.3783, 0.8342, 0.2596, 0.4505, 0.2143, 0.6791, 0.5136], [0.6643, 0.4283, 0.8630, 0.4127, 0.3536, 0.4288, 0.6292, 0.4873], [0.6172, 0.4089, 0.7182, 0.3220, 0.3551, 0.3615, 0.5789, 0.5728], [0.6835, 0.4250, 0.7887, 0.4249, 0.3495, 0.4700, 0.5967, 0.5164], [0.6390, 0.4108, 0.8453, 0.3511, 0.4372, 0.1673, 0.6323, 0.4891], [0.5556, 0.3422, 0.8279, 0.4370, 0.4475, 0.4216, 0.5880, 0.5037], [0.6281, 0.4205, 0.7479, 0.2551, 0.3832, 0.3778, 0.6559, 0.5277]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611], [0.6250, 0.4131, 0.8687, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6263, 0.4038, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804], [0.6170, 0.4102, 0.7468, 0.3695, 0.3462, 0.3767, 0.5238, 0.5823], [0.6201, 0.4017, 0.8871, 0.4621, 0.3517, 0.4675, 0.5999, 0.5106], [0.6273, 0.4110, 0.8900, 0.3817, 0.4187, 0.2167, 0.5858, 0.4835], [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097], [0.6205, 0.4062, 0.8338, 0.2683, 0.3675, 0.4283, 0.6338, 0.5250]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.1494324274826795 step: 38 running loss: 0.003932432302175776 Train Steps: 38/90 Loss: 0.0039 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6058, 0.3986, 0.8324, 0.4626, 0.3838, 0.4983, 0.5147, 0.5466], [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864], [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114], [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822], [0.6034, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6082, 0.3901, 0.8104, 0.4446, 0.3821, 0.4699, 0.5674, 0.5060], [0.6610, 0.4154, 0.8537, 0.5049, 0.3697, 0.4719, 0.6829, 0.5151], [0.6313, 0.4098, 0.8679, 0.4890, 0.4500, 0.4796, 0.6458, 0.4985], [0.6464, 0.4064, 0.8509, 0.4027, 0.3388, 0.3724, 0.5377, 0.5235], [0.7227, 0.4560, 0.6997, 0.2580, 0.3332, 0.3190, 0.5898, 0.4940], [0.6985, 0.4387, 0.7837, 0.2486, 0.4359, 0.2209, 0.6642, 0.5135], [0.6163, 0.3792, 0.8439, 0.4967, 0.4222, 0.4776, 0.5871, 0.5058], [0.6270, 0.3982, 0.8587, 0.4695, 0.4115, 0.5075, 0.6632, 0.5138]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6058, 0.3986, 0.8324, 0.4626, 0.3837, 0.4983, 0.5147, 0.5466], [0.6314, 0.4107, 0.8750, 0.5100, 0.3787, 0.4900, 0.7121, 0.5864], [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114], [0.6182, 0.4058, 0.8737, 0.4350, 0.3562, 0.3400, 0.5290, 0.5822], [0.6033, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6098, 0.3991, 0.8637, 0.4717, 0.4263, 0.4967, 0.5213, 0.5650], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.1505707015749067 step: 39 running loss: 0.0038607872198694027 Train Steps: 39/90 Loss: 0.0039 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683], [0.6166, 0.4008, 0.8563, 0.5667, 0.4388, 0.4933, 0.5575, 0.5567], [0.6198, 0.4115, 0.7762, 0.2717, 0.3713, 0.3200, 0.5837, 0.5683], [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667], [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378], [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6763, 0.4203, 0.8212, 0.3882, 0.3392, 0.4437, 0.5854, 0.5210], [0.5837, 0.3779, 0.8425, 0.5387, 0.4452, 0.5220, 0.5871, 0.5394], [0.6857, 0.4187, 0.7855, 0.3141, 0.3728, 0.3482, 0.5546, 0.5442], [0.6829, 0.4323, 0.7807, 0.3014, 0.3951, 0.3038, 0.6372, 0.4918], [0.6924, 0.4353, 0.8441, 0.4242, 0.3433, 0.4517, 0.5365, 0.5052], [0.6510, 0.3992, 0.8452, 0.3819, 0.3662, 0.4419, 0.6248, 0.5265], [0.6469, 0.4096, 0.8938, 0.4680, 0.4248, 0.3783, 0.6646, 0.4929], [0.7247, 0.4555, 0.8720, 0.4785, 0.3858, 0.3604, 0.6007, 0.4724]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6219, 0.4097, 0.8737, 0.3400, 0.3562, 0.4117, 0.5975, 0.5683], [0.6166, 0.4008, 0.8562, 0.5667, 0.4387, 0.4933, 0.5575, 0.5567], [0.6198, 0.4115, 0.7763, 0.2717, 0.3713, 0.3200, 0.5838, 0.5683], [0.6257, 0.4034, 0.8288, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.6257, 0.4167, 0.8775, 0.3433, 0.3562, 0.4133, 0.6200, 0.5667], [0.6307, 0.4029, 0.8988, 0.4817, 0.3938, 0.3500, 0.7311, 0.5378], [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.15187458659056574 step: 40 running loss: 0.0037968646647641435 Train Steps: 40/90 Loss: 0.0038 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617], [0.6148, 0.3996, 0.8488, 0.3867, 0.3488, 0.4067, 0.5863, 0.5000], [0.6179, 0.4082, 0.6688, 0.2667, 0.3588, 0.3317, 0.5750, 0.5783], [0.6261, 0.3987, 0.8688, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309], [0.6317, 0.4038, 0.8287, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986], [ nan, nan, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702], [0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7187, 0.4657, 0.8306, 0.2767, 0.4167, 0.2786, 0.6047, 0.5362], [0.6980, 0.4498, 0.8665, 0.3995, 0.3762, 0.4197, 0.5474, 0.5448], [0.6695, 0.4380, 0.9158, 0.4737, 0.3652, 0.5028, 0.6096, 0.5110], [0.7887, 0.4816, 0.7583, 0.3533, 0.3352, 0.4078, 0.5599, 0.5510], [0.6274, 0.4131, 0.9500, 0.5838, 0.4386, 0.6303, 0.6465, 0.5380], [0.7863, 0.5082, 0.9138, 0.6277, 0.3542, 0.5731, 0.5936, 0.5166], [0.5299, 0.3224, 0.9347, 0.3950, 0.4768, 0.3456, 0.6791, 0.5457], [0.6023, 0.3797, 0.7298, 0.2535, 0.3751, 0.2667, 0.5376, 0.5283]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6202, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617], [0.6148, 0.3996, 0.8487, 0.3867, 0.3487, 0.4067, 0.5863, 0.5000], [0.6179, 0.4082, 0.6687, 0.2667, 0.3587, 0.3317, 0.5750, 0.5783], [0.6261, 0.3987, 0.8687, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309], [0.6317, 0.4038, 0.8288, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986], [0.0000, 0.0000, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702], [0.6055, 0.4015, 0.7425, 0.2033, 0.4112, 0.1883, 0.5217, 0.4823]]], device='cuda:0') loss_train_step before backward: tensor(0.0102, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0102, device='cuda:0', grad_fn=) loss_train: 0.16205621568951756 step: 41 running loss: 0.003952590626573599 Train Steps: 41/90 Loss: 0.0040 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6092, 0.4001, 0.8638, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064], [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283], [0.6147, 0.4107, 0.8137, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412], [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170], [0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6168, 0.3942, 0.8755, 0.5418, 0.4090, 0.6057, 0.5533, 0.5243], [0.6289, 0.3965, 0.9262, 0.4993, 0.3608, 0.5306, 0.6536, 0.5308], [0.7041, 0.4402, 0.9332, 0.4216, 0.4070, 0.3354, 0.6179, 0.5408], [0.5029, 0.3301, 0.8102, 0.3966, 0.3733, 0.3770, 0.5184, 0.5792], [0.6606, 0.4101, 0.8494, 0.2925, 0.4726, 0.2737, 0.6511, 0.5564], [0.7221, 0.4742, 0.8928, 0.5254, 0.3563, 0.5033, 0.5423, 0.5139], [0.5314, 0.3336, 0.8073, 0.3495, 0.4271, 0.3116, 0.5455, 0.5472], [0.6883, 0.4347, 0.7809, 0.2945, 0.4331, 0.2745, 0.5999, 0.5528]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6092, 0.4001, 0.8637, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064], [0.6263, 0.4065, 0.9038, 0.4317, 0.3587, 0.4550, 0.6325, 0.5250], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6187, 0.5283], [0.6147, 0.4107, 0.8138, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412], [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170], [0.6108, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6143, 0.4040, 0.8238, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284]]], device='cuda:0') loss_train_step before backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train: 0.16464617906603962 step: 42 running loss: 0.003920147120619991 Train Steps: 42/90 Loss: 0.0039 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6063, 0.5567], [0.6260, 0.4120, 0.8013, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895], [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378], [ nan, nan, 0.7225, 0.2167, 0.3987, 0.2283, 0.5427, 0.5181], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717], [0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927], [0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5914, 0.3750, 0.9267, 0.4842, 0.4291, 0.5331, 0.5822, 0.5713], [0.6727, 0.4229, 0.8425, 0.2422, 0.4783, 0.1994, 0.6245, 0.5530], [0.6520, 0.4122, 0.9716, 0.4921, 0.4072, 0.3805, 0.6626, 0.5454], [0.4037, 0.2400, 0.7607, 0.2332, 0.4031, 0.2527, 0.5299, 0.5721], [0.6678, 0.4063, 0.9150, 0.5728, 0.4048, 0.5287, 0.5675, 0.5446], [0.7441, 0.4790, 0.9230, 0.5637, 0.3632, 0.5035, 0.5689, 0.5169], [0.6529, 0.4008, 0.8589, 0.2868, 0.4395, 0.2580, 0.6256, 0.5510], [0.5707, 0.3659, 0.9312, 0.4682, 0.4123, 0.6192, 0.5800, 0.5642]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6062, 0.5567], [0.6259, 0.4120, 0.8012, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895], [0.6307, 0.4029, 0.8988, 0.4817, 0.3938, 0.3500, 0.7311, 0.5378], [0.0000, 0.0000, 0.7225, 0.2167, 0.3988, 0.2283, 0.5427, 0.5181], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5412, 0.5717], [0.6321, 0.4048, 0.8737, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927], [0.6277, 0.4056, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116]]], device='cuda:0') loss_train_step before backward: tensor(0.0052, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0052, device='cuda:0', grad_fn=) loss_train: 0.16981427988503128 step: 43 running loss: 0.00394916929965189 Train Steps: 43/90 Loss: 0.0039 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200], [0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019], [ nan, nan, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983], [0.6198, 0.4130, 0.8762, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869], [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750], [0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5744, 0.3610, 0.8981, 0.3053, 0.4541, 0.2640, 0.6367, 0.5374], [0.6513, 0.4209, 0.9170, 0.4760, 0.3727, 0.5281, 0.6176, 0.5163], [0.3297, 0.2064, 0.7455, 0.2015, 0.4621, 0.1717, 0.5486, 0.5469], [0.6995, 0.4713, 0.9346, 0.4540, 0.3628, 0.4868, 0.5914, 0.5479], [0.5862, 0.3670, 0.9249, 0.5698, 0.4216, 0.4958, 0.5751, 0.5130], [0.5996, 0.3651, 0.8751, 0.3118, 0.4596, 0.2157, 0.5981, 0.5374], [0.6333, 0.4009, 0.9072, 0.5587, 0.4147, 0.4863, 0.5798, 0.5360], [0.5015, 0.3147, 0.8200, 0.2512, 0.4896, 0.1724, 0.5748, 0.5407]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200], [0.6161, 0.4099, 0.8737, 0.4383, 0.3787, 0.5483, 0.5605, 0.5019], [0.0000, 0.0000, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983], [0.6198, 0.4130, 0.8763, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869], [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750], [0.6286, 0.4078, 0.8062, 0.2267, 0.4787, 0.1533, 0.5953, 0.4913]]], device='cuda:0') loss_train_step before backward: tensor(0.0039, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0039, device='cuda:0', grad_fn=) loss_train: 0.1736702184425667 step: 44 running loss: 0.003947050419149243 Train Steps: 44/90 Loss: 0.0039 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5988, 0.5700], [0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693], [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767], [0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6188, 0.5400], [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500], [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450], [0.6286, 0.4060, 0.9188, 0.4333, 0.3675, 0.4167, 0.7034, 0.5528], [0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6262, 0.5367]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4946, 0.3198, 0.8330, 0.2777, 0.4325, 0.2888, 0.5966, 0.5509], [0.5412, 0.3576, 0.8926, 0.4817, 0.4078, 0.4237, 0.5483, 0.4943], [0.5436, 0.3551, 0.8983, 0.5374, 0.4359, 0.4431, 0.5623, 0.5453], [0.6110, 0.3883, 0.8518, 0.3097, 0.4324, 0.2751, 0.5913, 0.5272], [0.5764, 0.3854, 0.8313, 0.2675, 0.3994, 0.3353, 0.5966, 0.5245], [0.6086, 0.3816, 0.9137, 0.5039, 0.4031, 0.3933, 0.5720, 0.5005], [0.5202, 0.3427, 0.9587, 0.4177, 0.4165, 0.3402, 0.6581, 0.5295], [0.5557, 0.3643, 0.9287, 0.4716, 0.4642, 0.5385, 0.6326, 0.5063]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5987, 0.5700], [0.6038, 0.3946, 0.8413, 0.4883, 0.3562, 0.4550, 0.5266, 0.4693], [0.6163, 0.4001, 0.8788, 0.5033, 0.4013, 0.4633, 0.5337, 0.5767], [0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6187, 0.5400], [0.6197, 0.4051, 0.7812, 0.2650, 0.3512, 0.4050, 0.6112, 0.5500], [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450], [0.6286, 0.4060, 0.9187, 0.4333, 0.3675, 0.4167, 0.7034, 0.5528], [0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6263, 0.5367]]], device='cuda:0') loss_train_step before backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train: 0.17585839342791587 step: 45 running loss: 0.003907964298398131 Train Steps: 45/90 Loss: 0.0039 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6224, 0.4061, 0.8988, 0.4300, 0.3838, 0.4750, 0.6112, 0.5483], [0.6339, 0.4102, 0.8588, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089], [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611], [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550], [0.6152, 0.4131, 0.6863, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467], [0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467], [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5982, 0.3846, 0.9121, 0.4259, 0.3752, 0.5248, 0.6178, 0.4911], [0.5487, 0.3555, 0.8807, 0.2899, 0.4626, 0.1906, 0.6620, 0.4942], [0.5193, 0.3321, 0.8960, 0.4542, 0.4383, 0.4488, 0.5616, 0.5277], [0.4869, 0.3403, 0.9145, 0.4456, 0.4816, 0.4387, 0.6093, 0.5287], [0.5725, 0.3812, 0.7642, 0.2518, 0.3758, 0.2768, 0.5590, 0.5411], [0.5903, 0.3854, 0.7478, 0.2506, 0.3907, 0.2212, 0.5636, 0.5075], [0.5732, 0.3680, 0.8950, 0.5185, 0.4251, 0.4477, 0.5844, 0.5426], [0.3919, 0.2581, 0.7479, 0.2186, 0.4290, 0.1590, 0.5343, 0.5263]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6224, 0.4061, 0.8988, 0.4300, 0.3837, 0.4750, 0.6112, 0.5483], [0.6339, 0.4102, 0.8587, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089], [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611], [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550], [0.6152, 0.4131, 0.6862, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [0.6162, 0.4134, 0.6700, 0.2467, 0.3963, 0.2533, 0.5738, 0.5467], [0.6273, 0.4143, 0.8750, 0.5700, 0.3988, 0.4717, 0.6012, 0.5467], [0.6159, 0.4085, 0.6900, 0.2283, 0.4087, 0.1950, 0.5123, 0.5397]]], device='cuda:0') loss_train_step before backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train: 0.1786146817030385 step: 46 running loss: 0.003882927863109533 Train Steps: 46/90 Loss: 0.0039 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967], [0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297], [0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413], [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373], [ nan, nan, 0.8938, 0.2850, 0.4662, 0.3117, 0.7406, 0.5528], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6223, 0.4028, 0.8988, 0.4200, 0.3763, 0.5733, 0.6375, 0.5167], [0.6183, 0.4076, 0.8838, 0.4517, 0.3813, 0.4483, 0.5775, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5902, 0.4247, 0.8342, 0.4375, 0.4049, 0.2877, 0.5761, 0.4922], [0.5506, 0.3890, 0.8556, 0.3736, 0.4278, 0.2765, 0.5867, 0.5182], [0.4517, 0.3155, 0.8016, 0.1960, 0.5234, 0.2164, 0.6464, 0.5249], [0.5782, 0.3937, 0.7643, 0.4926, 0.3738, 0.4246, 0.5473, 0.5271], [0.3472, 0.2472, 0.8567, 0.2544, 0.4739, 0.2547, 0.6857, 0.5190], [0.5657, 0.3600, 0.8437, 0.4191, 0.3642, 0.3413, 0.5603, 0.4914], [0.5481, 0.3568, 0.8340, 0.3782, 0.3921, 0.5499, 0.6102, 0.4830], [0.5766, 0.3872, 0.8331, 0.4085, 0.3848, 0.4229, 0.5482, 0.5033]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967], [0.6364, 0.4165, 0.9087, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297], [0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413], [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373], [0.0000, 0.0000, 0.8938, 0.2850, 0.4663, 0.3117, 0.7406, 0.5528], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6223, 0.4028, 0.8988, 0.4200, 0.3762, 0.5733, 0.6375, 0.5167], [0.6183, 0.4076, 0.8838, 0.4517, 0.3812, 0.4483, 0.5775, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0054, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0054, device='cuda:0', grad_fn=) loss_train: 0.18397039466071874 step: 47 running loss: 0.003914263716185505 Train Steps: 47/90 Loss: 0.0039 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064], [0.6357, 0.4097, 0.9038, 0.3883, 0.4213, 0.2950, 0.6686, 0.5390], [0.6248, 0.4032, 0.7738, 0.1900, 0.4813, 0.1400, 0.5941, 0.4904], [0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136], [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5837, 0.5167], [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5356, 0.3720, 0.8169, 0.4604, 0.4491, 0.4540, 0.5478, 0.5239], [0.5339, 0.3525, 0.8122, 0.3496, 0.3886, 0.2890, 0.6373, 0.5348], [0.4764, 0.3318, 0.7208, 0.1869, 0.4643, 0.1142, 0.5922, 0.5236], [0.5076, 0.3286, 0.7992, 0.3146, 0.3590, 0.5195, 0.6077, 0.5179], [0.5510, 0.3645, 0.8016, 0.3514, 0.3689, 0.3138, 0.5630, 0.5140], [0.5081, 0.3545, 0.8147, 0.3891, 0.3914, 0.3503, 0.6489, 0.5292], [0.6319, 0.4548, 0.8447, 0.4583, 0.3428, 0.4245, 0.6191, 0.5207], [0.5233, 0.3743, 0.8006, 0.2742, 0.4410, 0.2303, 0.6163, 0.5454]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6128, 0.4022, 0.8737, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064], [0.6357, 0.4097, 0.9038, 0.3883, 0.4212, 0.2950, 0.6686, 0.5390], [0.6248, 0.4032, 0.7738, 0.1900, 0.4812, 0.1400, 0.5941, 0.4904], [0.6132, 0.4118, 0.8200, 0.3633, 0.3562, 0.5400, 0.5787, 0.5136], [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5838, 0.5167], [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274]]], device='cuda:0') loss_train_step before backward: tensor(0.0029, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0029, device='cuda:0', grad_fn=) loss_train: 0.18689397221896797 step: 48 running loss: 0.0038936244212284996 Train Steps: 48/90 Loss: 0.0039 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901], [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617], [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183], [0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783], [0.6131, 0.4037, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5783, 0.3915, 0.8595, 0.4291, 0.3632, 0.4066, 0.5616, 0.5293], [0.5297, 0.3724, 0.7324, 0.2156, 0.4202, 0.1592, 0.6418, 0.5288], [0.6896, 0.4471, 0.8466, 0.3719, 0.3241, 0.4163, 0.6480, 0.5045], [0.5473, 0.3957, 0.8311, 0.4912, 0.4214, 0.5023, 0.5809, 0.5467], [0.6278, 0.4192, 0.8578, 0.5170, 0.3710, 0.4178, 0.6087, 0.5449], [0.5861, 0.3984, 0.8378, 0.4504, 0.4305, 0.5174, 0.6835, 0.5589], [0.4583, 0.3411, 0.7667, 0.2237, 0.4115, 0.2177, 0.6164, 0.5766], [0.5240, 0.3749, 0.6977, 0.2363, 0.3559, 0.2672, 0.5716, 0.5624]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901], [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617], [0.6251, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6087, 0.5183], [0.6229, 0.4198, 0.7663, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783], [0.6131, 0.4036, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.1887947202194482 step: 49 running loss: 0.00385295347386629 Train Steps: 49/90 Loss: 0.0039 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901], [0.6304, 0.4024, 0.8925, 0.4800, 0.3937, 0.4817, 0.7485, 0.5297], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5737, 0.5350], [0.6200, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367], [0.6226, 0.4103, 0.8575, 0.3450, 0.4388, 0.2067, 0.5787, 0.5383]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5907, 0.4206, 0.8311, 0.4863, 0.3859, 0.4446, 0.6255, 0.5465], [0.4468, 0.3266, 0.7506, 0.2761, 0.3689, 0.2431, 0.5742, 0.5514], [0.5799, 0.3804, 0.8372, 0.4546, 0.3878, 0.4680, 0.6465, 0.5339], [0.6617, 0.4215, 0.7334, 0.2241, 0.4029, 0.1739, 0.6141, 0.5315], [0.6885, 0.4408, 0.8412, 0.3858, 0.3437, 0.4316, 0.6301, 0.5134], [0.6789, 0.4871, 0.8510, 0.4682, 0.3937, 0.3102, 0.6101, 0.5566], [0.5780, 0.3749, 0.8523, 0.4565, 0.3673, 0.5337, 0.6059, 0.5508], [0.5551, 0.4059, 0.8128, 0.3384, 0.4452, 0.2528, 0.6119, 0.5846]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901], [0.6304, 0.4024, 0.8925, 0.4800, 0.3938, 0.4817, 0.7485, 0.5297], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5788, 0.5117], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5738, 0.5350], [0.6199, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367], [0.6226, 0.4103, 0.8575, 0.3450, 0.4387, 0.2067, 0.5788, 0.5383]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.19071150361560285 step: 50 running loss: 0.003814230072312057 Train Steps: 50/90 Loss: 0.0038 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267], [0.6188, 0.4099, 0.7400, 0.2433, 0.3962, 0.2750, 0.6162, 0.5467], [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783], [0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6429, 0.4160, 0.8269, 0.2762, 0.4544, 0.2121, 0.6649, 0.5159], [0.6870, 0.4637, 0.7694, 0.2872, 0.3812, 0.3079, 0.6157, 0.5234], [0.3769, 0.2598, 0.7436, 0.2793, 0.4277, 0.2608, 0.5515, 0.5496], [0.7011, 0.4604, 0.8506, 0.3975, 0.3754, 0.3619, 0.5322, 0.5614], [0.5484, 0.3436, 0.7126, 0.2348, 0.4148, 0.2125, 0.5711, 0.4997], [0.6928, 0.4471, 0.8586, 0.2866, 0.3946, 0.2889, 0.6364, 0.4866], [0.7239, 0.4933, 0.8775, 0.5449, 0.4109, 0.5543, 0.5762, 0.5451], [0.6751, 0.4653, 0.7992, 0.3384, 0.3962, 0.3202, 0.6100, 0.5912]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267], [0.6188, 0.4099, 0.7400, 0.2433, 0.3963, 0.2750, 0.6162, 0.5467], [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783], [0.6054, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6257, 0.4034, 0.8288, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167]]], device='cuda:0') loss_train_step before backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train: 0.1935047539882362 step: 51 running loss: 0.003794210862514435 Train Steps: 51/90 Loss: 0.0038 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650], [ nan, nan, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583], [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6153, 0.4117, 0.8688, 0.5167, 0.4895, 0.5647, 0.5524, 0.5136], [0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583], [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7185, 0.4708, 0.8618, 0.5070, 0.4093, 0.3858, 0.6009, 0.5121], [0.8135, 0.4977, 0.8833, 0.4586, 0.3953, 0.5580, 0.6702, 0.5256], [0.7068, 0.4859, 0.8293, 0.3565, 0.3601, 0.2996, 0.6099, 0.5512], [0.3108, 0.2245, 0.7162, 0.2091, 0.4363, 0.1829, 0.5488, 0.5426], [0.7652, 0.4867, 0.8227, 0.5169, 0.3867, 0.4359, 0.5665, 0.5834], [0.6891, 0.4566, 0.8543, 0.4532, 0.4490, 0.4629, 0.5988, 0.5428], [0.6487, 0.4061, 0.9048, 0.4152, 0.3674, 0.4081, 0.5812, 0.5541], [0.6153, 0.4052, 0.7894, 0.2775, 0.4123, 0.2320, 0.5595, 0.5471]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650], [0.0000, 0.0000, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583], [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6154, 0.4117, 0.8687, 0.5167, 0.4895, 0.5647, 0.5524, 0.5136], [0.6212, 0.4033, 0.8938, 0.4167, 0.3812, 0.4267, 0.5612, 0.5583], [0.6182, 0.4099, 0.7812, 0.3000, 0.3938, 0.2367, 0.5325, 0.5750]]], device='cuda:0') loss_train_step before backward: tensor(0.0048, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0048, device='cuda:0', grad_fn=) loss_train: 0.19832616718485951 step: 52 running loss: 0.0038139647535549905 Train Steps: 52/90 Loss: 0.0038 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433], [ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767], [0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799], [0.6160, 0.4093, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808], [0.6131, 0.4037, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680], [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6246, 0.3846, 0.7061, 0.2208, 0.4330, 0.1798, 0.5548, 0.5012], [0.7532, 0.4474, 0.9285, 0.5140, 0.4293, 0.5499, 0.6269, 0.5106], [0.6529, 0.4466, 0.8533, 0.3775, 0.3637, 0.3183, 0.5681, 0.5332], [0.4785, 0.2866, 0.8067, 0.3179, 0.4354, 0.2605, 0.5388, 0.5638], [0.7072, 0.4532, 0.8808, 0.4216, 0.4210, 0.3280, 0.5693, 0.5640], [0.6810, 0.4414, 0.8535, 0.4714, 0.3980, 0.4581, 0.5754, 0.5703], [0.6754, 0.4343, 0.7393, 0.2802, 0.3996, 0.2901, 0.5359, 0.5503], [0.6758, 0.4444, 0.8548, 0.4092, 0.3865, 0.3137, 0.5375, 0.5456]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6054, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5412, 0.5433], [0.0000, 0.0000, 0.7613, 0.3250, 0.4038, 0.2533, 0.5437, 0.5767], [0.6250, 0.4236, 0.8637, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799], [0.6160, 0.4092, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808], [0.6131, 0.4036, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680], [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896]]], device='cuda:0') loss_train_step before backward: tensor(0.0060, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0060, device='cuda:0', grad_fn=) loss_train: 0.2043646709062159 step: 53 running loss: 0.003855937186909734 Train Steps: 53/90 Loss: 0.0039 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6127, 0.4118, 0.8650, 0.5083, 0.4088, 0.5367, 0.5300, 0.5456], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6125, 0.3974, 0.7725, 0.2517, 0.3538, 0.3317, 0.5887, 0.5500], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170], [0.6222, 0.3957, 0.8838, 0.5017, 0.3937, 0.4600, 0.5900, 0.5017], [0.6164, 0.4119, 0.7913, 0.2650, 0.3538, 0.3500, 0.5614, 0.5038]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6823, 0.4182, 0.8925, 0.5127, 0.3848, 0.3467, 0.5078, 0.5746], [0.6324, 0.3610, 0.8778, 0.5452, 0.4411, 0.5536, 0.5033, 0.5969], [0.6364, 0.3702, 0.8149, 0.3233, 0.4203, 0.2426, 0.5644, 0.5335], [0.6659, 0.3918, 0.7969, 0.2866, 0.3929, 0.3467, 0.5385, 0.5664], [0.6106, 0.3540, 0.8478, 0.4741, 0.3870, 0.4879, 0.5421, 0.5649], [0.6444, 0.3753, 0.8584, 0.2748, 0.4906, 0.1783, 0.6106, 0.5320], [0.6796, 0.4049, 0.9135, 0.5595, 0.4094, 0.4721, 0.5515, 0.5398], [0.7185, 0.4458, 0.8179, 0.2925, 0.3749, 0.3738, 0.5725, 0.5191]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6127, 0.4118, 0.8650, 0.5083, 0.4087, 0.5367, 0.5300, 0.5456], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6125, 0.3974, 0.7725, 0.2517, 0.3537, 0.3317, 0.5888, 0.5500], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170], [0.6222, 0.3957, 0.8838, 0.5017, 0.3938, 0.4600, 0.5900, 0.5017], [0.6164, 0.4119, 0.7912, 0.2650, 0.3537, 0.3500, 0.5614, 0.5038]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.20535756752360612 step: 54 running loss: 0.003802917917103817 Train Steps: 54/90 Loss: 0.0038 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693], [ nan, nan, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533], [0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5917, 0.3723, 0.8891, 0.4751, 0.4165, 0.5489, 0.6182, 0.5760], [0.7117, 0.4290, 0.8914, 0.4276, 0.4067, 0.2642, 0.5423, 0.5531], [0.5924, 0.3717, 0.8861, 0.4923, 0.3642, 0.4655, 0.5262, 0.5317], [0.2581, 0.1246, 0.7177, 0.2037, 0.4130, 0.1838, 0.5022, 0.5447], [0.6910, 0.4236, 0.8848, 0.5404, 0.4288, 0.5179, 0.5153, 0.5370], [0.7144, 0.4450, 0.8453, 0.4453, 0.4477, 0.3143, 0.5251, 0.5701], [0.5903, 0.3559, 0.7124, 0.2420, 0.3795, 0.2064, 0.4764, 0.5346], [0.6276, 0.3709, 0.8845, 0.5027, 0.3586, 0.4799, 0.5622, 0.5465]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6241, 0.4142, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.6226, 0.4098, 0.8913, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6038, 0.3946, 0.8413, 0.4883, 0.3562, 0.4550, 0.5266, 0.4693], [0.0000, 0.0000, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533], [0.6097, 0.3988, 0.8650, 0.5250, 0.4212, 0.5200, 0.5675, 0.5050], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6149, 0.4054, 0.6712, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733]]], device='cuda:0') loss_train_step before backward: tensor(0.0023, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0023, device='cuda:0', grad_fn=) loss_train: 0.20769805239979178 step: 55 running loss: 0.0037763282254507594 Train Steps: 55/90 Loss: 0.0038 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.3993, 0.8519, 0.4923, 0.3962, 0.4717, 0.6013, 0.5433], [0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683], [0.6166, 0.4008, 0.8563, 0.5667, 0.4388, 0.4933, 0.5575, 0.5567], [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947], [0.6136, 0.4117, 0.8700, 0.5167, 0.4188, 0.5083, 0.5147, 0.5495], [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6147, 0.4112, 0.7988, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5744, 0.3695, 0.8922, 0.5000, 0.3746, 0.4588, 0.5700, 0.5711], [0.6601, 0.4320, 0.8350, 0.3069, 0.3937, 0.2802, 0.5901, 0.5480], [0.5936, 0.3623, 0.8516, 0.5368, 0.4231, 0.4702, 0.5124, 0.5805], [0.5631, 0.3485, 0.7975, 0.2137, 0.4321, 0.2081, 0.6074, 0.5147], [0.6031, 0.3717, 0.8932, 0.5464, 0.4023, 0.4963, 0.5162, 0.5645], [0.5450, 0.3490, 0.8726, 0.5144, 0.4051, 0.4960, 0.5226, 0.5375], [0.6051, 0.3576, 0.8927, 0.4888, 0.4008, 0.5298, 0.5622, 0.5188], [0.4384, 0.3046, 0.7921, 0.3440, 0.3716, 0.2838, 0.4716, 0.5635]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6200, 0.3993, 0.8519, 0.4923, 0.3963, 0.4717, 0.6012, 0.5433], [0.6200, 0.4098, 0.8238, 0.2917, 0.4013, 0.2967, 0.6000, 0.5683], [0.6166, 0.4008, 0.8562, 0.5667, 0.4387, 0.4933, 0.5575, 0.5567], [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947], [0.6136, 0.4117, 0.8700, 0.5167, 0.4187, 0.5083, 0.5147, 0.5495], [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6147, 0.4112, 0.7987, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.20929075684398413 step: 56 running loss: 0.0037373349436425735 Train Steps: 56/90 Loss: 0.0037 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717], [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366], [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374], [0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5484, 0.3537, 0.8765, 0.5648, 0.4005, 0.4816, 0.5476, 0.5668], [0.5897, 0.3917, 0.8650, 0.5099, 0.4313, 0.4779, 0.5345, 0.5749], [0.5643, 0.3977, 0.8293, 0.5187, 0.4009, 0.4767, 0.5955, 0.5563], [0.6404, 0.4238, 0.8493, 0.3751, 0.3361, 0.3527, 0.5680, 0.5031], [0.5447, 0.3584, 0.8739, 0.5105, 0.4286, 0.4851, 0.5295, 0.5788], [0.5144, 0.3560, 0.8684, 0.3694, 0.3730, 0.3466, 0.5790, 0.5290], [0.4598, 0.3012, 0.8132, 0.3359, 0.4055, 0.2180, 0.5332, 0.5154], [0.4954, 0.3271, 0.8524, 0.3360, 0.3561, 0.3671, 0.6126, 0.5512]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6273, 0.4143, 0.8750, 0.5700, 0.3988, 0.4717, 0.6012, 0.5467], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5412, 0.5717], [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366], [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374], [0.6143, 0.4040, 0.8238, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.2111712924670428 step: 57 running loss: 0.003704759516965663 Train Steps: 57/90 Loss: 0.0037 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.6090, 0.4010, 0.7838, 0.3483, 0.3538, 0.3783, 0.5462, 0.5077], [0.6260, 0.4120, 0.8013, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700], [0.6357, 0.4097, 0.9038, 0.3883, 0.4213, 0.2950, 0.6686, 0.5390], [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5698, 0.3906, 0.7905, 0.3307, 0.4393, 0.2544, 0.5648, 0.5202], [0.5052, 0.3361, 0.8728, 0.5273, 0.3377, 0.4399, 0.5142, 0.5462], [0.5575, 0.3894, 0.8096, 0.3553, 0.3398, 0.4197, 0.5326, 0.4992], [0.5576, 0.3766, 0.8028, 0.2573, 0.4620, 0.2038, 0.6238, 0.4937], [0.5404, 0.3751, 0.8065, 0.2962, 0.4041, 0.3262, 0.5811, 0.5038], [0.4746, 0.3273, 0.7100, 0.2725, 0.4144, 0.2445, 0.4917, 0.5355], [0.5059, 0.3507, 0.8818, 0.4093, 0.3965, 0.3580, 0.6559, 0.5396], [0.5314, 0.3847, 0.8582, 0.5777, 0.3785, 0.6152, 0.6551, 0.5945]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6229, 0.4107, 0.8138, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6151, 0.4125, 0.8737, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.6090, 0.4010, 0.7837, 0.3483, 0.3537, 0.3783, 0.5462, 0.5077], [0.6259, 0.4120, 0.8012, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.0000, 0.0000, 0.6413, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700], [0.6357, 0.4097, 0.9038, 0.3883, 0.4212, 0.2950, 0.6686, 0.5390], [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795]]], device='cuda:0') loss_train_step before backward: tensor(0.0075, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0075, device='cuda:0', grad_fn=) loss_train: 0.21865464956499636 step: 58 running loss: 0.003769907751120627 Train Steps: 58/90 Loss: 0.0038 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6218, 0.4098, 0.7238, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350], [0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980], [0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843], [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567], [0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233], [0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485], [0.6131, 0.4064, 0.8638, 0.5200, 0.4788, 0.4783, 0.5258, 0.5867], [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5859, 0.4131, 0.7635, 0.2248, 0.4391, 0.2565, 0.6437, 0.5214], [0.5467, 0.3847, 0.7767, 0.2812, 0.3527, 0.3002, 0.5714, 0.4764], [0.5477, 0.3962, 0.9153, 0.4810, 0.3778, 0.5369, 0.6626, 0.4993], [0.5260, 0.3667, 0.8673, 0.4650, 0.3777, 0.4376, 0.5970, 0.5427], [0.4763, 0.3250, 0.8709, 0.5843, 0.4274, 0.4781, 0.5866, 0.5502], [0.5177, 0.3731, 0.7484, 0.2848, 0.3712, 0.3071, 0.5603, 0.5259], [0.6167, 0.4277, 0.8827, 0.5338, 0.4489, 0.4727, 0.6133, 0.5438], [0.4654, 0.3342, 0.7150, 0.2396, 0.3987, 0.2010, 0.5648, 0.5176]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6218, 0.4098, 0.7237, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350], [0.6079, 0.3964, 0.7420, 0.2958, 0.3562, 0.2917, 0.5351, 0.4980], [0.6264, 0.4049, 0.8988, 0.4633, 0.3812, 0.4983, 0.6326, 0.4843], [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567], [0.6222, 0.3937, 0.8350, 0.5617, 0.4137, 0.4600, 0.5800, 0.5233], [0.6123, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485], [0.6132, 0.4063, 0.8637, 0.5200, 0.4787, 0.4783, 0.5258, 0.5867], [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.2205590239027515 step: 59 running loss: 0.0037382885407246016 Train Steps: 59/90 Loss: 0.0037 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082], [0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467], [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433], [0.6125, 0.3974, 0.7725, 0.2517, 0.3538, 0.3317, 0.5887, 0.5500], [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5837, 0.5583], [0.6131, 0.4064, 0.8638, 0.5200, 0.4788, 0.4783, 0.5258, 0.5867]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5923, 0.3933, 0.8322, 0.5039, 0.4399, 0.5371, 0.6048, 0.4884], [0.5682, 0.3870, 0.8590, 0.5505, 0.3958, 0.4336, 0.6032, 0.5265], [0.6281, 0.4465, 0.7253, 0.2506, 0.4124, 0.2161, 0.6342, 0.5426], [0.5458, 0.3981, 0.7589, 0.2239, 0.3483, 0.3101, 0.6219, 0.4943], [0.5405, 0.3728, 0.7586, 0.2376, 0.3597, 0.2994, 0.5904, 0.5146], [0.5155, 0.3839, 0.8459, 0.3394, 0.3507, 0.4217, 0.7081, 0.4893], [0.5667, 0.4076, 0.8640, 0.4323, 0.4619, 0.4661, 0.6003, 0.5212], [0.6148, 0.4273, 0.8491, 0.5168, 0.4372, 0.4141, 0.5909, 0.5302]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082], [0.6273, 0.4143, 0.8750, 0.5700, 0.3988, 0.4717, 0.6012, 0.5467], [0.6193, 0.4078, 0.7287, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6175, 0.4091, 0.7862, 0.2800, 0.3638, 0.3583, 0.6187, 0.5433], [0.6125, 0.3974, 0.7725, 0.2517, 0.3537, 0.3317, 0.5888, 0.5500], [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5838, 0.5583], [0.6132, 0.4063, 0.8637, 0.5200, 0.4787, 0.4783, 0.5258, 0.5867]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.22198802302591503 step: 60 running loss: 0.0036998003837652505 Train Steps: 60/90 Loss: 0.0037 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.7725, 0.2611, 0.3675, 0.2733, 0.5413, 0.5167], [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170], [0.6171, 0.4127, 0.8900, 0.4800, 0.4325, 0.5783, 0.5769, 0.5090], [0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233], [0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320], [0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.1789, 0.1370, 0.7120, 0.2195, 0.3751, 0.2633, 0.5566, 0.5312], [0.5631, 0.3876, 0.8010, 0.1853, 0.4524, 0.1427, 0.6686, 0.5063], [0.6861, 0.4646, 0.8735, 0.4319, 0.4315, 0.5165, 0.5684, 0.5173], [0.5888, 0.4002, 0.8069, 0.5105, 0.4057, 0.4213, 0.5696, 0.5447], [0.7186, 0.4990, 0.8426, 0.4124, 0.3694, 0.4953, 0.7160, 0.5340], [0.6200, 0.4549, 0.8096, 0.5021, 0.4174, 0.4357, 0.6127, 0.5518], [0.6523, 0.4633, 0.8407, 0.4444, 0.4231, 0.4848, 0.6290, 0.5298], [0.6444, 0.4142, 0.7149, 0.1780, 0.3949, 0.1900, 0.5875, 0.5107]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.7725, 0.2611, 0.3675, 0.2733, 0.5412, 0.5167], [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170], [0.6171, 0.4127, 0.8900, 0.4800, 0.4325, 0.5783, 0.5769, 0.5090], [0.6222, 0.3937, 0.8350, 0.5617, 0.4137, 0.4600, 0.5800, 0.5233], [0.6250, 0.3993, 0.9137, 0.4333, 0.3762, 0.5217, 0.6995, 0.5320], [0.6229, 0.4066, 0.8512, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6199, 0.4059, 0.8700, 0.4900, 0.4162, 0.5000, 0.6162, 0.5467], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5788, 0.5117]]], device='cuda:0') loss_train_step before backward: tensor(0.0023, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0023, device='cuda:0', grad_fn=) loss_train: 0.22428614646196365 step: 61 running loss: 0.0036768220731469453 Train Steps: 61/90 Loss: 0.0037 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767], [0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6284, 0.4127, 0.8538, 0.5867, 0.4363, 0.5083, 0.6038, 0.5433], [ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567], [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917], [0.6222, 0.3957, 0.8838, 0.5017, 0.3937, 0.4600, 0.5900, 0.5017]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5691, 0.3808, 0.7564, 0.1777, 0.4299, 0.1889, 0.6466, 0.5016], [0.6328, 0.4316, 0.8196, 0.4589, 0.4082, 0.4745, 0.5711, 0.5704], [0.5339, 0.3510, 0.8157, 0.1961, 0.4921, 0.2346, 0.7271, 0.5324], [0.7159, 0.4650, 0.8122, 0.4496, 0.4231, 0.5123, 0.6721, 0.5393], [0.6640, 0.4476, 0.8020, 0.5129, 0.4045, 0.4800, 0.5805, 0.5495], [0.2912, 0.1803, 0.7068, 0.2265, 0.4071, 0.2261, 0.5405, 0.5527], [0.6205, 0.4289, 0.7558, 0.3432, 0.3518, 0.3197, 0.5453, 0.5625], [0.6474, 0.4268, 0.8590, 0.4616, 0.3832, 0.4418, 0.6134, 0.5029]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6163, 0.4001, 0.8788, 0.5033, 0.4013, 0.4633, 0.5337, 0.5767], [0.6293, 0.4097, 0.8800, 0.2517, 0.5263, 0.2600, 0.7430, 0.5378], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6284, 0.4127, 0.8537, 0.5867, 0.4363, 0.5083, 0.6037, 0.5433], [0.0000, 0.0000, 0.7425, 0.2117, 0.3938, 0.2433, 0.5437, 0.5567], [0.6179, 0.4118, 0.7278, 0.4237, 0.3587, 0.3400, 0.5675, 0.5917], [0.6222, 0.3957, 0.8838, 0.5017, 0.3938, 0.4600, 0.5900, 0.5017]]], device='cuda:0') loss_train_step before backward: tensor(0.0032, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0032, device='cuda:0', grad_fn=) loss_train: 0.22752876160666347 step: 62 running loss: 0.003669818735591346 Train Steps: 62/90 Loss: 0.0037 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822], [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5837, 0.5317], [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550], [0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5988, 0.5667], [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6034, 0.3830, 0.8470, 0.4437, 0.3738, 0.3747, 0.5338, 0.5797], [0.5444, 0.3546, 0.8087, 0.3473, 0.3960, 0.3135, 0.6162, 0.5542], [0.5521, 0.3261, 0.8409, 0.3032, 0.4649, 0.2440, 0.6543, 0.5259], [0.5926, 0.3772, 0.6626, 0.2557, 0.4032, 0.2445, 0.5561, 0.5455], [0.5108, 0.3200, 0.7701, 0.3098, 0.4265, 0.2208, 0.5590, 0.5159], [0.6456, 0.4156, 0.7671, 0.2335, 0.4370, 0.2696, 0.6316, 0.5587], [0.6134, 0.4167, 0.8674, 0.5409, 0.4430, 0.5818, 0.7387, 0.5563], [0.6006, 0.3781, 0.6835, 0.2165, 0.4343, 0.2681, 0.5758, 0.5471]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6182, 0.4058, 0.8737, 0.4350, 0.3562, 0.3400, 0.5290, 0.5822], [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5838, 0.5317], [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5362, 0.5550], [0.6143, 0.4040, 0.8238, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5987, 0.5667], [0.6339, 0.4123, 0.8637, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6108, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.22869582765270025 step: 63 running loss: 0.0036300925024238136 Train Steps: 63/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631], [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563], [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621], [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467], [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750], [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6140, 0.3829, 0.8532, 0.2711, 0.4873, 0.2148, 0.6442, 0.5450], [0.6847, 0.4150, 0.8289, 0.3931, 0.3653, 0.4806, 0.6040, 0.5475], [0.7392, 0.4679, 0.8294, 0.5288, 0.4050, 0.5031, 0.6651, 0.5992], [0.4096, 0.2648, 0.7645, 0.3139, 0.3828, 0.2889, 0.5430, 0.5730], [0.1937, 0.0940, 0.8403, 0.2669, 0.5400, 0.2014, 0.6737, 0.5628], [0.7099, 0.4543, 0.8390, 0.3263, 0.3754, 0.4002, 0.5924, 0.5534], [0.7283, 0.4502, 0.8550, 0.4972, 0.4617, 0.4719, 0.5788, 0.5715], [0.7993, 0.4937, 0.9025, 0.4761, 0.4007, 0.3892, 0.6428, 0.5437]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6275, 0.4111, 0.8462, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631], [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.0000, 0.0000, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563], [0.0000, 0.0000, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621], [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467], [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750], [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019]]], device='cuda:0') loss_train_step before backward: tensor(0.0062, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0062, device='cuda:0', grad_fn=) loss_train: 0.23489011928904802 step: 64 running loss: 0.0036701581138913753 Train Steps: 64/90 Loss: 0.0037 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.9088, 0.3783, 0.4562, 0.2617, 0.6741, 0.5575], [0.6263, 0.4030, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771], [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896], [0.6117, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103], [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117], [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.2212, 0.1345, 0.9286, 0.3959, 0.4843, 0.2741, 0.6916, 0.6015], [0.6566, 0.4074, 0.9196, 0.4697, 0.3683, 0.5070, 0.6453, 0.4970], [0.5537, 0.3543, 0.8277, 0.3884, 0.3790, 0.2970, 0.5461, 0.5881], [0.5645, 0.3503, 0.6897, 0.2247, 0.3856, 0.2232, 0.5269, 0.5686], [0.6891, 0.4353, 0.8483, 0.5375, 0.4398, 0.4830, 0.5719, 0.5832], [0.6096, 0.3528, 0.7507, 0.2245, 0.4239, 0.2051, 0.5894, 0.5566], [0.6337, 0.3920, 0.8727, 0.5009, 0.4832, 0.4534, 0.5873, 0.5605], [0.6439, 0.4225, 0.8797, 0.4280, 0.4242, 0.4819, 0.6083, 0.5413]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.9087, 0.3783, 0.4563, 0.2617, 0.6741, 0.5575], [0.6263, 0.4029, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771], [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896], [0.6116, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103], [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5788, 0.5117], [0.6102, 0.4005, 0.8687, 0.5100, 0.4812, 0.5400, 0.5404, 0.5064], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533]]], device='cuda:0') loss_train_step before backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train: 0.23684400937054306 step: 65 running loss: 0.003643753990316047 Train Steps: 65/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6332, 0.4118, 0.9238, 0.4267, 0.4012, 0.4733, 0.7525, 0.5436], [0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773], [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [ nan, nan, 0.7225, 0.2167, 0.3987, 0.2283, 0.5427, 0.5181], [0.6226, 0.4103, 0.8575, 0.3450, 0.4388, 0.2067, 0.5787, 0.5383], [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183], [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133], [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7355, 0.4587, 0.9337, 0.4735, 0.4093, 0.4814, 0.7177, 0.5522], [0.5557, 0.3274, 0.7331, 0.2659, 0.4116, 0.2238, 0.4978, 0.5754], [0.7013, 0.4191, 0.9234, 0.5755, 0.3711, 0.4749, 0.5756, 0.5564], [0.0530, 0.0123, 0.7388, 0.2372, 0.4230, 0.2440, 0.5240, 0.5477], [0.5336, 0.3270, 0.8849, 0.3829, 0.4673, 0.2350, 0.5589, 0.5687], [0.6730, 0.3989, 0.9308, 0.5037, 0.4046, 0.6039, 0.6622, 0.5154], [0.6858, 0.4156, 0.8866, 0.5624, 0.4506, 0.5007, 0.5818, 0.5358], [0.5166, 0.3204, 0.9286, 0.3357, 0.4373, 0.2783, 0.6617, 0.5548]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6332, 0.4118, 0.9237, 0.4267, 0.4013, 0.4733, 0.7525, 0.5436], [0.6147, 0.4026, 0.6600, 0.2467, 0.4087, 0.2150, 0.5489, 0.5773], [0.6222, 0.4171, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.0000, 0.0000, 0.7225, 0.2167, 0.3988, 0.2283, 0.5427, 0.5181], [0.6226, 0.4103, 0.8575, 0.3450, 0.4387, 0.2067, 0.5788, 0.5383], [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183], [0.6202, 0.4053, 0.8637, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133], [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609]]], device='cuda:0') loss_train_step before backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train: 0.2385742465266958 step: 66 running loss: 0.003614761311010542 Train Steps: 66/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297], [ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633], [0.6034, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084], [ nan, nan, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901], [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083], [0.6097, 0.4024, 0.8488, 0.3717, 0.3875, 0.5517, 0.5836, 0.5591]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6671, 0.4208, 0.9785, 0.5062, 0.4243, 0.3026, 0.6318, 0.5373], [0.1790, 0.0846, 0.8194, 0.3341, 0.4027, 0.3083, 0.5173, 0.5482], [0.6645, 0.4287, 0.7940, 0.3145, 0.3603, 0.3541, 0.5978, 0.5310], [0.2190, 0.1470, 0.7626, 0.2580, 0.4547, 0.2251, 0.5737, 0.5546], [0.5184, 0.3228, 0.7564, 0.3006, 0.4239, 0.2514, 0.5888, 0.5536], [0.5250, 0.3274, 0.8387, 0.3473, 0.3684, 0.2694, 0.5364, 0.5339], [0.6058, 0.3866, 0.9575, 0.5627, 0.4092, 0.5112, 0.6149, 0.5469], [0.6870, 0.4104, 0.9199, 0.4403, 0.4319, 0.5612, 0.6627, 0.5029]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6364, 0.4165, 0.9087, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297], [0.0000, 0.0000, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633], [0.6033, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084], [0.0000, 0.0000, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901], [0.6186, 0.4060, 0.8750, 0.5050, 0.3537, 0.4367, 0.5813, 0.6083], [0.6097, 0.4024, 0.8487, 0.3717, 0.3875, 0.5517, 0.5836, 0.5591]]], device='cuda:0') loss_train_step before backward: tensor(0.0042, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0042, device='cuda:0', grad_fn=) loss_train: 0.24276340811047703 step: 67 running loss: 0.003623334449410105 Train Steps: 67/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250], [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600], [0.6250, 0.4146, 0.8838, 0.3933, 0.3588, 0.4283, 0.6162, 0.5367], [0.6193, 0.4108, 0.7438, 0.2700, 0.3650, 0.3683, 0.6238, 0.5717], [ nan, nan, 0.6512, 0.1717, 0.4100, 0.1983, 0.5253, 0.5240], [0.6286, 0.4055, 0.9000, 0.4717, 0.3763, 0.4683, 0.7018, 0.5494], [0.6182, 0.3930, 0.8841, 0.3892, 0.3556, 0.4967, 0.6222, 0.5279], [0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5988, 0.5667]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5577, 0.3404, 0.9322, 0.5424, 0.4028, 0.5637, 0.5783, 0.5157], [0.5639, 0.3888, 0.7683, 0.2950, 0.3763, 0.3433, 0.5770, 0.5662], [0.5509, 0.3533, 0.9084, 0.4569, 0.3602, 0.3963, 0.5602, 0.4995], [0.6154, 0.3938, 0.8092, 0.3178, 0.3638, 0.3636, 0.5955, 0.5416], [0.0305, 0.0186, 0.7465, 0.2533, 0.4329, 0.2216, 0.5314, 0.5542], [0.5792, 0.3591, 0.9353, 0.5333, 0.3668, 0.4853, 0.6384, 0.5251], [0.5785, 0.3509, 0.9181, 0.4518, 0.3553, 0.4828, 0.6251, 0.5212], [0.5584, 0.3610, 0.8354, 0.2901, 0.4220, 0.2634, 0.5679, 0.5586]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250], [0.6132, 0.4066, 0.7259, 0.2402, 0.3587, 0.3300, 0.6000, 0.5600], [0.6250, 0.4146, 0.8838, 0.3933, 0.3587, 0.4283, 0.6162, 0.5367], [0.6193, 0.4108, 0.7437, 0.2700, 0.3650, 0.3683, 0.6237, 0.5717], [0.0000, 0.0000, 0.6513, 0.1717, 0.4100, 0.1983, 0.5253, 0.5240], [0.6286, 0.4055, 0.9000, 0.4717, 0.3762, 0.4683, 0.7018, 0.5494], [0.6182, 0.3930, 0.8841, 0.3892, 0.3556, 0.4967, 0.6222, 0.5279], [0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5987, 0.5667]]], device='cuda:0') loss_train_step before backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train: 0.24445695418398827 step: 68 running loss: 0.003594955208588063 Train Steps: 68/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598], [ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383], [0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235], [0.6250, 0.4013, 0.8525, 0.5417, 0.4037, 0.5117, 0.6325, 0.5017], [0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200], [0.6208, 0.4082, 0.8538, 0.3067, 0.3588, 0.3717, 0.6112, 0.5517], [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.5516, 0.3771, 0.8252, 0.4969, 0.4173, 0.4475, 0.5955, 0.5628], [-0.0498, -0.0271, 0.7035, 0.1818, 0.3952, 0.2227, 0.5223, 0.5519], [ 0.4824, 0.3117, 0.8709, 0.2582, 0.4267, 0.2320, 0.6016, 0.5198], [ 0.5551, 0.3592, 0.8599, 0.4213, 0.3784, 0.5226, 0.6311, 0.5210], [ 0.5545, 0.3603, 0.8491, 0.4894, 0.3806, 0.5112, 0.6097, 0.5152], [ 0.5212, 0.3685, 0.8626, 0.5007, 0.3574, 0.4597, 0.5946, 0.5317], [ 0.5741, 0.3873, 0.8403, 0.2797, 0.3233, 0.3746, 0.6037, 0.5376], [ 0.5667, 0.3745, 0.8423, 0.4916, 0.4029, 0.5023, 0.5392, 0.5470]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598], [0.0000, 0.0000, 0.7268, 0.2333, 0.4125, 0.1933, 0.5113, 0.5383], [0.6259, 0.4156, 0.8813, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235], [0.6250, 0.4013, 0.8525, 0.5417, 0.4038, 0.5117, 0.6325, 0.5017], [0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200], [0.6208, 0.4082, 0.8537, 0.3067, 0.3587, 0.3717, 0.6112, 0.5517], [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461]]], device='cuda:0') loss_train_step before backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train: 0.24625502328854054 step: 69 running loss: 0.003568913380993341 Train Steps: 69/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6260, 0.4120, 0.8013, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895], [0.6357, 0.4118, 0.8400, 0.2500, 0.5413, 0.1633, 0.6725, 0.5586], [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882], [0.6188, 0.4099, 0.7400, 0.2433, 0.3962, 0.2750, 0.6162, 0.5467], [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6362, 0.5000], [0.6293, 0.3982, 0.8700, 0.5300, 0.3763, 0.4717, 0.7050, 0.5297]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4102, 0.2683, 0.7072, 0.2315, 0.3801, 0.3003, 0.5461, 0.5529], [0.4654, 0.3223, 0.7671, 0.2061, 0.4613, 0.2163, 0.6026, 0.5000], [0.3718, 0.2384, 0.8290, 0.2430, 0.4869, 0.2548, 0.6139, 0.5142], [0.4628, 0.3322, 0.7671, 0.2106, 0.4033, 0.2363, 0.5672, 0.5524], [0.4736, 0.3394, 0.8669, 0.4077, 0.3232, 0.4130, 0.5212, 0.5193], [0.5163, 0.3695, 0.7215, 0.2309, 0.3828, 0.3317, 0.5854, 0.5381], [0.5423, 0.3440, 0.8990, 0.5151, 0.3579, 0.5819, 0.5909, 0.5004], [0.5993, 0.3707, 0.8675, 0.5264, 0.3271, 0.5180, 0.6483, 0.5365]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6193, 0.4050, 0.7312, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6259, 0.4120, 0.8012, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895], [0.6357, 0.4118, 0.8400, 0.2500, 0.5412, 0.1633, 0.6725, 0.5586], [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882], [0.6188, 0.4099, 0.7400, 0.2433, 0.3963, 0.2750, 0.6162, 0.5467], [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6363, 0.5000], [0.6293, 0.3982, 0.8700, 0.5300, 0.3762, 0.4717, 0.7050, 0.5297]]], device='cuda:0') loss_train_step before backward: tensor(0.0054, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0054, device='cuda:0', grad_fn=) loss_train: 0.25164759380277246 step: 70 running loss: 0.003594965625753892 Train Steps: 70/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6118, 0.4052, 0.8463, 0.3917, 0.3538, 0.3450, 0.5053, 0.5593], [0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995], [0.6353, 0.4128, 0.8488, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320], [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411], [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633], [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4371, 0.3352, 0.8117, 0.3662, 0.3465, 0.3646, 0.4915, 0.5341], [0.4829, 0.3549, 0.7093, 0.1705, 0.4160, 0.2348, 0.6135, 0.5019], [0.3030, 0.1993, 0.7985, 0.2163, 0.5106, 0.2201, 0.6479, 0.5032], [0.4934, 0.3381, 0.6827, 0.2094, 0.3904, 0.2574, 0.5596, 0.5345], [0.7015, 0.4689, 0.8686, 0.4039, 0.3264, 0.4478, 0.6475, 0.4829], [0.4930, 0.3494, 0.7581, 0.2368, 0.4055, 0.2341, 0.5452, 0.5235], [0.5343, 0.3676, 0.8534, 0.4128, 0.4094, 0.5396, 0.5989, 0.5252], [0.5998, 0.4077, 0.8622, 0.5105, 0.3764, 0.4379, 0.5306, 0.5128]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6118, 0.4052, 0.8462, 0.3917, 0.3537, 0.3450, 0.5053, 0.5593], [0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995], [0.6353, 0.4128, 0.8487, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6193, 0.4050, 0.7312, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320], [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411], [0.6189, 0.4049, 0.8888, 0.4417, 0.4212, 0.5200, 0.5987, 0.5633], [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881]]], device='cuda:0') loss_train_step before backward: tensor(0.0053, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0053, device='cuda:0', grad_fn=) loss_train: 0.2569181363796815 step: 71 running loss: 0.003618565301122275 Train Steps: 71/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683], [0.6147, 0.4112, 0.7988, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550], [0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019], [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6063, 0.5617], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7440, 0.5025, 0.8211, 0.2773, 0.4304, 0.2292, 0.6325, 0.4862], [0.5843, 0.4262, 0.7705, 0.4431, 0.4198, 0.4516, 0.6558, 0.5235], [0.5360, 0.3710, 0.8328, 0.4776, 0.3894, 0.3353, 0.5496, 0.5669], [0.3983, 0.3095, 0.7359, 0.2769, 0.4089, 0.2535, 0.5024, 0.5584], [0.6206, 0.4284, 0.8121, 0.3538, 0.3933, 0.5180, 0.5663, 0.5119], [0.6116, 0.4225, 0.8465, 0.4352, 0.3794, 0.4521, 0.5980, 0.5525], [0.6379, 0.4404, 0.8386, 0.3539, 0.3735, 0.4274, 0.6420, 0.5186], [0.5376, 0.3719, 0.8287, 0.2723, 0.4434, 0.2788, 0.6565, 0.5224]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6277, 0.4036, 0.8687, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5412, 0.5683], [0.6147, 0.4112, 0.7987, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550], [0.6161, 0.4099, 0.8737, 0.4383, 0.3787, 0.5483, 0.5605, 0.5019], [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6062, 0.5617], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355]]], device='cuda:0') loss_train_step before backward: tensor(0.0032, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0032, device='cuda:0', grad_fn=) loss_train: 0.26006837061140686 step: 72 running loss: 0.0036120607029362065 Train Steps: 72/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400], [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690], [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6063, 0.5617]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5909, 0.4132, 0.8230, 0.5111, 0.4550, 0.3941, 0.5952, 0.5519], [0.6030, 0.4113, 0.8529, 0.4071, 0.4819, 0.4526, 0.6298, 0.5334], [0.6145, 0.4201, 0.8440, 0.3966, 0.4589, 0.4681, 0.6126, 0.5336], [0.6355, 0.4346, 0.7490, 0.2479, 0.3614, 0.2378, 0.5285, 0.5285], [0.6995, 0.5085, 0.8196, 0.2540, 0.4860, 0.1383, 0.6536, 0.5236], [0.6006, 0.4164, 0.8114, 0.3042, 0.3536, 0.4077, 0.6028, 0.5390], [0.6246, 0.4432, 0.8034, 0.4662, 0.4182, 0.4467, 0.6713, 0.5607], [0.6711, 0.4687, 0.8610, 0.4468, 0.3593, 0.3986, 0.5906, 0.5496]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6229, 0.4066, 0.8512, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6251, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6087, 0.5183], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6168, 0.4029, 0.8523, 0.3417, 0.3587, 0.5000, 0.6125, 0.5400], [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690], [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6062, 0.5617]]], device='cuda:0') loss_train_step before backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train: 0.2622584825148806 step: 73 running loss: 0.0035925819522586383 Train Steps: 73/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6262, 0.5367], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6095, 0.3970, 0.8688, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103], [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667], [0.6289, 0.4024, 0.9088, 0.4567, 0.3937, 0.5633, 0.7058, 0.5609], [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6593, 0.4456, 0.8539, 0.4580, 0.4526, 0.5123, 0.6356, 0.5641], [0.8243, 0.5577, 0.7724, 0.3118, 0.3512, 0.2677, 0.6180, 0.5296], [0.6100, 0.3944, 0.8633, 0.4584, 0.3923, 0.4094, 0.6911, 0.5411], [0.7037, 0.4569, 0.8566, 0.4849, 0.4779, 0.3771, 0.5572, 0.5403], [0.6089, 0.4352, 0.8558, 0.4276, 0.4084, 0.3731, 0.5672, 0.5224], [0.7343, 0.5111, 0.8361, 0.3372, 0.3965, 0.3233, 0.6683, 0.5543], [0.6134, 0.4223, 0.8518, 0.4732, 0.4324, 0.4890, 0.7007, 0.5811], [0.5786, 0.4307, 0.8439, 0.4056, 0.3870, 0.3162, 0.5765, 0.5565]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6263, 0.5367], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6095, 0.3970, 0.8687, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103], [0.6257, 0.4167, 0.8775, 0.3433, 0.3562, 0.4133, 0.6200, 0.5667], [0.6289, 0.4024, 0.9087, 0.4567, 0.3938, 0.5633, 0.7058, 0.5609], [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350]]], device='cuda:0') loss_train_step before backward: tensor(0.0032, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0032, device='cuda:0', grad_fn=) loss_train: 0.26550832379143685 step: 74 running loss: 0.0035879503215059034 Train Steps: 74/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6124, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6250, 0.4106, 0.8700, 0.3717, 0.3588, 0.4967, 0.6038, 0.5167]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7221, 0.4680, 0.8072, 0.5104, 0.3845, 0.4499, 0.5519, 0.5637], [0.6449, 0.4268, 0.8583, 0.5027, 0.4076, 0.5221, 0.6426, 0.5507], [0.7298, 0.5031, 0.7766, 0.2269, 0.4749, 0.1387, 0.6136, 0.5198], [0.4523, 0.3037, 0.8340, 0.2722, 0.5095, 0.1814, 0.6730, 0.5421], [0.7061, 0.4926, 0.8780, 0.4993, 0.3746, 0.4107, 0.6573, 0.5228], [0.7205, 0.4687, 0.8657, 0.5145, 0.4142, 0.5632, 0.6663, 0.5195], [0.7471, 0.4946, 0.8512, 0.4310, 0.4183, 0.2610, 0.5567, 0.5546], [0.6749, 0.4517, 0.8457, 0.3970, 0.3774, 0.4914, 0.6339, 0.5649]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6123, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.0000, 0.0000, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6226, 0.4098, 0.8913, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6250, 0.4105, 0.8700, 0.3717, 0.3587, 0.4967, 0.6037, 0.5167]]], device='cuda:0') loss_train_step before backward: tensor(0.0066, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0066, device='cuda:0', grad_fn=) loss_train: 0.27209125307854265 step: 75 running loss: 0.003627883374380569 Train Steps: 75/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050], [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332], [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283], [0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148], [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133], [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6798, 0.4367, 0.8905, 0.5339, 0.4184, 0.5330, 0.5787, 0.5165], [0.6814, 0.4195, 0.9055, 0.4748, 0.3659, 0.4066, 0.6685, 0.5359], [0.6670, 0.4253, 0.8463, 0.5316, 0.3967, 0.4579, 0.5615, 0.5652], [0.6028, 0.3785, 0.8851, 0.3418, 0.4280, 0.3274, 0.6745, 0.5367], [0.6783, 0.4034, 0.8593, 0.4999, 0.4756, 0.4750, 0.5447, 0.5214], [0.6905, 0.4342, 0.9075, 0.4974, 0.3788, 0.4901, 0.6273, 0.5221], [0.7945, 0.5029, 0.8942, 0.5036, 0.3687, 0.4840, 0.6507, 0.5171], [0.6484, 0.4060, 0.8931, 0.4606, 0.3984, 0.4084, 0.6942, 0.5413]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6097, 0.3988, 0.8650, 0.5250, 0.4212, 0.5200, 0.5675, 0.5050], [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332], [0.6212, 0.4159, 0.8675, 0.5783, 0.4087, 0.4317, 0.5612, 0.5917], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6084, 0.4008, 0.8587, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283], [0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148], [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133], [0.6346, 0.4144, 0.9087, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.27344133728183806 step: 76 running loss: 0.003597912332655764 Train Steps: 76/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6198, 0.4101, 0.8838, 0.5283, 0.3763, 0.5267, 0.5913, 0.5567], [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507], [0.6246, 0.4008, 0.8757, 0.5088, 0.4101, 0.5392, 0.6644, 0.5133], [0.6250, 0.4103, 0.8950, 0.4400, 0.3912, 0.5650, 0.6050, 0.5133], [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301], [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4302, 0.2588, 0.7816, 0.3117, 0.4392, 0.2675, 0.5540, 0.5430], [0.6821, 0.4257, 0.9284, 0.6394, 0.3763, 0.5872, 0.6291, 0.5547], [0.7801, 0.5038, 0.8543, 0.3555, 0.4846, 0.2312, 0.6044, 0.5342], [0.7169, 0.4645, 0.7590, 0.2761, 0.4219, 0.2503, 0.6018, 0.5394], [0.6197, 0.3730, 0.9462, 0.5948, 0.3995, 0.5777, 0.6588, 0.5362], [0.7014, 0.4260, 0.9745, 0.5130, 0.4179, 0.6349, 0.6801, 0.5268], [0.6904, 0.4121, 0.9245, 0.5656, 0.4052, 0.5342, 0.6547, 0.5416], [0.6215, 0.3756, 0.9185, 0.4246, 0.3443, 0.5273, 0.6487, 0.5402]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6198, 0.4101, 0.8838, 0.5283, 0.3762, 0.5267, 0.5913, 0.5567], [0.6229, 0.4107, 0.8138, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6072, 0.4029, 0.7038, 0.2150, 0.3913, 0.2267, 0.5516, 0.5507], [0.6246, 0.4008, 0.8757, 0.5088, 0.4101, 0.5392, 0.6644, 0.5133], [0.6250, 0.4103, 0.8950, 0.4400, 0.3913, 0.5650, 0.6050, 0.5133], [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301], [0.6168, 0.4029, 0.8523, 0.3417, 0.3587, 0.5000, 0.6125, 0.5400]]], device='cuda:0') loss_train_step before backward: tensor(0.0037, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0037, device='cuda:0', grad_fn=) loss_train: 0.27711278037168086 step: 77 running loss: 0.0035988672775542968 Train Steps: 77/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167], [ nan, nan, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583], [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170], [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667], [0.6261, 0.3987, 0.8688, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309], [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317], [0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983], [0.6136, 0.4085, 0.6688, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6488, 0.4085, 0.9239, 0.4528, 0.3710, 0.4379, 0.6246, 0.5328], [0.1281, 0.0702, 0.7535, 0.2990, 0.4289, 0.2780, 0.5352, 0.5354], [0.7497, 0.4416, 0.8740, 0.3162, 0.4783, 0.2598, 0.6582, 0.5157], [0.7471, 0.4585, 0.8960, 0.4175, 0.3671, 0.4120, 0.5890, 0.5436], [0.6345, 0.3725, 0.9412, 0.5425, 0.4347, 0.6288, 0.6831, 0.5394], [0.7109, 0.4313, 0.9336, 0.5875, 0.3706, 0.4795, 0.5975, 0.5337], [0.6542, 0.3831, 0.9247, 0.5839, 0.4145, 0.5602, 0.6040, 0.5012], [0.6355, 0.3950, 0.7365, 0.2905, 0.3895, 0.3092, 0.5418, 0.5230]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6263, 0.5167], [0.0000, 0.0000, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583], [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170], [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5987, 0.5667], [0.6261, 0.3987, 0.8687, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309], [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317], [0.6219, 0.3934, 0.8687, 0.5267, 0.4313, 0.4967, 0.5987, 0.4983], [0.6136, 0.4085, 0.6687, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783]]], device='cuda:0') loss_train_step before backward: tensor(0.0027, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0027, device='cuda:0', grad_fn=) loss_train: 0.27983771497383714 step: 78 running loss: 0.0035876630124850916 Train Steps: 78/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123], [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841], [0.6317, 0.4038, 0.8287, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986], [0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517], [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6289, 0.4081, 0.8720, 0.3487, 0.3900, 0.3183, 0.6703, 0.5376], [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6002, 0.3573, 0.9323, 0.4695, 0.4067, 0.5875, 0.6304, 0.5184], [0.6629, 0.4128, 0.8068, 0.3113, 0.3855, 0.2987, 0.5554, 0.5189], [0.5464, 0.3306, 0.9350, 0.4519, 0.4138, 0.4141, 0.6858, 0.5519], [0.6337, 0.3700, 0.9142, 0.5837, 0.3704, 0.5209, 0.6383, 0.5275], [0.5967, 0.3291, 0.9264, 0.4726, 0.4439, 0.5952, 0.5902, 0.5238], [0.5697, 0.3110, 0.8690, 0.5939, 0.4222, 0.5291, 0.5780, 0.5865], [0.6790, 0.4061, 0.9367, 0.3756, 0.4064, 0.3228, 0.6563, 0.5449], [0.5955, 0.3229, 0.9189, 0.5381, 0.4464, 0.5383, 0.5872, 0.5075]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123], [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841], [0.6317, 0.4038, 0.8288, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986], [0.6122, 0.4006, 0.8850, 0.4217, 0.4087, 0.5517, 0.6062, 0.5517], [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6289, 0.4081, 0.8720, 0.3487, 0.3900, 0.3183, 0.6703, 0.5376], [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.28146227076649666 step: 79 running loss: 0.0035628135540062868 Train Steps: 79/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6346, 0.4165, 0.9138, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471], [0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843], [ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633], [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6261, 0.4066, 0.8325, 0.2150, 0.4763, 0.2667, 0.7002, 0.5633], [0.6109, 0.4015, 0.7668, 0.3639, 0.3513, 0.3667, 0.5200, 0.5641], [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5119, 0.3425, 0.9265, 0.4695, 0.3875, 0.4551, 0.6786, 0.5260], [0.6343, 0.3938, 0.9262, 0.5178, 0.3744, 0.5502, 0.6083, 0.4991], [0.1974, 0.1064, 0.7876, 0.3150, 0.3990, 0.3172, 0.5073, 0.5274], [0.6346, 0.4311, 0.8191, 0.3264, 0.4551, 0.3047, 0.5770, 0.5565], [0.6159, 0.3697, 0.8444, 0.6172, 0.3703, 0.5573, 0.5787, 0.4977], [0.7068, 0.4193, 0.8361, 0.2817, 0.4721, 0.2977, 0.6492, 0.5381], [0.6086, 0.3895, 0.8116, 0.4116, 0.3440, 0.3997, 0.5280, 0.5469], [0.5484, 0.3277, 0.9446, 0.4219, 0.4743, 0.3307, 0.6781, 0.5497]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6346, 0.4165, 0.9137, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471], [0.6264, 0.4049, 0.8988, 0.4633, 0.3812, 0.4983, 0.6326, 0.4843], [0.0000, 0.0000, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633], [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6261, 0.4066, 0.8325, 0.2150, 0.4762, 0.2667, 0.7002, 0.5633], [0.6109, 0.4015, 0.7668, 0.3639, 0.3512, 0.3667, 0.5200, 0.5641], [0.6271, 0.4040, 0.9137, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413]]], device='cuda:0') loss_train_step before backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train: 0.28381356899626553 step: 80 running loss: 0.0035476696124533192 Train Steps: 80/90 Loss: 0.0035 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6357, 0.4097, 0.9038, 0.3883, 0.4213, 0.2950, 0.6686, 0.5390], [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899], [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397], [0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567], [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717], [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240], [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700], [ nan, nan, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7318, 0.4761, 0.9288, 0.4338, 0.4220, 0.3790, 0.6864, 0.5237], [0.7237, 0.4574, 0.9326, 0.5243, 0.3879, 0.5031, 0.6863, 0.5338], [0.5232, 0.3285, 0.7327, 0.2817, 0.4121, 0.2591, 0.5487, 0.5335], [0.6395, 0.4100, 0.9086, 0.5295, 0.4753, 0.5576, 0.5786, 0.5316], [0.6670, 0.4423, 0.7967, 0.3256, 0.3888, 0.3411, 0.5705, 0.5651], [0.3493, 0.2167, 0.7714, 0.2595, 0.4272, 0.2022, 0.5383, 0.5256], [0.2071, 0.1384, 0.7526, 0.2718, 0.4370, 0.2432, 0.5629, 0.5550], [0.3592, 0.2399, 0.7592, 0.3142, 0.3981, 0.2982, 0.5345, 0.5534]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6357, 0.4097, 0.9038, 0.3883, 0.4212, 0.2950, 0.6686, 0.5390], [0.6346, 0.4144, 0.9087, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899], [0.6159, 0.4085, 0.6900, 0.2283, 0.4087, 0.1950, 0.5123, 0.5397], [0.6112, 0.4029, 0.8637, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567], [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717], [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240], [0.0000, 0.0000, 0.6413, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700], [0.0000, 0.0000, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933]]], device='cuda:0') loss_train_step before backward: tensor(0.0076, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0076, device='cuda:0', grad_fn=) loss_train: 0.29142689728178084 step: 81 running loss: 0.003597862929404702 Train Steps: 81/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917], [0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960], [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100], [0.6127, 0.4118, 0.8650, 0.5083, 0.4088, 0.5367, 0.5300, 0.5456], [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533], [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283], [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5037, 0.3305, 0.8141, 0.5500, 0.4043, 0.4282, 0.5797, 0.5804], [0.6664, 0.4200, 0.8563, 0.2748, 0.4429, 0.2194, 0.6277, 0.5476], [0.5060, 0.3348, 0.8612, 0.4524, 0.4320, 0.5289, 0.5627, 0.5324], [0.4562, 0.3147, 0.8126, 0.4894, 0.3906, 0.4595, 0.6114, 0.5158], [0.4270, 0.2897, 0.8398, 0.4804, 0.4076, 0.5038, 0.5437, 0.5549], [0.4705, 0.3189, 0.7083, 0.2346, 0.3568, 0.2952, 0.5956, 0.5650], [0.5339, 0.3557, 0.8859, 0.3443, 0.3659, 0.3642, 0.6454, 0.5390], [0.5492, 0.3757, 0.8003, 0.2359, 0.4695, 0.1793, 0.6644, 0.5451]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6248, 0.4185, 0.8500, 0.5767, 0.4462, 0.4550, 0.5612, 0.5917], [0.6259, 0.4156, 0.8813, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960], [0.6155, 0.4124, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6299, 0.4008, 0.8450, 0.5350, 0.4212, 0.5000, 0.6350, 0.5100], [0.6127, 0.4118, 0.8650, 0.5083, 0.4087, 0.5367, 0.5300, 0.5456], [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533], [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283], [0.6266, 0.4065, 0.8313, 0.2467, 0.4787, 0.1733, 0.6313, 0.5133]]], device='cuda:0') loss_train_step before backward: tensor(0.0033, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0033, device='cuda:0', grad_fn=) loss_train: 0.29475518641993403 step: 82 running loss: 0.003594575444145537 Train Steps: 82/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6198, 0.4115, 0.7762, 0.2717, 0.3713, 0.3200, 0.5837, 0.5683], [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6063, 0.5567], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750], [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567], [0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626], [0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748], [0.6060, 0.3924, 0.8450, 0.5717, 0.4200, 0.5217, 0.5253, 0.4752], [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5219, 0.3660, 0.7790, 0.2321, 0.3643, 0.2673, 0.5809, 0.5409], [0.5195, 0.3591, 0.8312, 0.3807, 0.4510, 0.4457, 0.5938, 0.5612], [0.5682, 0.3913, 0.6999, 0.2407, 0.3874, 0.2158, 0.6085, 0.5697], [0.1475, 0.1097, 0.6956, 0.2113, 0.4007, 0.1901, 0.5284, 0.5563], [0.5345, 0.3710, 0.8218, 0.4980, 0.3978, 0.4215, 0.6464, 0.4959], [0.5906, 0.4026, 0.8150, 0.4026, 0.3914, 0.3041, 0.6396, 0.5236], [0.4948, 0.3323, 0.8203, 0.5143, 0.4422, 0.4626, 0.5775, 0.5119], [0.6208, 0.4077, 0.7593, 0.2987, 0.4598, 0.2385, 0.5518, 0.5721]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6198, 0.4115, 0.7763, 0.2717, 0.3713, 0.3200, 0.5838, 0.5683], [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6062, 0.5567], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750], [0.0000, 0.0000, 0.7515, 0.2708, 0.3988, 0.2267, 0.5163, 0.5567], [0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626], [0.6336, 0.4191, 0.8938, 0.5167, 0.3938, 0.3517, 0.7343, 0.5748], [0.6060, 0.3924, 0.8450, 0.5717, 0.4200, 0.5217, 0.5253, 0.4752], [0.6299, 0.4303, 0.7962, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222]]], device='cuda:0') loss_train_step before backward: tensor(0.0032, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0032, device='cuda:0', grad_fn=) loss_train: 0.29796311957761645 step: 83 running loss: 0.0035899171033447765 Train Steps: 83/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6110, 0.3984, 0.8750, 0.4933, 0.4625, 0.4950, 0.5578, 0.5676], [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6152, 0.4131, 0.6863, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [0.6179, 0.4040, 0.7412, 0.1850, 0.3825, 0.2783, 0.5837, 0.5600], [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5181, 0.3539, 0.8390, 0.4595, 0.4247, 0.4804, 0.6175, 0.5306], [0.4836, 0.3487, 0.7682, 0.2725, 0.3916, 0.2523, 0.6011, 0.5581], [0.5561, 0.3936, 0.8207, 0.4423, 0.4917, 0.4533, 0.5457, 0.5479], [0.5201, 0.3584, 0.8710, 0.3262, 0.3958, 0.2486, 0.6272, 0.5657], [0.4927, 0.3735, 0.6764, 0.2544, 0.3768, 0.2560, 0.5766, 0.5698], [0.4877, 0.3685, 0.7300, 0.2531, 0.3999, 0.2156, 0.5919, 0.5682], [0.4900, 0.3573, 0.7261, 0.2373, 0.4414, 0.1238, 0.5944, 0.5538], [0.5658, 0.4001, 0.8245, 0.4372, 0.4641, 0.5247, 0.6171, 0.5283]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6110, 0.3984, 0.8750, 0.4933, 0.4625, 0.4950, 0.5578, 0.5676], [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6152, 0.4131, 0.6862, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [0.6179, 0.4040, 0.7412, 0.1850, 0.3825, 0.2783, 0.5838, 0.5600], [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150]]], device='cuda:0') loss_train_step before backward: tensor(0.0027, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0027, device='cuda:0', grad_fn=) loss_train: 0.3006193444598466 step: 84 running loss: 0.0035788017197600787 Train Steps: 84/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5613, 0.5900], [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200], [0.6275, 0.4050, 0.9038, 0.3767, 0.3838, 0.3533, 0.7074, 0.5575], [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167], [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917], [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320], [0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5401, 0.3604, 0.7687, 0.5018, 0.4050, 0.4576, 0.5831, 0.5146], [0.6179, 0.4281, 0.7783, 0.3962, 0.3832, 0.3099, 0.5524, 0.5581], [0.5014, 0.3646, 0.8290, 0.4376, 0.4013, 0.4111, 0.6210, 0.5631], [0.5474, 0.3766, 0.8633, 0.2975, 0.4234, 0.2974, 0.6899, 0.5606], [0.5735, 0.4356, 0.7765, 0.2658, 0.3967, 0.2545, 0.5893, 0.5601], [0.5691, 0.4091, 0.7424, 0.3332, 0.3733, 0.2847, 0.5574, 0.5884], [0.4941, 0.3615, 0.7728, 0.1914, 0.4640, 0.2244, 0.6177, 0.5739], [0.5280, 0.3710, 0.8275, 0.4217, 0.3933, 0.3405, 0.5366, 0.5362]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5612, 0.5900], [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200], [0.6275, 0.4050, 0.9038, 0.3767, 0.3837, 0.3533, 0.7074, 0.5575], [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167], [0.6179, 0.4118, 0.7278, 0.4237, 0.3587, 0.3400, 0.5675, 0.5917], [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320], [0.6127, 0.4084, 0.8737, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397]]], device='cuda:0') loss_train_step before backward: tensor(0.0027, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0027, device='cuda:0', grad_fn=) loss_train: 0.3032755733001977 step: 85 running loss: 0.0035679479211787966 Train Steps: 85/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378], [0.6138, 0.4101, 0.8800, 0.5083, 0.4637, 0.5950, 0.5587, 0.5077], [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915], [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407], [0.6115, 0.3998, 0.7063, 0.2383, 0.4037, 0.1950, 0.5320, 0.4993], [0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200], [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5689, 0.3898, 0.6745, 0.2629, 0.3627, 0.2696, 0.5545, 0.5796], [0.5396, 0.3793, 0.8791, 0.4103, 0.4065, 0.3312, 0.6782, 0.5538], [0.6021, 0.4020, 0.8272, 0.4906, 0.4783, 0.5328, 0.5690, 0.5541], [0.6105, 0.4080, 0.8580, 0.4126, 0.3853, 0.2962, 0.6221, 0.5206], [0.6513, 0.4556, 0.8229, 0.3616, 0.3921, 0.2947, 0.5291, 0.5429], [0.4928, 0.3361, 0.6731, 0.1926, 0.4042, 0.1822, 0.5665, 0.5280], [0.5830, 0.4160, 0.8283, 0.5111, 0.3814, 0.4505, 0.5676, 0.5232], [0.5854, 0.4016, 0.7740, 0.2581, 0.3595, 0.2813, 0.6416, 0.5413]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6307, 0.4029, 0.8988, 0.4817, 0.3938, 0.3500, 0.7311, 0.5378], [0.6138, 0.4101, 0.8800, 0.5083, 0.4638, 0.5950, 0.5587, 0.5077], [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915], [0.6133, 0.4094, 0.8495, 0.4028, 0.3587, 0.3200, 0.5003, 0.5407], [0.6115, 0.3998, 0.7063, 0.2383, 0.4038, 0.1950, 0.5320, 0.4993], [0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200], [0.6135, 0.3994, 0.7912, 0.3050, 0.3625, 0.3050, 0.5838, 0.5050]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.3045844287844375 step: 86 running loss: 0.0035416794044702033 Train Steps: 86/90 Loss: 0.0035 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6179, 0.4008, 0.8600, 0.4015, 0.3932, 0.2515, 0.5711, 0.5438], [0.6267, 0.4080, 0.8438, 0.2633, 0.4763, 0.1800, 0.6259, 0.5240], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986], [0.6173, 0.4114, 0.7325, 0.2500, 0.4213, 0.1917, 0.5338, 0.5700], [0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517], [ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933], [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6938, 0.4627, 0.7745, 0.3595, 0.3673, 0.2659, 0.5422, 0.5378], [0.7675, 0.5108, 0.8090, 0.2835, 0.4165, 0.2215, 0.5889, 0.5316], [0.7989, 0.5596, 0.8286, 0.5296, 0.3607, 0.5234, 0.5537, 0.5094], [0.6746, 0.4583, 0.8689, 0.3938, 0.4130, 0.2602, 0.6126, 0.5131], [0.4435, 0.2818, 0.6643, 0.2294, 0.3795, 0.2038, 0.5428, 0.5347], [0.7686, 0.5124, 0.8225, 0.4737, 0.4038, 0.5151, 0.5450, 0.4887], [0.3795, 0.2579, 0.6317, 0.2452, 0.3483, 0.2038, 0.5323, 0.5397], [0.2499, 0.1584, 0.8249, 0.2532, 0.4639, 0.2239, 0.6906, 0.5346]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6179, 0.4008, 0.8600, 0.4015, 0.3932, 0.2515, 0.5711, 0.5438], [0.6267, 0.4080, 0.8438, 0.2633, 0.4762, 0.1800, 0.6259, 0.5240], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986], [0.6173, 0.4114, 0.7325, 0.2500, 0.4212, 0.1917, 0.5337, 0.5700], [0.6175, 0.3957, 0.8700, 0.4817, 0.4663, 0.5133, 0.5800, 0.5517], [0.0000, 0.0000, 0.6688, 0.2513, 0.4112, 0.2117, 0.5193, 0.5933], [0.0000, 0.0000, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552]]], device='cuda:0') loss_train_step before backward: tensor(0.0087, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0087, device='cuda:0', grad_fn=) loss_train: 0.31329219427425414 step: 87 running loss: 0.0036010597043017716 Train Steps: 87/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6136, 0.4085, 0.6688, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783], [0.6131, 0.4064, 0.8638, 0.5200, 0.4788, 0.4783, 0.5258, 0.5867], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6267, 0.4080, 0.8438, 0.2633, 0.4763, 0.1800, 0.6259, 0.5240], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136], [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7099, 0.4488, 0.8918, 0.5150, 0.3862, 0.4957, 0.6400, 0.4723], [0.5831, 0.3690, 0.6823, 0.2410, 0.3665, 0.2161, 0.5593, 0.5425], [0.6480, 0.4269, 0.8681, 0.5214, 0.4285, 0.4433, 0.5788, 0.5561], [0.7184, 0.4610, 0.8753, 0.4972, 0.4210, 0.5490, 0.6182, 0.5110], [0.6838, 0.4470, 0.8567, 0.2835, 0.4509, 0.1811, 0.6441, 0.5405], [0.5792, 0.3969, 0.7905, 0.3778, 0.3279, 0.3846, 0.5698, 0.5413], [0.6387, 0.4136, 0.8354, 0.3753, 0.3431, 0.4887, 0.6001, 0.5222], [0.6564, 0.4403, 0.8673, 0.5548, 0.3776, 0.4223, 0.5660, 0.5180]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6136, 0.4085, 0.6687, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783], [0.6132, 0.4063, 0.8637, 0.5200, 0.4787, 0.4783, 0.5258, 0.5867], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6267, 0.4080, 0.8438, 0.2633, 0.4762, 0.1800, 0.6259, 0.5240], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6132, 0.4118, 0.8200, 0.3633, 0.3562, 0.5400, 0.5787, 0.5136], [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.3143081188900396 step: 88 running loss: 0.003571683169204996 Train Steps: 88/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783], [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301], [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5503, 0.3433, 0.8042, 0.2656, 0.4390, 0.2441, 0.5905, 0.5752], [0.4924, 0.2846, 0.7159, 0.2282, 0.3960, 0.2076, 0.5492, 0.5356], [0.7602, 0.4789, 0.9267, 0.5469, 0.4050, 0.5384, 0.5807, 0.4911], [0.7297, 0.4698, 0.8829, 0.5198, 0.3922, 0.5400, 0.6037, 0.5108], [0.5250, 0.3183, 0.7260, 0.2407, 0.4251, 0.2064, 0.5545, 0.5222], [0.6908, 0.4439, 0.9060, 0.5334, 0.3466, 0.3828, 0.5711, 0.5140], [0.6375, 0.4331, 0.9175, 0.4850, 0.3439, 0.4380, 0.6430, 0.5255], [0.7003, 0.4664, 0.9140, 0.5508, 0.4439, 0.5048, 0.5472, 0.5362]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6229, 0.4198, 0.7663, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783], [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301], [0.6154, 0.4112, 0.7038, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6140, 0.4070, 0.8700, 0.5000, 0.4613, 0.4900, 0.5260, 0.5852]]], device='cuda:0') loss_train_step before backward: tensor(0.0025, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0025, device='cuda:0', grad_fn=) loss_train: 0.31682477600406855 step: 89 running loss: 0.0035598289438659387 Train Steps: 89/90 Loss: 0.0036 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5837, 0.5317], [0.6059, 0.4002, 0.7562, 0.2767, 0.3538, 0.3033, 0.5529, 0.5455], [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6276, 0.4120, 0.8738, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892], [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332], [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6138, 0.5883], [ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6008, 0.3638, 0.8715, 0.3505, 0.3895, 0.3336, 0.5966, 0.5717], [0.6301, 0.4162, 0.7736, 0.2896, 0.3732, 0.3400, 0.5808, 0.5527], [0.7209, 0.4506, 0.9106, 0.5676, 0.3906, 0.4272, 0.5947, 0.5073], [0.6660, 0.4047, 0.9162, 0.3639, 0.4761, 0.2391, 0.6249, 0.5336], [0.7238, 0.4595, 0.8319, 0.5199, 0.3889, 0.5274, 0.6024, 0.5363], [0.7352, 0.4680, 0.9079, 0.4925, 0.4598, 0.5879, 0.5617, 0.5313], [0.2352, 0.1203, 0.7305, 0.2232, 0.4102, 0.2230, 0.5104, 0.5519], [0.7288, 0.4646, 0.8875, 0.5587, 0.3863, 0.4118, 0.5313, 0.5509]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5838, 0.5317], [0.6059, 0.4002, 0.7563, 0.2767, 0.3537, 0.3033, 0.5529, 0.5455], [0.6272, 0.4071, 0.8737, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6276, 0.4120, 0.8737, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892], [0.6289, 0.4019, 0.8112, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332], [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6137, 0.5883], [0.0000, 0.0000, 0.7268, 0.2333, 0.4125, 0.1933, 0.5113, 0.5383], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5412, 0.5683]]], device='cuda:0') loss_train_step before backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train: 0.31965578126255423 step: 90 running loss: 0.003551730902917269 Valid Steps: 10/10 Loss: nan 36 -------------------------------------------------- Epoch: 2 Train Loss: 0.0036 Valid Loss: nan -------------------------------------------------- size of train loader is: 90 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103], [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750], [0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983], [0.6200, 0.4049, 0.8638, 0.5617, 0.4125, 0.5100, 0.6013, 0.5317]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6002, 0.3815, 0.8933, 0.5012, 0.4067, 0.4831, 0.5530, 0.5773], [0.6410, 0.4013, 0.9355, 0.4723, 0.4100, 0.4706, 0.5617, 0.5201], [0.5869, 0.3487, 0.8109, 0.3059, 0.3688, 0.3311, 0.5628, 0.5760], [0.6995, 0.4231, 0.9138, 0.5172, 0.4408, 0.4911, 0.5576, 0.5447], [0.6746, 0.4112, 0.9438, 0.5046, 0.4605, 0.5980, 0.6225, 0.5350], [0.3981, 0.2286, 0.7361, 0.2425, 0.4561, 0.2127, 0.5571, 0.5786], [0.7015, 0.4287, 0.9322, 0.5515, 0.4549, 0.4995, 0.5993, 0.5136], [0.6751, 0.4332, 0.8995, 0.5880, 0.4387, 0.5118, 0.5768, 0.5340]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6186, 0.4060, 0.8750, 0.5050, 0.3537, 0.4367, 0.5813, 0.6083], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103], [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5512, 0.5750], [0.6219, 0.3934, 0.8687, 0.5267, 0.4313, 0.4967, 0.5987, 0.4983], [0.6199, 0.4049, 0.8637, 0.5617, 0.4125, 0.5100, 0.6012, 0.5317]]], device='cuda:0') loss_train_step before backward: tensor(0.0023, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0023, device='cuda:0', grad_fn=) loss_train: 0.002288524992763996 step: 1 running loss: 0.002288524992763996 Train Steps: 1/90 Loss: 0.0023 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5563, 0.5533], [0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799], [0.6030, 0.3969, 0.7988, 0.3917, 0.3450, 0.3667, 0.5266, 0.4700], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6084, 0.3981, 0.8588, 0.5233, 0.4600, 0.5367, 0.5680, 0.5006], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6224, 0.4179, 0.8700, 0.5683, 0.4037, 0.4683, 0.5650, 0.5600]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5669, 0.3610, 0.8866, 0.5139, 0.4175, 0.4738, 0.5647, 0.5553], [0.5978, 0.3696, 0.8827, 0.4233, 0.4464, 0.3263, 0.5813, 0.5786], [0.5948, 0.3882, 0.8695, 0.4031, 0.3815, 0.3790, 0.5696, 0.5258], [0.5551, 0.3560, 0.9126, 0.4513, 0.4443, 0.5119, 0.5792, 0.5305], [0.4847, 0.2875, 0.7358, 0.2585, 0.4195, 0.2411, 0.5638, 0.5675], [0.5462, 0.3488, 0.8821, 0.5407, 0.4778, 0.5131, 0.5600, 0.5296], [0.6726, 0.4098, 0.9118, 0.5105, 0.4304, 0.5799, 0.6213, 0.5177], [0.6760, 0.4343, 0.8956, 0.5829, 0.4141, 0.4797, 0.5930, 0.5445]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5562, 0.5533], [0.6250, 0.4236, 0.8637, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799], [0.6030, 0.3969, 0.7987, 0.3917, 0.3450, 0.3667, 0.5266, 0.4700], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6084, 0.3981, 0.8587, 0.5233, 0.4600, 0.5367, 0.5680, 0.5006], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6224, 0.4179, 0.8700, 0.5683, 0.4038, 0.4683, 0.5650, 0.5600]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.0036328728310763836 step: 2 running loss: 0.0018164364155381918 Train Steps: 2/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6267, 0.4080, 0.8438, 0.2633, 0.4763, 0.1800, 0.6259, 0.5240], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [ nan, nan, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999], [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6359, 0.4082, 0.8793, 0.6344, 0.4539, 0.5514, 0.5787, 0.5307], [0.6908, 0.4302, 0.8753, 0.3369, 0.4884, 0.2461, 0.6141, 0.5311], [0.5758, 0.3436, 0.8435, 0.3540, 0.4212, 0.2781, 0.5877, 0.5331], [0.3117, 0.1714, 0.7735, 0.2756, 0.4538, 0.2254, 0.5412, 0.5513], [0.5630, 0.3703, 0.7986, 0.3373, 0.4174, 0.3504, 0.5790, 0.6113], [0.6433, 0.3916, 0.7148, 0.3738, 0.3802, 0.3490, 0.5193, 0.5787], [0.6158, 0.4063, 0.9110, 0.5084, 0.3871, 0.4496, 0.5415, 0.5369], [0.5936, 0.3585, 0.8877, 0.3856, 0.3611, 0.4360, 0.6028, 0.5406]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6229, 0.4066, 0.8512, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6267, 0.4080, 0.8438, 0.2633, 0.4762, 0.1800, 0.6259, 0.5240], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.0000, 0.0000, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999], [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290]]], device='cuda:0') loss_train_step before backward: tensor(0.0035, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0035, device='cuda:0', grad_fn=) loss_train: 0.00717794056981802 step: 3 running loss: 0.0023926468566060066 Train Steps: 3/90 Loss: 0.0024 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507], [0.6206, 0.4001, 0.8900, 0.3933, 0.3588, 0.3567, 0.5837, 0.5083], [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100], [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501], [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290], [0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6296, 0.4197, 0.6843, 0.2552, 0.4132, 0.2502, 0.5549, 0.5667], [0.4891, 0.3154, 0.9120, 0.4422, 0.3716, 0.3887, 0.5510, 0.5275], [0.5897, 0.3899, 0.8455, 0.5802, 0.4190, 0.5284, 0.5896, 0.5246], [0.6052, 0.3880, 0.7896, 0.2977, 0.4430, 0.2461, 0.5919, 0.5701], [0.5638, 0.3631, 0.7524, 0.2583, 0.4390, 0.2381, 0.6076, 0.5492], [0.5158, 0.3366, 0.9171, 0.5129, 0.3831, 0.5358, 0.5797, 0.4969], [0.5390, 0.3384, 0.7859, 0.3426, 0.3885, 0.3116, 0.5779, 0.5692], [0.5654, 0.3864, 0.8075, 0.4019, 0.3687, 0.3412, 0.5067, 0.5633]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6072, 0.4029, 0.7038, 0.2150, 0.3913, 0.2267, 0.5516, 0.5507], [0.6206, 0.4001, 0.8900, 0.3933, 0.3587, 0.3567, 0.5838, 0.5083], [0.6299, 0.4008, 0.8450, 0.5350, 0.4212, 0.5000, 0.6350, 0.5100], [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501], [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290], [0.6264, 0.4049, 0.8988, 0.4633, 0.3812, 0.4983, 0.6326, 0.4843], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5412, 0.5433]]], device='cuda:0') loss_train_step before backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train: 0.008925004513002932 step: 4 running loss: 0.002231251128250733 Train Steps: 4/90 Loss: 0.0022 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6275, 0.4050, 0.9038, 0.3767, 0.3838, 0.3533, 0.7074, 0.5575], [0.6339, 0.4149, 0.8800, 0.5000, 0.3900, 0.5283, 0.7541, 0.5424], [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879], [0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667], [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6362, 0.5200], [0.6129, 0.4063, 0.8738, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6101, 0.4008, 0.8775, 0.3697, 0.3979, 0.3186, 0.6877, 0.5551], [0.5939, 0.4007, 0.8892, 0.5176, 0.3852, 0.5240, 0.6640, 0.5635], [0.6112, 0.4214, 0.7873, 0.2553, 0.3898, 0.2494, 0.6111, 0.5374], [0.5689, 0.3956, 0.8435, 0.4460, 0.3896, 0.5423, 0.5555, 0.5107], [0.6235, 0.4064, 0.7775, 0.2624, 0.4293, 0.2045, 0.6141, 0.5317], [0.5731, 0.3912, 0.8131, 0.3550, 0.3671, 0.3545, 0.5398, 0.5531], [0.5109, 0.3331, 0.8373, 0.3655, 0.3707, 0.5275, 0.5966, 0.5326], [0.6127, 0.4086, 0.8259, 0.5347, 0.4222, 0.4452, 0.5156, 0.5748]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6275, 0.4050, 0.9038, 0.3767, 0.3837, 0.3533, 0.7074, 0.5575], [0.6339, 0.4149, 0.8800, 0.5000, 0.3900, 0.5283, 0.7541, 0.5424], [0.6257, 0.4034, 0.8288, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879], [0.6277, 0.4056, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667], [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6363, 0.5200], [0.6130, 0.4063, 0.8737, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.009910384309478104 step: 5 running loss: 0.001982076861895621 Train Steps: 5/90 Loss: 0.0020 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767], [0.6193, 0.3930, 0.8949, 0.4437, 0.3852, 0.5435, 0.6263, 0.5263], [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402], [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6028, 0.4081, 0.8787, 0.4410, 0.3613, 0.3906, 0.6559, 0.5075], [0.5579, 0.3736, 0.8574, 0.4347, 0.3937, 0.4630, 0.5563, 0.5520], [0.6174, 0.4114, 0.8715, 0.4025, 0.3697, 0.5316, 0.6222, 0.5028], [0.5251, 0.3515, 0.8400, 0.4720, 0.3916, 0.4560, 0.5975, 0.5433], [0.6194, 0.4004, 0.8417, 0.4841, 0.3773, 0.4897, 0.6316, 0.4948], [0.4978, 0.3611, 0.7664, 0.2651, 0.3495, 0.2716, 0.5151, 0.5461], [0.6735, 0.4743, 0.7903, 0.2001, 0.4035, 0.2606, 0.6323, 0.5188], [0.5516, 0.3580, 0.7681, 0.2637, 0.4117, 0.2454, 0.5465, 0.5668]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6163, 0.4001, 0.8788, 0.5033, 0.4013, 0.4633, 0.5337, 0.5767], [0.6193, 0.3930, 0.8949, 0.4437, 0.3852, 0.5435, 0.6263, 0.5263], [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402], [0.6275, 0.4024, 0.8500, 0.5383, 0.3913, 0.4883, 0.6288, 0.5100], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6182, 0.4099, 0.7812, 0.3000, 0.3938, 0.2367, 0.5325, 0.5750]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.011184405419044197 step: 6 running loss: 0.0018640675698406994 Train Steps: 6/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6152, 0.4131, 0.6863, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800], [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6185, 0.4042, 0.7700, 0.2250, 0.4062, 0.2117, 0.5763, 0.5150], [0.6234, 0.4179, 0.7825, 0.3450, 0.3813, 0.2867, 0.5675, 0.5617], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6380, 0.4439, 0.7222, 0.2509, 0.3482, 0.3219, 0.5899, 0.5464], [0.5228, 0.3732, 0.7454, 0.2522, 0.3416, 0.2843, 0.5357, 0.5352], [0.6708, 0.4600, 0.7047, 0.2072, 0.4197, 0.1954, 0.5537, 0.5195], [0.6443, 0.4543, 0.8250, 0.2313, 0.3901, 0.2928, 0.6525, 0.5063], [0.6395, 0.4302, 0.7451, 0.2097, 0.3914, 0.2084, 0.6203, 0.5268], [0.5345, 0.3629, 0.7980, 0.3231, 0.3868, 0.2772, 0.6029, 0.5457], [0.4908, 0.3311, 0.9332, 0.4952, 0.3968, 0.5405, 0.6285, 0.5112], [0.5857, 0.3792, 0.9133, 0.4411, 0.3189, 0.3980, 0.6537, 0.4978]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6152, 0.4131, 0.6862, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5462, 0.5800], [0.6154, 0.4112, 0.7038, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6184, 0.4042, 0.7700, 0.2250, 0.4062, 0.2117, 0.5763, 0.5150], [0.6234, 0.4179, 0.7825, 0.3450, 0.3812, 0.2867, 0.5675, 0.5617], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.012649990734644234 step: 7 running loss: 0.0018071415335206048 Train Steps: 7/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320], [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496], [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400], [0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791], [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033], [0.6261, 0.3987, 0.8688, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6730, 0.4779, 0.7829, 0.1793, 0.3983, 0.2202, 0.6320, 0.5176], [0.6574, 0.4479, 0.8553, 0.2471, 0.3966, 0.2033, 0.6424, 0.5357], [0.5816, 0.3918, 0.8557, 0.3419, 0.3306, 0.4048, 0.6079, 0.5148], [0.5688, 0.3792, 0.8812, 0.4301, 0.3675, 0.5155, 0.6587, 0.5246], [0.6893, 0.4635, 0.6720, 0.2585, 0.3296, 0.2493, 0.5315, 0.5631], [0.5993, 0.3999, 0.8540, 0.4222, 0.3700, 0.4330, 0.6404, 0.4731], [0.5385, 0.3690, 0.8455, 0.4344, 0.3897, 0.4744, 0.5790, 0.5234], [0.6363, 0.4504, 0.8473, 0.4180, 0.4055, 0.4664, 0.6760, 0.5129]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320], [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496], [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6137, 0.5400], [0.6254, 0.3993, 0.8988, 0.4767, 0.3988, 0.5517, 0.6955, 0.5285], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791], [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033], [0.6261, 0.3987, 0.8687, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309]]], device='cuda:0') loss_train_step before backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train: 0.014307374716736376 step: 8 running loss: 0.001788421839592047 Train Steps: 8/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633], [0.6197, 0.4118, 0.8688, 0.5517, 0.4037, 0.5233, 0.5875, 0.5600], [0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748], [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650], [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6075, 0.4180, 0.8209, 0.1754, 0.4414, 0.2094, 0.7048, 0.5392], [0.5856, 0.4044, 0.6639, 0.1650, 0.3894, 0.1484, 0.5538, 0.5401], [0.6596, 0.4292, 0.8553, 0.4984, 0.3863, 0.5089, 0.6441, 0.4961], [0.6291, 0.4204, 0.8378, 0.4888, 0.3686, 0.5141, 0.6246, 0.5025], [0.6524, 0.4352, 0.8420, 0.4286, 0.3649, 0.2985, 0.6814, 0.5026], [0.6159, 0.4232, 0.7951, 0.2934, 0.3295, 0.2710, 0.5499, 0.5150], [0.6264, 0.4401, 0.7368, 0.1997, 0.3638, 0.2359, 0.5624, 0.5183], [0.5910, 0.3714, 0.8494, 0.4495, 0.3998, 0.4437, 0.6143, 0.4924]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6270, 0.4266, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [0.6192, 0.4128, 0.8512, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633], [0.6197, 0.4118, 0.8687, 0.5517, 0.4038, 0.5233, 0.5875, 0.5600], [0.6336, 0.4191, 0.8938, 0.5167, 0.3938, 0.3517, 0.7343, 0.5748], [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5412, 0.5433], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650], [0.6126, 0.3954, 0.8537, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350]]], device='cuda:0') loss_train_step before backward: tensor(0.0023, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0023, device='cuda:0', grad_fn=) loss_train: 0.016634083236567676 step: 9 running loss: 0.0018482314707297417 Train Steps: 9/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272], [0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983], [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852], [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519], [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650], [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301], [0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050], [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6524, 0.4339, 0.8155, 0.3268, 0.3706, 0.2373, 0.5624, 0.5438], [0.5945, 0.3936, 0.8238, 0.4429, 0.4082, 0.4306, 0.6347, 0.4983], [0.5720, 0.4040, 0.8387, 0.4393, 0.4390, 0.4357, 0.5802, 0.5505], [0.6188, 0.4136, 0.8323, 0.4499, 0.4364, 0.4505, 0.5917, 0.5264], [0.6634, 0.4575, 0.7849, 0.3285, 0.3516, 0.2794, 0.5788, 0.5534], [0.5991, 0.3866, 0.8214, 0.4122, 0.3813, 0.4463, 0.6662, 0.5194], [0.6810, 0.4477, 0.8417, 0.3520, 0.3407, 0.4455, 0.6495, 0.5353], [0.6106, 0.3994, 0.7506, 0.2554, 0.3855, 0.2089, 0.5507, 0.5628]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6186, 0.4080, 0.8625, 0.3483, 0.3787, 0.2650, 0.5320, 0.5272], [0.6219, 0.3934, 0.8687, 0.5267, 0.4313, 0.4967, 0.5987, 0.4983], [0.6140, 0.4070, 0.8700, 0.5000, 0.4613, 0.4900, 0.5260, 0.5852], [0.6100, 0.4016, 0.8600, 0.5067, 0.4613, 0.5233, 0.5086, 0.5519], [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650], [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301], [0.6200, 0.4112, 0.8863, 0.4100, 0.3638, 0.4917, 0.6087, 0.6050], [0.6182, 0.4099, 0.7812, 0.3000, 0.3938, 0.2367, 0.5325, 0.5750]]], device='cuda:0') loss_train_step before backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train: 0.018363903160206974 step: 10 running loss: 0.0018363903160206973 Train Steps: 10/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6317, 0.4038, 0.8287, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986], [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917], [0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6670, 0.4386, 0.8303, 0.5567, 0.4043, 0.4091, 0.6370, 0.5482], [0.6776, 0.4518, 0.7636, 0.4052, 0.3734, 0.2846, 0.5586, 0.5967], [0.6549, 0.4292, 0.8408, 0.4363, 0.4181, 0.5187, 0.6062, 0.5413], [0.6972, 0.4272, 0.8284, 0.4156, 0.3770, 0.4023, 0.5834, 0.5589], [0.6584, 0.4397, 0.8381, 0.2707, 0.4301, 0.2266, 0.6898, 0.5920], [0.6660, 0.4326, 0.8762, 0.4353, 0.4584, 0.5287, 0.6234, 0.5476], [0.5939, 0.3899, 0.8410, 0.4262, 0.3975, 0.3439, 0.5154, 0.5441], [0.5964, 0.3956, 0.8295, 0.4933, 0.4590, 0.4670, 0.5879, 0.5507]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6317, 0.4038, 0.8288, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986], [0.6179, 0.4118, 0.7278, 0.4237, 0.3587, 0.3400, 0.5675, 0.5917], [0.6064, 0.4019, 0.8650, 0.4517, 0.4038, 0.5367, 0.5703, 0.5609], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6286, 0.4086, 0.8408, 0.2801, 0.4162, 0.2800, 0.6725, 0.5393], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6127, 0.4084, 0.8700, 0.4467, 0.3988, 0.4317, 0.5013, 0.5471], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.019579105195589364 step: 11 running loss: 0.0017799186541444876 Train Steps: 11/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683], [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500], [0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050], [0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954], [0.6200, 0.3999, 0.8653, 0.5207, 0.4100, 0.5125, 0.5975, 0.5103]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6627, 0.4582, 0.8446, 0.3431, 0.4621, 0.3512, 0.6252, 0.6020], [0.6581, 0.4269, 0.7875, 0.3111, 0.3887, 0.4321, 0.5941, 0.5865], [0.6116, 0.3716, 0.7028, 0.2405, 0.4561, 0.2085, 0.5293, 0.5720], [0.6217, 0.4010, 0.9149, 0.5132, 0.4004, 0.4406, 0.5760, 0.5551], [0.7454, 0.4856, 0.7083, 0.3050, 0.4564, 0.2597, 0.4888, 0.6232], [0.6429, 0.3958, 0.9150, 0.5988, 0.4104, 0.5090, 0.6470, 0.5615], [0.6573, 0.4318, 0.8670, 0.4660, 0.4384, 0.3084, 0.5093, 0.5605], [0.6276, 0.4006, 0.9061, 0.5831, 0.4815, 0.5650, 0.5835, 0.5617]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6200, 0.4098, 0.8238, 0.2917, 0.4013, 0.2967, 0.6000, 0.5683], [0.6197, 0.4051, 0.7812, 0.2650, 0.3512, 0.4050, 0.6112, 0.5500], [0.6054, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050], [0.6147, 0.4026, 0.6600, 0.2467, 0.4087, 0.2150, 0.5489, 0.5773], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.6137, 0.4038, 0.8562, 0.4050, 0.3812, 0.2550, 0.5106, 0.4954], [0.6200, 0.3999, 0.8653, 0.5207, 0.4100, 0.5125, 0.5975, 0.5103]]], device='cuda:0') loss_train_step before backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train: 0.021589155890978873 step: 12 running loss: 0.0017990963242482394 Train Steps: 12/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136], [0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500], [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.6118, 0.4052, 0.8463, 0.3917, 0.3538, 0.3450, 0.5053, 0.5593]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6850, 0.4247, 0.7308, 0.2875, 0.4270, 0.2783, 0.5280, 0.5645], [0.6193, 0.3962, 0.8800, 0.6018, 0.4218, 0.5049, 0.5980, 0.5801], [0.6418, 0.3852, 0.8744, 0.3716, 0.3916, 0.4108, 0.6206, 0.5730], [0.6372, 0.4026, 0.9109, 0.5178, 0.4621, 0.5772, 0.5763, 0.5733], [0.6267, 0.3887, 0.8751, 0.5489, 0.4650, 0.5104, 0.5475, 0.5908], [0.6339, 0.4063, 0.8848, 0.5980, 0.4351, 0.4774, 0.5554, 0.5471], [0.6724, 0.4247, 0.7273, 0.3193, 0.4394, 0.2667, 0.5483, 0.5913], [0.6507, 0.4327, 0.8642, 0.4348, 0.3746, 0.3852, 0.4934, 0.6010]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6043, 0.4022, 0.6888, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136], [0.6273, 0.4143, 0.8750, 0.5700, 0.3988, 0.4717, 0.6012, 0.5467], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5838, 0.5500], [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.6118, 0.4052, 0.8462, 0.3917, 0.3537, 0.3450, 0.5053, 0.5593]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.022897822898812592 step: 13 running loss: 0.0017613709922163533 Train Steps: 13/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [ nan, nan, 0.7648, 0.2722, 0.3962, 0.2183, 0.5060, 0.5422], [ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767], [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7103, 0.4486, 0.8747, 0.4249, 0.3902, 0.5114, 0.6438, 0.5376], [0.3599, 0.2209, 0.7311, 0.3070, 0.3967, 0.2738, 0.4546, 0.5411], [0.4195, 0.2813, 0.7883, 0.3753, 0.4167, 0.2952, 0.4734, 0.5861], [0.6465, 0.4145, 0.8845, 0.5313, 0.3612, 0.4709, 0.6170, 0.5398], [0.6977, 0.4486, 0.8552, 0.6218, 0.4135, 0.4887, 0.5198, 0.5217], [0.6612, 0.4368, 0.8415, 0.3068, 0.4765, 0.3075, 0.6446, 0.5897], [0.7176, 0.4319, 0.9126, 0.5278, 0.3973, 0.5687, 0.6204, 0.5488], [0.7001, 0.4696, 0.8438, 0.5102, 0.4606, 0.3406, 0.4802, 0.6126]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [0.0000, 0.0000, 0.7648, 0.2722, 0.3963, 0.2183, 0.5060, 0.5422], [0.0000, 0.0000, 0.7613, 0.3250, 0.4038, 0.2533, 0.5437, 0.5767], [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209]]], device='cuda:0') loss_train_step before backward: tensor(0.0086, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0086, device='cuda:0', grad_fn=) loss_train: 0.031523438286967576 step: 14 running loss: 0.002251674163354827 Train Steps: 14/90 Loss: 0.0023 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117], [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633], [0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683], [ nan, nan, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737], [0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485], [0.6147, 0.4112, 0.7988, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550], [0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5613, 0.5900]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6369, 0.3770, 0.8405, 0.3821, 0.3651, 0.3864, 0.5791, 0.5156], [0.6178, 0.3775, 0.9148, 0.4095, 0.3825, 0.3501, 0.5726, 0.5146], [0.7046, 0.4203, 0.8782, 0.6498, 0.4341, 0.5928, 0.5846, 0.5494], [0.6253, 0.4038, 0.8356, 0.3593, 0.4174, 0.3532, 0.6002, 0.5456], [0.2293, 0.1415, 0.8490, 0.2963, 0.5242, 0.2863, 0.6327, 0.5384], [0.6541, 0.4069, 0.7483, 0.3571, 0.3859, 0.3337, 0.5057, 0.5481], [0.5698, 0.3742, 0.8092, 0.4008, 0.3791, 0.3166, 0.4940, 0.5553], [0.6750, 0.4287, 0.8318, 0.5494, 0.3564, 0.4079, 0.5192, 0.5380]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117], [0.6192, 0.4128, 0.8512, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633], [0.6200, 0.4098, 0.8238, 0.2917, 0.4013, 0.2967, 0.6000, 0.5683], [0.0000, 0.0000, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737], [0.6123, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485], [0.6147, 0.4112, 0.7987, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550], [0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5612, 0.5900]]], device='cuda:0') loss_train_step before backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train: 0.03430542966816574 step: 15 running loss: 0.0022870286445443827 Train Steps: 15/90 Loss: 0.0023 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843], [0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383], [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5944, 0.3454, 0.9136, 0.5033, 0.3575, 0.5314, 0.6047, 0.5082], [0.5365, 0.3196, 0.7885, 0.3145, 0.3401, 0.3181, 0.5186, 0.5181], [0.6247, 0.3771, 0.9237, 0.5689, 0.3381, 0.4124, 0.5901, 0.5049], [0.5429, 0.3474, 0.8863, 0.5152, 0.4129, 0.5361, 0.5636, 0.5641], [0.5060, 0.3406, 0.8971, 0.5473, 0.4520, 0.5284, 0.5180, 0.5835], [0.5774, 0.3656, 0.8471, 0.3874, 0.3530, 0.4066, 0.5703, 0.5504], [0.5550, 0.3419, 0.7782, 0.2975, 0.4284, 0.1834, 0.5793, 0.5391], [0.4954, 0.2894, 0.7917, 0.2573, 0.4166, 0.2622, 0.6149, 0.5335]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6264, 0.4049, 0.8988, 0.4633, 0.3812, 0.4983, 0.6326, 0.4843], [0.6079, 0.3964, 0.7420, 0.2958, 0.3562, 0.2917, 0.5351, 0.4980], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6140, 0.4070, 0.8700, 0.5000, 0.4613, 0.4900, 0.5260, 0.5852], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383], [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290]]], device='cuda:0') loss_train_step before backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train: 0.03636402601841837 step: 16 running loss: 0.0022727516261511482 Train Steps: 16/90 Loss: 0.0023 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6199, 0.4093, 0.7913, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700], [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869], [0.6198, 0.4130, 0.8762, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103], [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967], [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100], [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4330, 0.2798, 0.8786, 0.2619, 0.4194, 0.2778, 0.6735, 0.5339], [0.5555, 0.3388, 0.8658, 0.5031, 0.3932, 0.5309, 0.6476, 0.5368], [0.4834, 0.3136, 0.8026, 0.2625, 0.3900, 0.2671, 0.5784, 0.5424], [0.5196, 0.3330, 0.8274, 0.2463, 0.4491, 0.2297, 0.6057, 0.4950], [0.5648, 0.3605, 0.8913, 0.4223, 0.3230, 0.4654, 0.5457, 0.5211], [0.5265, 0.3438, 0.8746, 0.5002, 0.3539, 0.3281, 0.5949, 0.5066], [0.5507, 0.3276, 0.8567, 0.5455, 0.3664, 0.5009, 0.5833, 0.4898], [0.5677, 0.3474, 0.8651, 0.5104, 0.3851, 0.4639, 0.5004, 0.5649]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6198, 0.4093, 0.7912, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700], [0.6276, 0.4095, 0.8238, 0.2250, 0.4663, 0.1783, 0.6171, 0.4869], [0.6198, 0.4130, 0.8763, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103], [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967], [0.6275, 0.4024, 0.8500, 0.5383, 0.3913, 0.4883, 0.6288, 0.5100], [0.6163, 0.4001, 0.8788, 0.5033, 0.4013, 0.4633, 0.5337, 0.5767]]], device='cuda:0') loss_train_step before backward: tensor(0.0027, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0027, device='cuda:0', grad_fn=) loss_train: 0.03901701339054853 step: 17 running loss: 0.0022951184347381488 Train Steps: 17/90 Loss: 0.0023 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6246, 0.4028, 0.8738, 0.4867, 0.4088, 0.5667, 0.6362, 0.5200], [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5370, 0.3631, 0.8025, 0.4733, 0.3694, 0.4301, 0.6613, 0.5039], [0.5555, 0.3662, 0.8576, 0.4481, 0.3509, 0.3076, 0.5604, 0.5309], [0.5755, 0.3779, 0.8860, 0.4477, 0.3929, 0.5262, 0.6205, 0.5168], [0.5609, 0.3712, 0.8877, 0.3128, 0.3627, 0.2935, 0.6108, 0.4896], [0.4438, 0.3181, 0.7941, 0.2994, 0.3447, 0.2704, 0.5179, 0.5191], [0.5578, 0.3616, 0.8624, 0.2708, 0.4258, 0.2182, 0.6350, 0.5023], [0.6400, 0.3798, 0.8733, 0.4429, 0.4123, 0.5609, 0.6441, 0.5230], [0.2633, 0.2200, 0.7622, 0.2249, 0.3862, 0.2525, 0.5937, 0.5282]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6311, 0.3998, 0.7975, 0.5767, 0.3837, 0.4850, 0.7327, 0.5343], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6250, 0.4131, 0.8687, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6246, 0.4028, 0.8737, 0.4867, 0.4087, 0.5667, 0.6363, 0.5200], [0.0000, 0.0000, 0.7525, 0.2291, 0.3837, 0.3017, 0.6050, 0.5667]]], device='cuda:0') loss_train_step before backward: tensor(0.0036, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0036, device='cuda:0', grad_fn=) loss_train: 0.04258008219767362 step: 18 running loss: 0.002365560122092979 Train Steps: 18/90 Loss: 0.0024 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6131, 0.4064, 0.8638, 0.5200, 0.4788, 0.4783, 0.5258, 0.5867], [0.6197, 0.3986, 0.8800, 0.4617, 0.4188, 0.4783, 0.5687, 0.5550], [ nan, nan, 0.8938, 0.2850, 0.4662, 0.3117, 0.7406, 0.5528], [0.6201, 0.4151, 0.8588, 0.5467, 0.3700, 0.3950, 0.5637, 0.5933], [0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783], [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320], [0.6296, 0.4045, 0.9138, 0.4100, 0.4232, 0.4242, 0.7422, 0.5297], [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5918, 0.3952, 0.8584, 0.4755, 0.4334, 0.4507, 0.5884, 0.5322], [0.5844, 0.3977, 0.8591, 0.4129, 0.3728, 0.4645, 0.6018, 0.5118], [0.2210, 0.1637, 0.9241, 0.2472, 0.4548, 0.2745, 0.7776, 0.5157], [0.6087, 0.4235, 0.8215, 0.5093, 0.3536, 0.3738, 0.5898, 0.5271], [0.5043, 0.3472, 0.7739, 0.2174, 0.4279, 0.2316, 0.6006, 0.5307], [0.5409, 0.3545, 0.8611, 0.3833, 0.3235, 0.4136, 0.6816, 0.4781], [0.6830, 0.4531, 0.8903, 0.3539, 0.3679, 0.4203, 0.7136, 0.4970], [0.5697, 0.3720, 0.8571, 0.4214, 0.4092, 0.4769, 0.5778, 0.4975]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6132, 0.4063, 0.8637, 0.5200, 0.4787, 0.4783, 0.5258, 0.5867], [0.6197, 0.3986, 0.8800, 0.4617, 0.4187, 0.4783, 0.5688, 0.5550], [0.0000, 0.0000, 0.8938, 0.2850, 0.4663, 0.3117, 0.7406, 0.5528], [0.6202, 0.4151, 0.8587, 0.5467, 0.3700, 0.3950, 0.5638, 0.5933], [0.6229, 0.4198, 0.7663, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783], [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320], [0.6296, 0.4045, 0.9137, 0.4100, 0.4232, 0.4242, 0.7422, 0.5297], [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097]]], device='cuda:0') loss_train_step before backward: tensor(0.0031, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0031, device='cuda:0', grad_fn=) loss_train: 0.045656067435629666 step: 19 running loss: 0.0024029509176647194 Train Steps: 19/90 Loss: 0.0024 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8850, 0.2817, 0.5112, 0.2183, 0.7184, 0.5436], [0.6277, 0.4083, 0.8350, 0.2717, 0.4562, 0.1800, 0.5918, 0.4878], [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [ nan, nan, 0.7648, 0.2722, 0.3962, 0.2183, 0.5060, 0.5422], [0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921], [0.6136, 0.4117, 0.8700, 0.5167, 0.4188, 0.5083, 0.5147, 0.5495], [0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667], [0.6260, 0.4120, 0.8013, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.3317, 0.2437, 0.8838, 0.2707, 0.5007, 0.2329, 0.7101, 0.5134], [0.4226, 0.2902, 0.8046, 0.2608, 0.4476, 0.2214, 0.6137, 0.5088], [0.5430, 0.3784, 0.8516, 0.2332, 0.5131, 0.2774, 0.7556, 0.5394], [0.1304, 0.1043, 0.7471, 0.2489, 0.3800, 0.2012, 0.5221, 0.5210], [0.6030, 0.3979, 0.8176, 0.2325, 0.4119, 0.2649, 0.6792, 0.5080], [0.7374, 0.4957, 0.8782, 0.5724, 0.4039, 0.4957, 0.5789, 0.5219], [0.6925, 0.4588, 0.7116, 0.2866, 0.3439, 0.3203, 0.5630, 0.5454], [0.6044, 0.3941, 0.7845, 0.2106, 0.4687, 0.1488, 0.6680, 0.5050]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8850, 0.2817, 0.5113, 0.2183, 0.7184, 0.5436], [0.6277, 0.4083, 0.8350, 0.2717, 0.4563, 0.1800, 0.5918, 0.4878], [0.6300, 0.4133, 0.8537, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.0000, 0.0000, 0.7648, 0.2722, 0.3963, 0.2183, 0.5060, 0.5422], [0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921], [0.6136, 0.4117, 0.8700, 0.5167, 0.4187, 0.5083, 0.5147, 0.5495], [0.6134, 0.4090, 0.6926, 0.2819, 0.3537, 0.3233, 0.5562, 0.5667], [0.6259, 0.4120, 0.8012, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895]]], device='cuda:0') loss_train_step before backward: tensor(0.0051, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0051, device='cuda:0', grad_fn=) loss_train: 0.050726136774756014 step: 20 running loss: 0.002536306838737801 Train Steps: 20/90 Loss: 0.0025 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333], [0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300], [0.6115, 0.4081, 0.6725, 0.2433, 0.4088, 0.1933, 0.5167, 0.5544], [0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397], [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261], [0.6201, 0.4151, 0.8588, 0.5467, 0.3700, 0.3950, 0.5637, 0.5933], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5901, 0.4298, 0.8875, 0.4076, 0.3987, 0.4621, 0.6725, 0.5381], [0.5370, 0.3682, 0.7592, 0.1734, 0.4341, 0.2684, 0.6666, 0.5472], [0.4572, 0.3172, 0.7158, 0.2226, 0.4405, 0.1772, 0.5711, 0.5436], [0.5756, 0.3987, 0.8808, 0.4527, 0.3977, 0.3508, 0.5557, 0.5388], [0.5095, 0.3654, 0.8806, 0.4264, 0.4335, 0.4443, 0.6113, 0.5361], [0.5600, 0.4047, 0.8448, 0.5110, 0.4104, 0.3563, 0.6075, 0.5725], [0.4382, 0.3084, 0.8828, 0.2070, 0.4856, 0.2378, 0.7775, 0.5535], [0.6409, 0.4133, 0.8676, 0.3542, 0.4018, 0.4103, 0.6507, 0.5465]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6137, 0.5333], [0.6207, 0.4081, 0.7663, 0.2067, 0.3963, 0.3200, 0.6313, 0.5300], [0.6115, 0.4081, 0.6725, 0.2433, 0.4087, 0.1933, 0.5167, 0.5544], [0.6127, 0.4084, 0.8737, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397], [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261], [0.6202, 0.4151, 0.8587, 0.5467, 0.3700, 0.3950, 0.5638, 0.5933], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131]]], device='cuda:0') loss_train_step before backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train: 0.053505434771068394 step: 21 running loss: 0.002547877846241352 Train Steps: 21/90 Loss: 0.0025 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6268, 0.4061, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300], [0.6069, 0.3975, 0.8625, 0.5083, 0.4388, 0.5483, 0.5650, 0.4967], [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268], [0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150], [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899], [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332], [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5056, 0.3462, 0.8709, 0.2943, 0.4799, 0.2926, 0.6936, 0.5203], [0.5236, 0.3603, 0.8550, 0.1726, 0.4822, 0.1785, 0.6520, 0.5347], [0.5216, 0.3837, 0.8388, 0.4630, 0.4735, 0.4920, 0.5264, 0.5561], [0.5107, 0.3523, 0.8397, 0.4314, 0.4522, 0.4588, 0.5876, 0.5638], [0.4940, 0.3262, 0.7804, 0.2140, 0.4554, 0.2161, 0.5925, 0.5326], [0.5580, 0.3879, 0.8799, 0.4109, 0.4304, 0.3769, 0.6884, 0.5656], [0.6593, 0.4375, 0.8602, 0.4501, 0.4064, 0.3761, 0.6315, 0.5474], [0.5514, 0.3732, 0.8135, 0.4741, 0.4225, 0.4865, 0.6201, 0.5624]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6268, 0.4060, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300], [0.6069, 0.3975, 0.8625, 0.5083, 0.4387, 0.5483, 0.5650, 0.4967], [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268], [0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150], [0.6346, 0.4144, 0.9087, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899], [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332], [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204]]], device='cuda:0') loss_train_step before backward: tensor(0.0025, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0025, device='cuda:0', grad_fn=) loss_train: 0.05599151493515819 step: 22 running loss: 0.002545068860689009 Train Steps: 22/90 Loss: 0.0025 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933], [0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717], [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517], [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6138, 0.5883], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6189, 0.3997, 0.8820, 0.4238, 0.4160, 0.3840, 0.6119, 0.5490], [0.0977, 0.0911, 0.6965, 0.2213, 0.4232, 0.1823, 0.5793, 0.5771], [0.6315, 0.4235, 0.7317, 0.2649, 0.4093, 0.2834, 0.5647, 0.5624], [0.6621, 0.4304, 0.8660, 0.4649, 0.4866, 0.4302, 0.5786, 0.5840], [0.6096, 0.3989, 0.8899, 0.3080, 0.4147, 0.2769, 0.6552, 0.5774], [0.6882, 0.4586, 0.8791, 0.4039, 0.4599, 0.5236, 0.6363, 0.5646], [0.6097, 0.4281, 0.8857, 0.4110, 0.4833, 0.5131, 0.6350, 0.5820], [0.6308, 0.4299, 0.8597, 0.4899, 0.3927, 0.3489, 0.5909, 0.5397]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6201, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.0000, 0.0000, 0.6688, 0.2513, 0.4112, 0.2117, 0.5193, 0.5933], [0.6123, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5412, 0.5717], [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6122, 0.4006, 0.8850, 0.4217, 0.4087, 0.5517, 0.6062, 0.5517], [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6137, 0.5883], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.05741637130267918 step: 23 running loss: 0.0024963639696817036 Train Steps: 23/90 Loss: 0.0025 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583], [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [ nan, nan, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533], [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323], [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064], [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309], [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6483, 0.4329, 0.8978, 0.4326, 0.3863, 0.4186, 0.5454, 0.5792], [0.6430, 0.4231, 0.7608, 0.3607, 0.3721, 0.4013, 0.5512, 0.5972], [0.6352, 0.3953, 0.7954, 0.2319, 0.4659, 0.2090, 0.6230, 0.5244], [0.0224, 0.0161, 0.6969, 0.2110, 0.4282, 0.1571, 0.5344, 0.5740], [0.6293, 0.4141, 0.7977, 0.2335, 0.4426, 0.1968, 0.6254, 0.5558], [0.6616, 0.4534, 0.8593, 0.5196, 0.4908, 0.4956, 0.5336, 0.5511], [0.6865, 0.4557, 0.8721, 0.4978, 0.3912, 0.3721, 0.6095, 0.5410], [0.7057, 0.4689, 0.8426, 0.5325, 0.4386, 0.5152, 0.6351, 0.5462]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6212, 0.4033, 0.8938, 0.4167, 0.3812, 0.4267, 0.5612, 0.5583], [0.6170, 0.4102, 0.7468, 0.3695, 0.3462, 0.3767, 0.5238, 0.5823], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.0000, 0.0000, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533], [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323], [0.6102, 0.4005, 0.8687, 0.5100, 0.4812, 0.5400, 0.5404, 0.5064], [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309], [0.6299, 0.4008, 0.8450, 0.5350, 0.4212, 0.5000, 0.6350, 0.5100]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.058127752621658146 step: 24 running loss: 0.0024219896925690896 Train Steps: 24/90 Loss: 0.0024 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822], [0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050], [ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617], [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285], [0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133], [0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100], [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6093, 0.3999, 0.8337, 0.4284, 0.3559, 0.3779, 0.5022, 0.5830], [0.6382, 0.4220, 0.8517, 0.4154, 0.3702, 0.5035, 0.5867, 0.5758], [0.2702, 0.1768, 0.7247, 0.2521, 0.4284, 0.2276, 0.5442, 0.5621], [0.6540, 0.4113, 0.8690, 0.3864, 0.4371, 0.2669, 0.6174, 0.5398], [0.6239, 0.3889, 0.8392, 0.5560, 0.4058, 0.5212, 0.6302, 0.5345], [0.6113, 0.4132, 0.8346, 0.5003, 0.4610, 0.4800, 0.5369, 0.5320], [0.7170, 0.4505, 0.8821, 0.4155, 0.3609, 0.4126, 0.6495, 0.5369], [0.6013, 0.3870, 0.8461, 0.4729, 0.3706, 0.4457, 0.5826, 0.5565]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6182, 0.4058, 0.8737, 0.4350, 0.3562, 0.3400, 0.5290, 0.5822], [0.6200, 0.4112, 0.8863, 0.4100, 0.3638, 0.4917, 0.6087, 0.6050], [0.0000, 0.0000, 0.7513, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617], [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285], [0.6308, 0.3990, 0.8687, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133], [0.6128, 0.4022, 0.8737, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100], [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450]]], device='cuda:0') loss_train_step before backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train: 0.06033143226522952 step: 25 running loss: 0.002413257290609181 Train Steps: 25/90 Loss: 0.0024 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183], [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717], [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183], [0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767], [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167], [0.6127, 0.4115, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495], [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938], [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6160, 0.3895, 0.7928, 0.3331, 0.3901, 0.3177, 0.5873, 0.5922], [0.5629, 0.3442, 0.7577, 0.2975, 0.3681, 0.2849, 0.5655, 0.5591], [0.6375, 0.3715, 0.9214, 0.4802, 0.3835, 0.5911, 0.6241, 0.5429], [0.6171, 0.3862, 0.8805, 0.5191, 0.4360, 0.5156, 0.5317, 0.5459], [0.6700, 0.4098, 0.8710, 0.5518, 0.3930, 0.5352, 0.5776, 0.5149], [0.5614, 0.3484, 0.7337, 0.3151, 0.3581, 0.3145, 0.5202, 0.5507], [0.7045, 0.4279, 0.8837, 0.5943, 0.3764, 0.4633, 0.6147, 0.5073], [0.4651, 0.2801, 0.6989, 0.2835, 0.3888, 0.2057, 0.5165, 0.5298]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6196, 0.4094, 0.7563, 0.2817, 0.3938, 0.3183, 0.6012, 0.6183], [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717], [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183], [0.6148, 0.4076, 0.8666, 0.4820, 0.4137, 0.5067, 0.5250, 0.5767], [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167], [0.6127, 0.4114, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495], [0.6334, 0.4037, 0.8637, 0.5733, 0.4013, 0.4717, 0.6369, 0.4938], [0.6159, 0.4085, 0.6900, 0.2283, 0.4087, 0.1950, 0.5123, 0.5397]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.06175887619610876 step: 26 running loss: 0.002375341392158029 Train Steps: 26/90 Loss: 0.0024 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5787, 0.5600], [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631], [ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563], [0.6304, 0.4024, 0.8925, 0.4800, 0.3937, 0.4817, 0.7485, 0.5297], [0.6034, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084], [0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6436, 0.3993, 0.8663, 0.5342, 0.4515, 0.5277, 0.5575, 0.5408], [0.6770, 0.3979, 0.8363, 0.4462, 0.3444, 0.5367, 0.5881, 0.5528], [0.2007, 0.1157, 0.7863, 0.3516, 0.3207, 0.3140, 0.5153, 0.5435], [0.6610, 0.3867, 0.8835, 0.5234, 0.3765, 0.5157, 0.6412, 0.5174], [0.6181, 0.3727, 0.7298, 0.2919, 0.3445, 0.3732, 0.5756, 0.5017], [0.6510, 0.3927, 0.8464, 0.4056, 0.3709, 0.3140, 0.5137, 0.5286], [0.6249, 0.3743, 0.8593, 0.4903, 0.3432, 0.3430, 0.5064, 0.5652], [0.6487, 0.4039, 0.7184, 0.2811, 0.4340, 0.1724, 0.5875, 0.5331]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5788, 0.5600], [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631], [0.0000, 0.0000, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563], [0.6304, 0.4024, 0.8925, 0.4800, 0.3938, 0.4817, 0.7485, 0.5297], [0.6033, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084], [0.6186, 0.4080, 0.8625, 0.3483, 0.3787, 0.2650, 0.5320, 0.5272], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383]]], device='cuda:0') loss_train_step before backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train: 0.06347480788826942 step: 27 running loss: 0.002350918810676645 Train Steps: 27/90 Loss: 0.0024 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611], [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283], [0.6117, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103], [0.6201, 0.4151, 0.8588, 0.5467, 0.3700, 0.3950, 0.5637, 0.5933]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6822, 0.4314, 0.8651, 0.4856, 0.3992, 0.4735, 0.5274, 0.5404], [0.5400, 0.3283, 0.8774, 0.4733, 0.4195, 0.5492, 0.6072, 0.5681], [0.4843, 0.2985, 0.8924, 0.5053, 0.3795, 0.5752, 0.6132, 0.5527], [0.6219, 0.3712, 0.8523, 0.3708, 0.3658, 0.2955, 0.6313, 0.5004], [0.6180, 0.4010, 0.8046, 0.3480, 0.3300, 0.3867, 0.5747, 0.5360], [0.6175, 0.3946, 0.8366, 0.5188, 0.4460, 0.4738, 0.5096, 0.5268], [0.5767, 0.3839, 0.6640, 0.2650, 0.3513, 0.2447, 0.5245, 0.5126], [0.6322, 0.4174, 0.8397, 0.5665, 0.3124, 0.4103, 0.5321, 0.5726]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611], [0.6184, 0.4079, 0.8838, 0.4617, 0.4837, 0.5650, 0.6175, 0.5850], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6277, 0.4036, 0.8687, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6084, 0.4008, 0.8587, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283], [0.6116, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103], [0.6202, 0.4151, 0.8587, 0.5467, 0.3700, 0.3950, 0.5638, 0.5933]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.06485963379964232 step: 28 running loss: 0.0023164154928443687 Train Steps: 28/90 Loss: 0.0023 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.3993, 0.8519, 0.4923, 0.3962, 0.4717, 0.6013, 0.5433], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567], [ nan, nan, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6118, 0.4052, 0.8463, 0.3917, 0.3538, 0.3450, 0.5053, 0.5593], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 6.7090e-01, 4.0646e-01, 8.8374e-01, 5.4438e-01, 3.8827e-01, 5.3380e-01, 5.8710e-01, 5.2107e-01], [ 6.3156e-01, 3.8918e-01, 8.8924e-01, 3.9016e-01, 4.5429e-01, 3.5994e-01, 7.0014e-01, 5.0003e-01], [ 7.1235e-01, 4.4388e-01, 8.7769e-01, 4.9250e-01, 3.6733e-01, 4.6729e-01, 5.2955e-01, 5.4011e-01], [-3.6269e-04, -8.9101e-03, 7.3422e-01, 3.1684e-01, 3.8392e-01, 2.9073e-01, 4.9513e-01, 5.4069e-01], [ 7.0830e-01, 4.4306e-01, 8.0668e-01, 3.2193e-01, 3.6984e-01, 3.3750e-01, 5.6709e-01, 5.1859e-01], [ 6.9812e-01, 4.5714e-01, 8.6463e-01, 4.2777e-01, 3.4170e-01, 3.8019e-01, 4.5121e-01, 5.2652e-01], [ 6.6152e-01, 4.4307e-01, 7.3589e-01, 3.0837e-01, 4.2011e-01, 2.6642e-01, 5.2988e-01, 5.5355e-01], [ 5.7453e-01, 3.4652e-01, 7.8571e-01, 2.3260e-01, 4.2730e-01, 2.6989e-01, 6.5486e-01, 5.0269e-01]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6200, 0.3993, 0.8519, 0.4923, 0.3963, 0.4717, 0.6012, 0.5433], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567], [0.0000, 0.0000, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6118, 0.4052, 0.8462, 0.3917, 0.3537, 0.3450, 0.5053, 0.5593], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.066403744276613 step: 29 running loss: 0.002289784285400448 Train Steps: 29/90 Loss: 0.0023 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767], [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.6317, 0.4038, 0.8287, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986], [0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283], [0.6214, 0.4112, 0.7838, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083], [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667], [0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6553, 0.4516, 0.8909, 0.4744, 0.3987, 0.3983, 0.5979, 0.5874], [0.6033, 0.3995, 0.9001, 0.3978, 0.3655, 0.4665, 0.5997, 0.5057], [0.6109, 0.3870, 0.8913, 0.4282, 0.3730, 0.3778, 0.5874, 0.5565], [0.6447, 0.4530, 0.8713, 0.6013, 0.3892, 0.4725, 0.6208, 0.5390], [0.6460, 0.4327, 0.8558, 0.5071, 0.4999, 0.4846, 0.5085, 0.5388], [0.6930, 0.4552, 0.7825, 0.2199, 0.3805, 0.2937, 0.5852, 0.5319], [0.1483, 0.1368, 0.7527, 0.2525, 0.4189, 0.2928, 0.5746, 0.5629], [0.6174, 0.3917, 0.7474, 0.2506, 0.4010, 0.2601, 0.5830, 0.5165]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767], [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.6317, 0.4038, 0.8288, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986], [0.6084, 0.4008, 0.8587, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283], [0.6214, 0.4112, 0.7837, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083], [0.0000, 0.0000, 0.7525, 0.2291, 0.3837, 0.3017, 0.6050, 0.5667], [0.6163, 0.4114, 0.7650, 0.2017, 0.3762, 0.2867, 0.5631, 0.5071]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.06759074423462152 step: 30 running loss: 0.0022530248078207176 Train Steps: 30/90 Loss: 0.0023 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378], [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118], [0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5563, 0.5533], [0.6275, 0.4157, 0.8337, 0.5800, 0.3763, 0.4200, 0.5547, 0.6125], [0.6246, 0.4028, 0.8738, 0.4867, 0.4088, 0.5667, 0.6362, 0.5200], [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617], [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350], [0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5891, 0.4129, 0.8819, 0.1990, 0.5162, 0.2382, 0.7617, 0.5226], [0.5691, 0.4271, 0.7440, 0.2872, 0.3934, 0.2544, 0.5267, 0.5962], [0.5291, 0.3691, 0.8776, 0.4669, 0.3964, 0.4697, 0.5507, 0.5498], [0.5685, 0.3836, 0.8484, 0.5101, 0.3768, 0.4332, 0.5894, 0.5859], [0.5461, 0.3610, 0.8937, 0.4030, 0.4117, 0.5739, 0.6300, 0.5297], [0.6028, 0.4109, 0.8958, 0.4430, 0.4658, 0.4558, 0.5568, 0.5362], [0.5917, 0.3764, 0.8950, 0.4554, 0.3693, 0.4597, 0.5308, 0.4896], [0.5577, 0.3902, 0.8747, 0.4571, 0.4388, 0.5051, 0.5908, 0.5223]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6293, 0.4097, 0.8800, 0.2517, 0.5263, 0.2600, 0.7430, 0.5378], [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118], [0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5562, 0.5533], [0.6275, 0.4157, 0.8338, 0.5800, 0.3762, 0.4200, 0.5547, 0.6125], [0.6246, 0.4028, 0.8737, 0.4867, 0.4087, 0.5667, 0.6363, 0.5200], [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617], [0.6143, 0.3982, 0.8650, 0.4883, 0.3913, 0.4317, 0.5315, 0.5350], [0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.06881934218108654 step: 31 running loss: 0.0022199787800350496 Train Steps: 31/90 Loss: 0.0022 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6193, 0.4108, 0.7438, 0.2700, 0.3650, 0.3683, 0.6238, 0.5717], [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719], [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633], [0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131], [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600], [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183], [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6275, 0.4265, 0.7716, 0.2781, 0.3984, 0.3497, 0.6382, 0.5523], [0.5203, 0.3401, 0.8799, 0.4272, 0.3907, 0.4228, 0.5961, 0.5414], [0.5463, 0.3887, 0.9214, 0.4432, 0.4414, 0.5566, 0.6011, 0.5751], [0.6827, 0.4398, 0.8401, 0.2632, 0.4333, 0.2702, 0.5930, 0.5047], [0.6210, 0.4231, 0.8930, 0.4080, 0.3893, 0.3765, 0.5890, 0.5142], [0.1107, 0.0948, 0.7827, 0.2740, 0.4395, 0.2777, 0.5472, 0.5458], [0.5985, 0.4182, 0.7871, 0.2891, 0.4246, 0.3038, 0.5919, 0.6043], [0.6370, 0.4413, 0.7423, 0.2375, 0.4879, 0.1454, 0.5941, 0.5261]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6193, 0.4108, 0.7437, 0.2700, 0.3650, 0.3683, 0.6237, 0.5717], [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719], [0.6189, 0.4049, 0.8888, 0.4417, 0.4212, 0.5200, 0.5987, 0.5633], [0.6205, 0.4015, 0.8350, 0.2717, 0.3988, 0.2550, 0.5788, 0.5133], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131], [0.0000, 0.0000, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600], [0.6196, 0.4094, 0.7563, 0.2817, 0.3938, 0.3183, 0.6012, 0.6183], [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.06997097912244499 step: 32 running loss: 0.002186593097576406 Train Steps: 32/90 Loss: 0.0022 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400], [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050], [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050], [0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426], [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823], [0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6632, 0.4778, 0.8182, 0.1933, 0.4617, 0.1867, 0.6462, 0.5282], [0.5586, 0.3796, 0.8881, 0.3963, 0.3744, 0.3900, 0.5972, 0.5318], [0.5068, 0.3652, 0.8657, 0.4687, 0.4695, 0.4946, 0.6065, 0.5649], [0.5647, 0.3709, 0.8731, 0.3147, 0.4259, 0.2559, 0.6572, 0.5229], [0.5499, 0.3703, 0.8740, 0.4694, 0.4646, 0.4603, 0.5615, 0.5666], [0.5191, 0.3648, 0.8962, 0.3705, 0.3918, 0.3294, 0.5904, 0.5462], [0.5249, 0.3715, 0.7566, 0.3213, 0.3711, 0.3640, 0.5613, 0.6067], [0.5703, 0.3967, 0.8596, 0.4826, 0.4569, 0.4958, 0.5331, 0.5503]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400], [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050], [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050], [0.6277, 0.4036, 0.8687, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6126, 0.3954, 0.8537, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426], [0.6170, 0.4102, 0.7468, 0.3695, 0.3462, 0.3767, 0.5238, 0.5823], [0.6097, 0.3988, 0.8650, 0.5250, 0.4212, 0.5200, 0.5675, 0.5050]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.0716163640609011 step: 33 running loss: 0.0021701928503303366 Train Steps: 33/90 Loss: 0.0022 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6200, 0.4049, 0.8638, 0.5617, 0.4125, 0.5100, 0.6013, 0.5317], [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6197, 0.4118, 0.8688, 0.5517, 0.4037, 0.5233, 0.5875, 0.5600], [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5498, 0.3801, 0.8887, 0.4040, 0.3866, 0.3922, 0.5649, 0.5455], [0.5553, 0.3956, 0.8445, 0.2139, 0.4880, 0.1568, 0.6218, 0.5161], [0.5328, 0.3725, 0.8469, 0.5384, 0.4104, 0.4539, 0.5756, 0.5725], [0.6221, 0.4063, 0.8781, 0.4012, 0.4715, 0.5037, 0.6208, 0.5813], [0.5698, 0.3907, 0.8111, 0.2811, 0.3746, 0.3330, 0.5808, 0.5588], [0.5970, 0.4106, 0.8380, 0.5187, 0.4095, 0.5040, 0.5817, 0.5924], [0.6124, 0.4144, 0.8904, 0.4044, 0.3616, 0.4242, 0.6757, 0.5401], [0.5840, 0.3941, 0.8406, 0.4465, 0.4335, 0.4914, 0.6647, 0.5600]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6199, 0.4049, 0.8637, 0.5617, 0.4125, 0.5100, 0.6012, 0.5317], [0.6184, 0.4079, 0.8838, 0.4617, 0.4837, 0.5650, 0.6175, 0.5850], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6197, 0.4118, 0.8687, 0.5517, 0.4038, 0.5233, 0.5875, 0.5600], [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.07260472874622792 step: 34 running loss: 0.0021354331984184682 Train Steps: 34/90 Loss: 0.0021 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533], [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.6126, 0.4039, 0.8237, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138], [0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700], [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533], [0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5184, 0.3429, 0.7628, 0.2571, 0.4802, 0.1860, 0.6074, 0.5515], [0.0977, 0.0458, 0.8108, 0.3233, 0.3749, 0.3155, 0.5274, 0.5635], [0.6940, 0.4294, 0.9553, 0.4917, 0.3541, 0.5029, 0.6596, 0.4610], [0.5996, 0.3737, 0.8759, 0.3087, 0.5299, 0.2640, 0.7339, 0.5147], [0.6866, 0.4262, 0.8580, 0.4324, 0.3571, 0.4155, 0.6056, 0.5862], [0.7120, 0.4739, 0.8066, 0.3997, 0.3782, 0.3399, 0.5830, 0.5756], [0.6466, 0.4199, 0.7163, 0.2580, 0.3951, 0.2921, 0.5666, 0.5387], [0.7154, 0.4815, 0.7369, 0.2808, 0.4385, 0.2124, 0.5525, 0.5242]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6245, 0.4100, 0.7763, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.0000, 0.0000, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533], [0.6263, 0.4038, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.6126, 0.4038, 0.8238, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138], [0.6212, 0.4171, 0.7875, 0.3633, 0.3812, 0.2933, 0.5675, 0.5700], [0.6114, 0.4018, 0.7212, 0.1967, 0.3762, 0.2700, 0.5875, 0.5533], [0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263]]], device='cuda:0') loss_train_step before backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train: 0.07426053285598755 step: 35 running loss: 0.0021217295101710726 Train Steps: 35/90 Loss: 0.0021 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6274, 0.4003, 0.8638, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661], [0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983], [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621], [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344], [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892], [0.6087, 0.3951, 0.8387, 0.5833, 0.4188, 0.4933, 0.5146, 0.4830], [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6699, 0.4275, 0.8024, 0.5265, 0.3900, 0.4471, 0.6596, 0.5230], [0.6794, 0.4385, 0.8418, 0.4789, 0.4136, 0.4619, 0.6039, 0.4783], [0.1114, 0.0595, 0.8397, 0.2376, 0.5027, 0.1908, 0.6890, 0.5289], [0.6909, 0.4345, 0.8839, 0.4388, 0.3304, 0.3869, 0.5642, 0.5191], [0.6291, 0.3867, 0.8446, 0.4202, 0.3504, 0.3528, 0.5102, 0.5479], [0.6586, 0.4093, 0.8303, 0.5168, 0.3792, 0.4555, 0.5674, 0.5253], [0.7494, 0.4655, 0.8382, 0.2952, 0.3746, 0.3281, 0.6696, 0.5156], [0.7090, 0.4590, 0.8363, 0.5264, 0.4275, 0.4275, 0.5747, 0.5815]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6274, 0.4003, 0.8637, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661], [0.6219, 0.3934, 0.8687, 0.5267, 0.4313, 0.4967, 0.5987, 0.4983], [0.0000, 0.0000, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621], [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344], [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892], [0.6087, 0.3951, 0.8388, 0.5833, 0.4187, 0.4933, 0.5146, 0.4830], [0.6255, 0.4017, 0.8687, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6248, 0.4185, 0.8500, 0.5767, 0.4462, 0.4550, 0.5612, 0.5917]]], device='cuda:0') loss_train_step before backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train: 0.07608700543642044 step: 36 running loss: 0.002113527928789457 Train Steps: 36/90 Loss: 0.0021 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6289, 0.4081, 0.8720, 0.3487, 0.3900, 0.3183, 0.6703, 0.5376], [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773], [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5988, 0.5283], [0.6157, 0.4102, 0.8513, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890], [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249], [0.6259, 0.4133, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6112, 0.3707, 0.7490, 0.3077, 0.3605, 0.2968, 0.5415, 0.5408], [0.5955, 0.3761, 0.8708, 0.3528, 0.3607, 0.2691, 0.6453, 0.5391], [0.5341, 0.3475, 0.8463, 0.5252, 0.4797, 0.4976, 0.5357, 0.5157], [0.5937, 0.3833, 0.8467, 0.5804, 0.3934, 0.4725, 0.5941, 0.5159], [0.5596, 0.3415, 0.8215, 0.3929, 0.3491, 0.3478, 0.5140, 0.5521], [0.5897, 0.3710, 0.8079, 0.2786, 0.3989, 0.2553, 0.6312, 0.4956], [0.6347, 0.4102, 0.8645, 0.5367, 0.3555, 0.4859, 0.6540, 0.5185], [0.5740, 0.3737, 0.8167, 0.2516, 0.4687, 0.1609, 0.6137, 0.5002]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6289, 0.4081, 0.8720, 0.3487, 0.3900, 0.3183, 0.6703, 0.5376], [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773], [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5987, 0.5283], [0.6157, 0.4102, 0.8512, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890], [0.6257, 0.4034, 0.8288, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249], [0.6259, 0.4132, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.07694085594266653 step: 37 running loss: 0.0020794825930450416 Train Steps: 37/90 Loss: 0.0021 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986], [0.6111, 0.4033, 0.8300, 0.3267, 0.3588, 0.3333, 0.5444, 0.5637], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283], [ nan, nan, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.5779, 0.3774, 0.8839, 0.4152, 0.4410, 0.2534, 0.5820, 0.5221], [ 0.6797, 0.4455, 0.8260, 0.3675, 0.3468, 0.3362, 0.5157, 0.5417], [ 0.6687, 0.4174, 0.8940, 0.4060, 0.3847, 0.2705, 0.5886, 0.5112], [-0.0912, -0.0921, 0.6975, 0.2365, 0.4422, 0.1867, 0.5306, 0.5320], [ 0.6604, 0.4093, 0.8252, 0.6504, 0.3691, 0.5073, 0.6017, 0.4858], [ 0.6601, 0.4308, 0.6697, 0.2957, 0.3913, 0.2468, 0.5048, 0.5545], [ 0.7503, 0.4870, 0.8212, 0.3324, 0.4210, 0.2805, 0.6182, 0.4864], [ 0.6137, 0.3834, 0.8276, 0.2994, 0.4318, 0.2749, 0.6676, 0.5525]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986], [0.6111, 0.4033, 0.8300, 0.3267, 0.3587, 0.3333, 0.5444, 0.5637], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6187, 0.5283], [0.0000, 0.0000, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5362, 0.5550], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.07825841219164431 step: 38 running loss: 0.0020594318997801133 Train Steps: 38/90 Loss: 0.0021 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6275, 0.4050, 0.9038, 0.3767, 0.3838, 0.3533, 0.7074, 0.5575], [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332], [0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297], [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850], [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5947, 0.3805, 0.8456, 0.4897, 0.4121, 0.4539, 0.5597, 0.4984], [0.3858, 0.2530, 0.7886, 0.3215, 0.3442, 0.2716, 0.4741, 0.5347], [0.5989, 0.3812, 0.8558, 0.5530, 0.3637, 0.4137, 0.6011, 0.5037], [0.6297, 0.4055, 0.8894, 0.3635, 0.3809, 0.3036, 0.6904, 0.5203], [0.6403, 0.4136, 0.8121, 0.4900, 0.3675, 0.4450, 0.6480, 0.5108], [0.5778, 0.3763, 0.8999, 0.4514, 0.4168, 0.2672, 0.6207, 0.5142], [0.6699, 0.4190, 0.8753, 0.4437, 0.4606, 0.5210, 0.6239, 0.5325], [0.5892, 0.3982, 0.8542, 0.4954, 0.4509, 0.4645, 0.5348, 0.5564]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6275, 0.4050, 0.9038, 0.3767, 0.3837, 0.3533, 0.7074, 0.5575], [0.6289, 0.4019, 0.8112, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332], [0.6364, 0.4165, 0.9087, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297], [0.6184, 0.4079, 0.8838, 0.4617, 0.4837, 0.5650, 0.6175, 0.5850], [0.6140, 0.4070, 0.8700, 0.5000, 0.4613, 0.4900, 0.5260, 0.5852]]], device='cuda:0') loss_train_step before backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train: 0.08022535732015967 step: 39 running loss: 0.0020570604441066584 Train Steps: 39/90 Loss: 0.0021 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791], [ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563], [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397], [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038], [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5837, 0.5583], [0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6195, 0.4118, 0.9066, 0.4095, 0.3811, 0.2271, 0.6301, 0.5351], [ 0.6931, 0.4617, 0.8825, 0.5103, 0.3938, 0.5106, 0.6760, 0.4836], [-0.0069, -0.0022, 0.7917, 0.3328, 0.3496, 0.2844, 0.5340, 0.5845], [ 0.5912, 0.3992, 0.6751, 0.2480, 0.3963, 0.1762, 0.5067, 0.5832], [ 0.6294, 0.4174, 0.8644, 0.5082, 0.4266, 0.4852, 0.5873, 0.5055], [ 0.7166, 0.4977, 0.8992, 0.4628, 0.4537, 0.5101, 0.6015, 0.5534], [ 0.4912, 0.3237, 0.8256, 0.3268, 0.4155, 0.2062, 0.5291, 0.5372], [ 0.7153, 0.4560, 0.8673, 0.5064, 0.4377, 0.4808, 0.5551, 0.5160]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6277, 0.4117, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791], [0.0000, 0.0000, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563], [0.6159, 0.4085, 0.6900, 0.2283, 0.4087, 0.1950, 0.5123, 0.5397], [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038], [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5838, 0.5583], [0.6143, 0.4040, 0.8238, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.08172035915777087 step: 40 running loss: 0.002043008978944272 Train Steps: 40/90 Loss: 0.0020 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5837, 0.5583], [0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683], [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567], [0.6274, 0.4003, 0.8638, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661], [0.6125, 0.3974, 0.7725, 0.2517, 0.3538, 0.3317, 0.5887, 0.5500], [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726], [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6063, 0.6083]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5758, 0.3786, 0.8423, 0.3171, 0.4899, 0.2170, 0.5905, 0.5425], [0.5740, 0.3879, 0.9308, 0.4643, 0.4837, 0.5307, 0.5876, 0.5373], [0.6391, 0.4176, 0.8629, 0.3083, 0.4227, 0.3055, 0.6264, 0.5473], [0.5625, 0.3663, 0.8680, 0.4720, 0.3909, 0.4157, 0.5760, 0.5533], [0.5354, 0.3332, 0.8488, 0.6208, 0.4072, 0.4879, 0.6330, 0.5155], [0.5616, 0.3440, 0.7892, 0.2749, 0.3800, 0.3443, 0.5898, 0.5582], [0.6584, 0.4165, 0.7801, 0.2906, 0.4001, 0.2840, 0.5664, 0.5037], [0.5920, 0.4034, 0.8760, 0.4120, 0.3817, 0.4544, 0.5925, 0.5546]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6229, 0.4107, 0.8138, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5838, 0.5583], [0.6200, 0.4098, 0.8238, 0.2917, 0.4013, 0.2967, 0.6000, 0.5683], [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567], [0.6274, 0.4003, 0.8637, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661], [0.6125, 0.3974, 0.7725, 0.2517, 0.3537, 0.3317, 0.5888, 0.5500], [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726], [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6062, 0.6083]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.08272404491435736 step: 41 running loss: 0.002017659632057497 Train Steps: 41/90 Loss: 0.0020 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6084, 0.3981, 0.8588, 0.5233, 0.4600, 0.5367, 0.5680, 0.5006], [0.6200, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6179, 0.4008, 0.8600, 0.4015, 0.3932, 0.2515, 0.5711, 0.5438], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290], [0.6275, 0.4157, 0.8337, 0.5800, 0.3763, 0.4200, 0.5547, 0.6125], [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5963, 0.6217]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5543, 0.3742, 0.8943, 0.5196, 0.4794, 0.5638, 0.5621, 0.4876], [0.6037, 0.3794, 0.9385, 0.4418, 0.4021, 0.4450, 0.5992, 0.5035], [0.6467, 0.4319, 0.8756, 0.3346, 0.4146, 0.2846, 0.5714, 0.5388], [0.5918, 0.3820, 0.8937, 0.4119, 0.3816, 0.4555, 0.5682, 0.5513], [0.6275, 0.4428, 0.8985, 0.5641, 0.4904, 0.4356, 0.6108, 0.5810], [0.5539, 0.3666, 0.8488, 0.2932, 0.3606, 0.4177, 0.6532, 0.5217], [0.6402, 0.4327, 0.8611, 0.5682, 0.3874, 0.4667, 0.5907, 0.5768], [0.5660, 0.3770, 0.7327, 0.2319, 0.4440, 0.2731, 0.5796, 0.6056]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6084, 0.3981, 0.8587, 0.5233, 0.4600, 0.5367, 0.5680, 0.5006], [0.6201, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6179, 0.4008, 0.8600, 0.4015, 0.3932, 0.2515, 0.5711, 0.5438], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290], [0.6275, 0.4157, 0.8338, 0.5800, 0.3762, 0.4200, 0.5547, 0.6125], [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5962, 0.6217]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.08353073458420113 step: 42 running loss: 0.0019888270139095505 Train Steps: 42/90 Loss: 0.0020 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954], [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6188, 0.5433], [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689], [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395], [0.6293, 0.3982, 0.8700, 0.5300, 0.3763, 0.4717, 0.7050, 0.5297], [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367], [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114], [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6746, 0.4319, 0.8910, 0.3957, 0.4231, 0.3041, 0.4968, 0.5360], [0.6110, 0.4048, 0.9118, 0.4690, 0.4132, 0.5844, 0.6055, 0.5478], [0.5277, 0.3620, 0.8001, 0.3551, 0.3683, 0.4127, 0.5339, 0.5777], [0.6432, 0.4046, 0.7831, 0.2371, 0.4205, 0.2584, 0.5683, 0.5560], [0.5422, 0.3705, 0.8758, 0.5354, 0.3853, 0.5174, 0.6447, 0.5465], [0.6417, 0.4331, 0.8154, 0.2463, 0.4465, 0.2785, 0.5989, 0.5654], [0.5830, 0.4069, 0.8817, 0.5367, 0.4828, 0.5881, 0.5900, 0.5317], [0.6231, 0.4210, 0.9309, 0.4411, 0.3923, 0.4972, 0.6554, 0.5362]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6137, 0.4038, 0.8562, 0.4050, 0.3812, 0.2550, 0.5106, 0.4954], [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6187, 0.5433], [0.6102, 0.4001, 0.7738, 0.3583, 0.3462, 0.3800, 0.5524, 0.5689], [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395], [0.6293, 0.3982, 0.8700, 0.5300, 0.3762, 0.4717, 0.7050, 0.5297], [0.6204, 0.4110, 0.7912, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367], [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114], [0.6263, 0.4065, 0.9038, 0.4317, 0.3587, 0.4550, 0.6325, 0.5250]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.08457283378811553 step: 43 running loss: 0.0019668100880957102 Train Steps: 43/90 Loss: 0.0020 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136], [0.6272, 0.4045, 0.8538, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6246, 0.4008, 0.8757, 0.5088, 0.4101, 0.5392, 0.6644, 0.5133], [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167], [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050], [0.6201, 0.4151, 0.8588, 0.5467, 0.3700, 0.3950, 0.5637, 0.5933]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5829, 0.3921, 0.8264, 0.3159, 0.3731, 0.5240, 0.5861, 0.5414], [0.6574, 0.4408, 0.8602, 0.5335, 0.3847, 0.4092, 0.5730, 0.5130], [0.6363, 0.4418, 0.8349, 0.5432, 0.3740, 0.5134, 0.6518, 0.5425], [0.5697, 0.3814, 0.8596, 0.4887, 0.4200, 0.5548, 0.6287, 0.5410], [0.6425, 0.4493, 0.8188, 0.2990, 0.3808, 0.4217, 0.5833, 0.5963], [0.6742, 0.4422, 0.8950, 0.3099, 0.3901, 0.3213, 0.6032, 0.5729], [0.6124, 0.4264, 0.8670, 0.3953, 0.3919, 0.5078, 0.5886, 0.5685], [0.5960, 0.4155, 0.8576, 0.5290, 0.3820, 0.4110, 0.5455, 0.5905]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6132, 0.4118, 0.8200, 0.3633, 0.3562, 0.5400, 0.5787, 0.5136], [0.6271, 0.4045, 0.8537, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6246, 0.4008, 0.8757, 0.5088, 0.4101, 0.5392, 0.6644, 0.5133], [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6037, 0.6167], [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6200, 0.4112, 0.8863, 0.4100, 0.3638, 0.4917, 0.6087, 0.6050], [0.6202, 0.4151, 0.8587, 0.5467, 0.3700, 0.3950, 0.5638, 0.5933]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.08523785881698132 step: 44 running loss: 0.0019372240640223026 Train Steps: 44/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6133, 0.4066, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358], [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6138, 0.5450], [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312], [0.6125, 0.3974, 0.7725, 0.2517, 0.3538, 0.3317, 0.5887, 0.5500], [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133], [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427], [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320], [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5574, 0.3708, 0.7304, 0.2770, 0.3952, 0.2549, 0.5315, 0.5325], [0.5851, 0.4072, 0.8954, 0.4923, 0.4134, 0.6094, 0.6331, 0.5524], [0.7421, 0.4717, 0.8658, 0.4165, 0.3477, 0.4763, 0.6099, 0.5323], [0.6102, 0.3866, 0.7828, 0.2630, 0.3653, 0.3681, 0.5750, 0.5749], [0.6527, 0.4293, 0.8477, 0.2623, 0.4776, 0.2003, 0.6449, 0.5496], [0.7032, 0.4611, 0.8211, 0.4001, 0.3383, 0.4056, 0.5390, 0.5727], [0.6002, 0.3951, 0.8682, 0.5551, 0.3725, 0.5488, 0.6113, 0.5342], [0.5485, 0.3695, 0.8289, 0.3966, 0.3556, 0.4300, 0.5648, 0.5475]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6133, 0.4065, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358], [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6137, 0.5450], [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312], [0.6125, 0.3974, 0.7725, 0.2517, 0.3537, 0.3317, 0.5888, 0.5500], [0.6266, 0.4065, 0.8313, 0.2467, 0.4787, 0.1733, 0.6313, 0.5133], [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427], [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320], [0.6058, 0.3978, 0.8288, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.08670229942072183 step: 45 running loss: 0.0019267177649049294 Train Steps: 45/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6312, 0.5250], [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519], [0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153], [0.6250, 0.4103, 0.8950, 0.4400, 0.3912, 0.5650, 0.6050, 0.5133], [ nan, nan, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983], [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6801, 0.4410, 0.8683, 0.5659, 0.3392, 0.4688, 0.5803, 0.5495], [0.6600, 0.4212, 0.8819, 0.3557, 0.3762, 0.3373, 0.6782, 0.5341], [0.6769, 0.4397, 0.8591, 0.5386, 0.4188, 0.5211, 0.5393, 0.5377], [0.6893, 0.4491, 0.8465, 0.5392, 0.4218, 0.5473, 0.6143, 0.5291], [0.7045, 0.4556, 0.8790, 0.4567, 0.3647, 0.5870, 0.6346, 0.5346], [0.2349, 0.1266, 0.7190, 0.2019, 0.4143, 0.1636, 0.5454, 0.5321], [0.6809, 0.4381, 0.6721, 0.2559, 0.3333, 0.2282, 0.5364, 0.5570], [0.6265, 0.4215, 0.8322, 0.3383, 0.3416, 0.3138, 0.5845, 0.5514]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6222, 0.4171, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6313, 0.5250], [0.6100, 0.4016, 0.8600, 0.5067, 0.4613, 0.5233, 0.5086, 0.5519], [0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153], [0.6250, 0.4103, 0.8950, 0.4400, 0.3913, 0.5650, 0.6050, 0.5133], [0.0000, 0.0000, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983], [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train: 0.08878360188100487 step: 46 running loss: 0.0019300783017609754 Train Steps: 46/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164], [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459], [ nan, nan, 0.6793, 0.2110, 0.4012, 0.2167, 0.5112, 0.5583], [0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6164, 0.4119, 0.7913, 0.2650, 0.3538, 0.3500, 0.5614, 0.5038], [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578], [0.6277, 0.4083, 0.8350, 0.2717, 0.4562, 0.1800, 0.5918, 0.4878]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7055, 0.4578, 0.8519, 0.5879, 0.3761, 0.3916, 0.5511, 0.5944], [0.7292, 0.4616, 0.8460, 0.5709, 0.3622, 0.4798, 0.6770, 0.4983], [0.0664, 0.0312, 0.7004, 0.2303, 0.4082, 0.1988, 0.5215, 0.5506], [0.6563, 0.4132, 0.8240, 0.3713, 0.3363, 0.5313, 0.5989, 0.5189], [0.6488, 0.4153, 0.8015, 0.3351, 0.3190, 0.3517, 0.6047, 0.5187], [0.7151, 0.4625, 0.7876, 0.2750, 0.3320, 0.3584, 0.5924, 0.4999], [0.6842, 0.4461, 0.6953, 0.2469, 0.3832, 0.1930, 0.5175, 0.5420], [0.6422, 0.3944, 0.8310, 0.2808, 0.4452, 0.2402, 0.5940, 0.5098]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164], [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459], [0.0000, 0.0000, 0.6793, 0.2110, 0.4013, 0.2167, 0.5113, 0.5583], [0.6132, 0.4118, 0.8200, 0.3633, 0.3562, 0.5400, 0.5787, 0.5136], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6164, 0.4119, 0.7912, 0.2650, 0.3537, 0.3500, 0.5614, 0.5038], [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578], [0.6277, 0.4083, 0.8350, 0.2717, 0.4563, 0.1800, 0.5918, 0.4878]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.09005057823378593 step: 47 running loss: 0.0019159697496550197 Train Steps: 47/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150], [0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110], [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167], [0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617], [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791], [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350], [0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550], [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6486, 0.4020, 0.7514, 0.2730, 0.4087, 0.2175, 0.5757, 0.5366], [0.6411, 0.4125, 0.8790, 0.4814, 0.3593, 0.5086, 0.5643, 0.5240], [0.6816, 0.4637, 0.7844, 0.3338, 0.3328, 0.2747, 0.5542, 0.5338], [0.6323, 0.3974, 0.7978, 0.5710, 0.4151, 0.4028, 0.5774, 0.5467], [0.5792, 0.3522, 0.8565, 0.4896, 0.3740, 0.4974, 0.6258, 0.4819], [0.6051, 0.3964, 0.8591, 0.4106, 0.3487, 0.3718, 0.5399, 0.5469], [0.5822, 0.3830, 0.7570, 0.2944, 0.3506, 0.4014, 0.6029, 0.5421], [0.3315, 0.2145, 0.6800, 0.2316, 0.4172, 0.1543, 0.5413, 0.5392]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150], [0.6186, 0.4129, 0.8900, 0.4567, 0.3938, 0.5417, 0.5734, 0.5110], [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167], [0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617], [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791], [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350], [0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550], [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533]]], device='cuda:0') loss_train_step before backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train: 0.09267542359884828 step: 48 running loss: 0.0019307379916426726 Train Steps: 48/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197], [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6030, 0.3969, 0.7988, 0.3917, 0.3450, 0.3667, 0.5266, 0.4700], [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633], [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750], [0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6052, 0.3889, 0.7422, 0.3402, 0.3384, 0.3641, 0.5236, 0.5260], [0.6399, 0.4004, 0.7969, 0.3118, 0.4094, 0.1803, 0.5158, 0.5109], [0.5728, 0.3562, 0.8510, 0.4912, 0.4429, 0.4912, 0.5814, 0.5052], [0.5621, 0.3503, 0.8764, 0.4556, 0.4044, 0.5215, 0.6876, 0.5352], [0.6308, 0.4028, 0.8012, 0.3729, 0.3408, 0.3155, 0.5477, 0.5014], [0.6521, 0.4160, 0.8504, 0.5561, 0.4178, 0.4928, 0.5914, 0.5374], [0.5205, 0.3282, 0.6716, 0.2172, 0.4256, 0.1459, 0.5180, 0.5464], [0.6104, 0.3535, 0.8501, 0.5325, 0.3972, 0.4680, 0.6415, 0.4927]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6143, 0.4040, 0.8238, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197], [0.6339, 0.4102, 0.9087, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6030, 0.3969, 0.7987, 0.3917, 0.3450, 0.3667, 0.5266, 0.4700], [0.6192, 0.4128, 0.8512, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633], [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5512, 0.5750], [0.6308, 0.3990, 0.8687, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.09376236831303686 step: 49 running loss: 0.0019135177206742217 Train Steps: 49/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6147, 0.4112, 0.7988, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550], [0.6265, 0.4088, 0.8025, 0.1850, 0.4163, 0.2500, 0.6290, 0.4947], [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6403, 0.4320, 0.7649, 0.3199, 0.3983, 0.2335, 0.4917, 0.5320], [0.5692, 0.3730, 0.7705, 0.2126, 0.4442, 0.1903, 0.5995, 0.5071], [0.5403, 0.3438, 0.8224, 0.4247, 0.4037, 0.4621, 0.5542, 0.4882], [0.5514, 0.3663, 0.8637, 0.4957, 0.3862, 0.3813, 0.6040, 0.5017], [0.6259, 0.3877, 0.8745, 0.4758, 0.3988, 0.3611, 0.6208, 0.5046], [0.5511, 0.3389, 0.8570, 0.5071, 0.4225, 0.5230, 0.6351, 0.5395], [0.5870, 0.3975, 0.7331, 0.3451, 0.3606, 0.3595, 0.5102, 0.5234], [0.6117, 0.3787, 0.8476, 0.4176, 0.4175, 0.4941, 0.5977, 0.5509]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6147, 0.4112, 0.7987, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550], [0.6265, 0.4088, 0.8025, 0.1850, 0.4162, 0.2500, 0.6290, 0.4947], [0.6068, 0.3963, 0.8650, 0.4317, 0.4038, 0.5083, 0.5253, 0.4999], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6199, 0.4102, 0.8950, 0.4417, 0.4013, 0.5367, 0.6112, 0.5967]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.09513336268719286 step: 50 running loss: 0.0019026672537438571 Train Steps: 50/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207], [0.6198, 0.4101, 0.8838, 0.5283, 0.3763, 0.5267, 0.5913, 0.5567], [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6339, 0.4118, 0.7988, 0.5800, 0.3912, 0.4583, 0.7343, 0.5760], [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385], [0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5637, 0.5083], [0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5291, 0.3672, 0.7294, 0.2730, 0.3650, 0.3765, 0.5397, 0.5277], [0.5886, 0.3764, 0.8469, 0.5366, 0.3838, 0.5122, 0.5506, 0.5424], [0.5200, 0.3578, 0.8842, 0.4014, 0.3746, 0.3781, 0.5141, 0.5359], [0.5578, 0.3773, 0.7889, 0.2673, 0.4389, 0.2098, 0.5802, 0.5160], [0.5594, 0.3759, 0.8057, 0.5207, 0.3739, 0.4169, 0.6299, 0.5353], [0.6054, 0.4064, 0.7891, 0.2735, 0.4530, 0.2321, 0.6273, 0.5279], [0.6244, 0.3972, 0.8628, 0.5094, 0.3892, 0.4647, 0.5318, 0.5040], [0.5981, 0.3899, 0.8339, 0.2679, 0.5203, 0.2410, 0.6810, 0.5113]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207], [0.6198, 0.4101, 0.8838, 0.5283, 0.3762, 0.5267, 0.5913, 0.5567], [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6339, 0.4118, 0.7987, 0.5800, 0.3913, 0.4583, 0.7343, 0.5760], [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385], [0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5638, 0.5083], [0.6293, 0.4097, 0.8800, 0.2517, 0.5263, 0.2600, 0.7430, 0.5378]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.09657937183510512 step: 51 running loss: 0.0018937131732373552 Train Steps: 51/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367], [0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300], [0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019], [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417], [0.6353, 0.4128, 0.9138, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667], [0.6030, 0.3969, 0.7988, 0.3917, 0.3450, 0.3667, 0.5266, 0.4700]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6311, 0.4221, 0.8473, 0.2995, 0.4195, 0.2622, 0.6039, 0.5202], [0.7261, 0.5016, 0.7322, 0.2343, 0.3933, 0.3053, 0.5926, 0.5256], [0.6551, 0.4545, 0.8713, 0.4405, 0.3823, 0.5575, 0.5866, 0.5082], [0.6100, 0.4321, 0.7779, 0.2898, 0.4315, 0.2631, 0.6073, 0.5497], [0.4372, 0.2929, 0.9085, 0.3609, 0.4803, 0.3184, 0.6927, 0.5634], [0.6183, 0.4245, 0.8867, 0.5138, 0.3870, 0.4361, 0.6212, 0.5193], [0.1437, 0.1349, 0.7505, 0.2220, 0.3988, 0.2872, 0.5788, 0.5707], [0.6677, 0.4697, 0.8175, 0.4028, 0.3731, 0.3748, 0.5594, 0.5090]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367], [0.6207, 0.4081, 0.7663, 0.2067, 0.3963, 0.3200, 0.6313, 0.5300], [0.6161, 0.4099, 0.8737, 0.4383, 0.3787, 0.5483, 0.5605, 0.5019], [0.6243, 0.4128, 0.7763, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417], [0.6353, 0.4128, 0.9137, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.0000, 0.0000, 0.7525, 0.2291, 0.3837, 0.3017, 0.6050, 0.5667], [0.6030, 0.3969, 0.7987, 0.3917, 0.3450, 0.3667, 0.5266, 0.4700]]], device='cuda:0') loss_train_step before backward: tensor(0.0023, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0023, device='cuda:0', grad_fn=) loss_train: 0.0988936327630654 step: 52 running loss: 0.00190180063005895 Train Steps: 52/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6202, 0.4079, 0.8025, 0.2500, 0.3763, 0.3217, 0.6125, 0.5533], [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400], [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200], [0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683], [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332], [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367], [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274], [0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5353, 0.3823, 0.7938, 0.2598, 0.3850, 0.3384, 0.6041, 0.5397], [0.5435, 0.3924, 0.8487, 0.3284, 0.3678, 0.4744, 0.6533, 0.5549], [0.6516, 0.4569, 0.8807, 0.4752, 0.3808, 0.4926, 0.6206, 0.5455], [0.5597, 0.3985, 0.8237, 0.2700, 0.4093, 0.3204, 0.6297, 0.5742], [0.6297, 0.4217, 0.9052, 0.4837, 0.4001, 0.4403, 0.6905, 0.5351], [0.5590, 0.3859, 0.8605, 0.2797, 0.4072, 0.2892, 0.6317, 0.5315], [0.5734, 0.4131, 0.8632, 0.3005, 0.4720, 0.2418, 0.6259, 0.5304], [0.6172, 0.4364, 0.8233, 0.4241, 0.3716, 0.4831, 0.6026, 0.5529]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6202, 0.4079, 0.8025, 0.2500, 0.3762, 0.3217, 0.6125, 0.5533], [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400], [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200], [0.6200, 0.4098, 0.8238, 0.2917, 0.4013, 0.2967, 0.6000, 0.5683], [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332], [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367], [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274], [0.6091, 0.3997, 0.8314, 0.4334, 0.3787, 0.4550, 0.5213, 0.5656]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.09974715590942651 step: 53 running loss: 0.001882021809611821 Train Steps: 53/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6147, 0.4081, 0.8538, 0.3400, 0.3663, 0.3150, 0.5142, 0.4875], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6115, 0.4081, 0.6725, 0.2433, 0.4088, 0.1933, 0.5167, 0.5544], [0.6275, 0.4050, 0.9038, 0.3767, 0.3838, 0.3533, 0.7074, 0.5575], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600], [0.6293, 0.3982, 0.8700, 0.5300, 0.3763, 0.4717, 0.7050, 0.5297], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6275, 0.4306, 0.8911, 0.3299, 0.3758, 0.3486, 0.5929, 0.5505], [0.5658, 0.4011, 0.9082, 0.4369, 0.4359, 0.5329, 0.6209, 0.5843], [0.7188, 0.4968, 0.7221, 0.2397, 0.4145, 0.2572, 0.5564, 0.5696], [0.6036, 0.4003, 0.9182, 0.3324, 0.4068, 0.3903, 0.7308, 0.5581], [0.5868, 0.3989, 0.9414, 0.4409, 0.3893, 0.4048, 0.6734, 0.5355], [0.6512, 0.4482, 0.9158, 0.5223, 0.4085, 0.5194, 0.6330, 0.5632], [0.6254, 0.4223, 0.8987, 0.4838, 0.3649, 0.5039, 0.7165, 0.5501], [0.5276, 0.3468, 0.8923, 0.3737, 0.3886, 0.5233, 0.6415, 0.5414]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6147, 0.4081, 0.8537, 0.3400, 0.3663, 0.3150, 0.5142, 0.4875], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6115, 0.4081, 0.6725, 0.2433, 0.4087, 0.1933, 0.5167, 0.5544], [0.6275, 0.4050, 0.9038, 0.3767, 0.3837, 0.3533, 0.7074, 0.5575], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600], [0.6293, 0.3982, 0.8700, 0.5300, 0.3762, 0.4717, 0.7050, 0.5297], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.10123581916559488 step: 54 running loss: 0.0018747373919554607 Train Steps: 54/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.7192, 0.2346, 0.4037, 0.2050, 0.5138, 0.5650], [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6063, 0.6083], [ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567], [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167], [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6275, 0.4048, 0.8488, 0.2883, 0.4463, 0.2033, 0.6321, 0.5155], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.2724, 0.1738, 0.7532, 0.2123, 0.3912, 0.2749, 0.5379, 0.5518], [0.7120, 0.4843, 0.8970, 0.3871, 0.3452, 0.4813, 0.6276, 0.5857], [0.2392, 0.1629, 0.7932, 0.2384, 0.3899, 0.2904, 0.5817, 0.5716], [0.7815, 0.5277, 0.8028, 0.2982, 0.3764, 0.3326, 0.6129, 0.6341], [0.7014, 0.4486, 0.9404, 0.5730, 0.3902, 0.5327, 0.6732, 0.5283], [0.6658, 0.4433, 0.8476, 0.2791, 0.4124, 0.2763, 0.6565, 0.5462], [0.7444, 0.4732, 0.8730, 0.2678, 0.4320, 0.2968, 0.6825, 0.5032], [0.6798, 0.4357, 0.8951, 0.2959, 0.4816, 0.3000, 0.7251, 0.5187]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.7192, 0.2346, 0.4038, 0.2050, 0.5138, 0.5650], [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6062, 0.6083], [0.0000, 0.0000, 0.7425, 0.2117, 0.3938, 0.2433, 0.5437, 0.5567], [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167], [0.6334, 0.4037, 0.8637, 0.5733, 0.4013, 0.4717, 0.6369, 0.4938], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6275, 0.4048, 0.8487, 0.2883, 0.4462, 0.2033, 0.6321, 0.5155], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343]]], device='cuda:0') loss_train_step before backward: tensor(0.0053, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0053, device='cuda:0', grad_fn=) loss_train: 0.1065472379559651 step: 55 running loss: 0.0019372225082902746 Train Steps: 55/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882], [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600], [0.6289, 0.4024, 0.9088, 0.4567, 0.3937, 0.5633, 0.7058, 0.5609], [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332], [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200], [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170], [0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5691, 0.3698, 0.9165, 0.4752, 0.4648, 0.4836, 0.5714, 0.5409], [0.5765, 0.3781, 0.8772, 0.3890, 0.3343, 0.3800, 0.5563, 0.5370], [0.6562, 0.4239, 0.7415, 0.2279, 0.3311, 0.3317, 0.6071, 0.5846], [0.6207, 0.3919, 0.9110, 0.4470, 0.3838, 0.5880, 0.6730, 0.5643], [0.6225, 0.3718, 0.9335, 0.4698, 0.3647, 0.3998, 0.6734, 0.5408], [0.6486, 0.4084, 0.9166, 0.4715, 0.3560, 0.4715, 0.6107, 0.5510], [0.6746, 0.4125, 0.8632, 0.2376, 0.4625, 0.2011, 0.6717, 0.5324], [0.5819, 0.3607, 0.9126, 0.4763, 0.3888, 0.3742, 0.6958, 0.5375]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6128, 0.4022, 0.8737, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882], [0.6132, 0.4066, 0.7259, 0.2402, 0.3587, 0.3300, 0.6000, 0.5600], [0.6289, 0.4024, 0.9087, 0.4567, 0.3938, 0.5633, 0.7058, 0.5609], [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332], [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200], [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170], [0.6336, 0.4191, 0.8938, 0.5167, 0.3938, 0.3517, 0.7343, 0.5748]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.10731903073610738 step: 56 running loss: 0.0019164112631447747 Train Steps: 56/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459], [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533], [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116], [0.6111, 0.4033, 0.8300, 0.3267, 0.3588, 0.3333, 0.5444, 0.5637], [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5637, 0.5083]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6403, 0.3795, 0.8980, 0.5622, 0.3909, 0.4747, 0.7285, 0.5281], [0.6636, 0.4230, 0.7843, 0.2408, 0.3759, 0.2624, 0.5990, 0.5570], [0.4426, 0.2650, 0.7819, 0.2717, 0.4223, 0.2305, 0.5698, 0.5799], [0.6475, 0.3757, 0.8123, 0.2576, 0.3823, 0.2330, 0.6141, 0.4856], [0.5844, 0.3666, 0.9030, 0.3338, 0.3551, 0.3270, 0.5759, 0.5615], [0.5405, 0.3511, 0.7870, 0.2629, 0.4318, 0.2087, 0.6067, 0.5856], [0.6199, 0.3592, 0.9353, 0.5180, 0.4166, 0.4974, 0.6105, 0.5446], [0.6197, 0.3734, 0.9563, 0.5340, 0.3956, 0.4882, 0.6105, 0.5443]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459], [0.6114, 0.4018, 0.7212, 0.1967, 0.3762, 0.2700, 0.5875, 0.5533], [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6165, 0.4106, 0.7575, 0.1733, 0.3837, 0.2650, 0.5680, 0.5116], [0.6111, 0.4033, 0.8300, 0.3267, 0.3587, 0.3333, 0.5444, 0.5637], [0.6218, 0.4185, 0.7337, 0.2650, 0.4625, 0.1950, 0.5688, 0.5800], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5638, 0.5083]]], device='cuda:0') loss_train_step before backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train: 0.10952302097575739 step: 57 running loss: 0.0019214565083466209 Train Steps: 57/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283], [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841], [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6365, 0.3743, 0.8675, 0.4952, 0.4117, 0.5146, 0.6097, 0.5063], [0.5609, 0.3361, 0.8477, 0.4395, 0.3576, 0.3986, 0.4763, 0.5326], [0.6198, 0.3910, 0.8201, 0.2905, 0.4859, 0.1939, 0.6222, 0.5119], [0.5869, 0.3545, 0.8819, 0.3543, 0.3917, 0.2430, 0.6047, 0.5221], [0.5527, 0.3526, 0.8673, 0.5385, 0.3781, 0.4188, 0.5554, 0.5453], [0.5270, 0.3048, 0.7687, 0.2961, 0.3601, 0.3038, 0.5805, 0.5237], [0.6488, 0.4108, 0.8584, 0.4499, 0.4041, 0.3312, 0.6700, 0.5362], [0.6019, 0.3609, 0.8609, 0.4872, 0.4596, 0.4562, 0.5581, 0.5371]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6127, 0.4084, 0.8700, 0.4467, 0.3988, 0.4317, 0.5013, 0.5471], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6187, 0.5283], [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841], [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.11065401375526562 step: 58 running loss: 0.0019078278233666487 Train Steps: 58/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332], [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667], [0.6325, 0.4165, 0.9000, 0.4617, 0.3813, 0.4900, 0.7485, 0.5447]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6700, 0.4326, 0.8465, 0.3759, 0.4052, 0.2541, 0.5767, 0.4936], [0.6088, 0.3693, 0.8447, 0.4587, 0.3971, 0.4752, 0.5430, 0.5230], [0.6091, 0.3827, 0.7853, 0.5151, 0.3908, 0.4268, 0.6323, 0.5240], [0.6106, 0.3826, 0.8802, 0.4890, 0.3949, 0.3110, 0.5897, 0.5250], [0.6589, 0.4166, 0.8583, 0.4883, 0.4730, 0.5148, 0.5638, 0.5646], [0.6102, 0.3873, 0.8047, 0.5130, 0.3933, 0.4348, 0.6198, 0.5365], [0.5807, 0.3623, 0.8555, 0.3527, 0.3793, 0.3696, 0.5950, 0.5563], [0.5650, 0.3652, 0.8887, 0.4821, 0.4247, 0.4604, 0.6303, 0.5394]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6311, 0.3998, 0.7975, 0.5767, 0.3837, 0.4850, 0.7327, 0.5343], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6289, 0.4019, 0.8112, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332], [0.6257, 0.4167, 0.8775, 0.3433, 0.3562, 0.4133, 0.6200, 0.5667], [0.6325, 0.4165, 0.9000, 0.4617, 0.3812, 0.4900, 0.7485, 0.5447]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.11230159673141316 step: 59 running loss: 0.0019034168937527654 Train Steps: 59/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385], [0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6238, 0.5350], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400], [0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250], [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5976, 0.3865, 0.7750, 0.2826, 0.4411, 0.2298, 0.6490, 0.5309], [0.5556, 0.3312, 0.8283, 0.3899, 0.3939, 0.5214, 0.6137, 0.5141], [0.6032, 0.3724, 0.7987, 0.5199, 0.4078, 0.4533, 0.5771, 0.5130], [0.5418, 0.3389, 0.7947, 0.2667, 0.4400, 0.1755, 0.5935, 0.5198], [0.5723, 0.3736, 0.8127, 0.3381, 0.3666, 0.2736, 0.5842, 0.5161], [0.6249, 0.4152, 0.7927, 0.4392, 0.4423, 0.2585, 0.5467, 0.5873], [0.5716, 0.3700, 0.8444, 0.4802, 0.4240, 0.4934, 0.5635, 0.5217], [0.5619, 0.3722, 0.7979, 0.4927, 0.3437, 0.3396, 0.5290, 0.5390]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385], [0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6237, 0.5350], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400], [0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250], [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5638, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.11372045130701736 step: 60 running loss: 0.001895340855116956 Train Steps: 60/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343], [0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267], [0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6238, 0.5350], [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869], [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650], [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667], [0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6158, 0.4215, 0.8445, 0.3986, 0.4052, 0.2963, 0.6601, 0.5171], [0.6921, 0.4721, 0.8280, 0.5742, 0.4215, 0.4660, 0.6214, 0.5356], [0.6141, 0.4159, 0.7788, 0.2511, 0.4412, 0.1787, 0.6674, 0.5213], [0.6449, 0.4211, 0.8393, 0.3993, 0.3831, 0.5536, 0.6376, 0.5329], [0.6885, 0.4565, 0.8322, 0.3377, 0.4438, 0.2241, 0.6034, 0.5038], [0.6514, 0.4302, 0.8149, 0.5266, 0.4088, 0.4930, 0.5682, 0.5626], [0.1568, 0.1365, 0.7287, 0.2559, 0.3713, 0.2906, 0.6051, 0.5638], [0.6785, 0.4605, 0.8245, 0.5355, 0.4814, 0.4635, 0.5484, 0.5490]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6275, 0.4003, 0.9100, 0.3783, 0.4387, 0.3133, 0.7058, 0.5343], [0.6229, 0.4066, 0.8512, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267], [0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6237, 0.5350], [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869], [0.6098, 0.3991, 0.8637, 0.4717, 0.4263, 0.4967, 0.5213, 0.5650], [0.0000, 0.0000, 0.7525, 0.2291, 0.3837, 0.3017, 0.6050, 0.5667], [0.6084, 0.4008, 0.8587, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.11536864080699161 step: 61 running loss: 0.0018912891935572395 Train Steps: 61/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6218, 0.4137, 0.7263, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783], [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183], [0.6145, 0.4007, 0.8775, 0.4533, 0.4562, 0.5533, 0.6088, 0.5533], [0.6125, 0.4035, 0.7825, 0.3100, 0.3463, 0.4900, 0.5832, 0.5637], [0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283], [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783], [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312], [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6308, 0.4368, 0.7588, 0.2792, 0.4103, 0.2608, 0.6189, 0.5617], [0.6197, 0.4305, 0.7728, 0.3165, 0.4232, 0.3204, 0.6322, 0.5996], [0.6046, 0.4186, 0.8744, 0.4556, 0.4631, 0.5276, 0.6331, 0.5470], [0.5291, 0.3804, 0.7644, 0.3500, 0.3569, 0.4627, 0.6394, 0.5231], [0.6324, 0.4379, 0.8579, 0.5266, 0.5035, 0.4578, 0.5769, 0.5433], [0.6046, 0.4234, 0.8726, 0.4919, 0.3836, 0.4204, 0.5508, 0.5633], [0.5945, 0.4069, 0.8636, 0.4195, 0.3478, 0.3993, 0.6374, 0.5284], [0.6733, 0.4624, 0.8645, 0.5685, 0.3725, 0.4319, 0.6626, 0.5058]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6218, 0.4137, 0.7262, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783], [0.6196, 0.4094, 0.7563, 0.2817, 0.3938, 0.3183, 0.6012, 0.6183], [0.6145, 0.4007, 0.8775, 0.4533, 0.4563, 0.5533, 0.6087, 0.5533], [0.6125, 0.4035, 0.7825, 0.3100, 0.3462, 0.4900, 0.5832, 0.5637], [0.6084, 0.4008, 0.8587, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283], [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783], [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312], [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.11608802812406793 step: 62 running loss: 0.0018723875503881924 Train Steps: 62/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6273, 0.4100, 0.7137, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6362, 0.5200], [0.6245, 0.4115, 0.8700, 0.4883, 0.4625, 0.5517, 0.6100, 0.5217], [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413], [0.6117, 0.4019, 0.8538, 0.4067, 0.3513, 0.3583, 0.5663, 0.5133], [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183], [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4617, 0.3219, 0.7192, 0.2454, 0.3976, 0.2679, 0.6069, 0.5669], [0.6731, 0.4611, 0.7892, 0.2466, 0.4606, 0.2235, 0.6119, 0.5364], [0.5985, 0.4165, 0.8816, 0.3927, 0.3924, 0.5346, 0.6360, 0.5497], [0.6380, 0.4531, 0.8811, 0.4980, 0.4677, 0.5485, 0.6470, 0.5719], [0.5778, 0.3894, 0.9415, 0.3813, 0.4868, 0.3180, 0.7241, 0.5422], [0.6515, 0.4434, 0.8502, 0.4100, 0.3677, 0.3639, 0.5602, 0.5561], [0.6955, 0.4848, 0.9014, 0.4948, 0.4301, 0.5923, 0.6377, 0.5528], [0.5887, 0.4287, 0.7657, 0.3732, 0.3652, 0.4024, 0.5393, 0.5895]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6273, 0.4099, 0.7138, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6363, 0.5200], [0.6245, 0.4115, 0.8700, 0.4883, 0.4625, 0.5517, 0.6100, 0.5217], [0.6271, 0.4040, 0.9137, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413], [0.6116, 0.4019, 0.8537, 0.4067, 0.3512, 0.3583, 0.5663, 0.5133], [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183], [0.6170, 0.4102, 0.7468, 0.3695, 0.3462, 0.3767, 0.5238, 0.5823]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.11741927900584415 step: 63 running loss: 0.0018637980794578436 Train Steps: 63/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767], [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343], [0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263], [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6084, 0.3981, 0.8588, 0.5233, 0.4600, 0.5367, 0.5680, 0.5006], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5823, 0.3989, 0.7990, 0.2686, 0.4015, 0.3397, 0.6064, 0.5830], [0.5737, 0.3859, 0.9048, 0.3874, 0.4089, 0.3633, 0.6578, 0.5424], [0.6790, 0.4570, 0.7580, 0.2675, 0.4338, 0.2390, 0.5607, 0.5454], [0.5070, 0.3484, 0.7983, 0.2682, 0.4036, 0.3215, 0.6074, 0.5580], [0.4992, 0.3424, 0.7282, 0.2410, 0.4013, 0.2917, 0.5396, 0.5811], [0.5697, 0.3986, 0.8845, 0.5299, 0.4589, 0.5847, 0.5493, 0.5481], [0.6039, 0.4162, 0.8152, 0.2024, 0.4612, 0.2351, 0.6230, 0.5262], [0.6638, 0.4316, 0.9216, 0.4938, 0.4082, 0.4667, 0.6866, 0.5826]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6230, 0.4152, 0.7588, 0.2283, 0.4013, 0.2883, 0.6200, 0.5767], [0.6275, 0.4003, 0.9100, 0.3783, 0.4387, 0.3133, 0.7058, 0.5343], [0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263], [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6084, 0.3981, 0.8587, 0.5233, 0.4600, 0.5367, 0.5680, 0.5006], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6346, 0.4144, 0.9087, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.11882246885215864 step: 64 running loss: 0.0018566010758149787 Train Steps: 64/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6203, 0.4021, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [0.6200, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913], [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461], [0.6230, 0.4113, 0.7213, 0.1983, 0.4325, 0.2367, 0.6262, 0.5400], [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621], [0.6148, 0.3996, 0.8488, 0.3867, 0.3488, 0.4067, 0.5863, 0.5000], [0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267], [0.6133, 0.4066, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7349, 0.4675, 0.9156, 0.5079, 0.3783, 0.4450, 0.5984, 0.5317], [0.6379, 0.4151, 0.8800, 0.5041, 0.3912, 0.5593, 0.5922, 0.5318], [0.6555, 0.4495, 0.8806, 0.5197, 0.4423, 0.5433, 0.5737, 0.5766], [0.5777, 0.3901, 0.7510, 0.2130, 0.4300, 0.2646, 0.6502, 0.5579], [0.3214, 0.2092, 0.8647, 0.2586, 0.5082, 0.2396, 0.6842, 0.5704], [0.6745, 0.4494, 0.8771, 0.3857, 0.3560, 0.4396, 0.5858, 0.5374], [0.6192, 0.3965, 0.8787, 0.2707, 0.4508, 0.3048, 0.6362, 0.5435], [0.5929, 0.4056, 0.7187, 0.2436, 0.3843, 0.2642, 0.5445, 0.5340]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6203, 0.4020, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [0.6199, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913], [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461], [0.6230, 0.4113, 0.7212, 0.1983, 0.4325, 0.2367, 0.6263, 0.5400], [0.0000, 0.0000, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621], [0.6148, 0.3996, 0.8487, 0.3867, 0.3487, 0.4067, 0.5863, 0.5000], [0.6266, 0.4067, 0.8587, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267], [0.6133, 0.4065, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358]]], device='cuda:0') loss_train_step before backward: tensor(0.0032, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0032, device='cuda:0', grad_fn=) loss_train: 0.12204406381351873 step: 65 running loss: 0.001877600981746442 Train Steps: 65/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6124, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402], [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467], [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378], [0.6110, 0.3984, 0.8750, 0.4933, 0.4625, 0.4950, 0.5578, 0.5676], [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6549, 0.4234, 0.8641, 0.4460, 0.3999, 0.4543, 0.6240, 0.5543], [0.5980, 0.3909, 0.7557, 0.2231, 0.3643, 0.2611, 0.6287, 0.5349], [0.6213, 0.4000, 0.9014, 0.4747, 0.3929, 0.4499, 0.5685, 0.5125], [0.6375, 0.4072, 0.9264, 0.4221, 0.4738, 0.5030, 0.6107, 0.5567], [0.6489, 0.4262, 0.9240, 0.4144, 0.4471, 0.5674, 0.6234, 0.5403], [0.6403, 0.4147, 0.8330, 0.4742, 0.3766, 0.4779, 0.7220, 0.5145], [0.6165, 0.4096, 0.8961, 0.4373, 0.4341, 0.4900, 0.6077, 0.5471], [0.6455, 0.4126, 0.8287, 0.3788, 0.3643, 0.4262, 0.5727, 0.5199]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6123, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402], [0.6162, 0.4134, 0.6700, 0.2467, 0.3963, 0.2533, 0.5738, 0.5467], [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378], [0.6110, 0.3984, 0.8750, 0.4933, 0.4625, 0.4950, 0.5578, 0.5676], [0.6155, 0.4124, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6311, 0.3998, 0.7975, 0.5767, 0.3837, 0.4850, 0.7327, 0.5343], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.12342670414363965 step: 66 running loss: 0.001870101577933934 Train Steps: 66/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378], [0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6286, 0.4060, 0.9188, 0.4333, 0.3675, 0.4167, 0.7034, 0.5528], [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100], [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6517, 0.3950, 0.8952, 0.2544, 0.5097, 0.2601, 0.6829, 0.5222], [0.5733, 0.3255, 0.7971, 0.2400, 0.4828, 0.2072, 0.5727, 0.5022], [0.5915, 0.3806, 0.7819, 0.2226, 0.3704, 0.3152, 0.5553, 0.5218], [0.6712, 0.3939, 0.9119, 0.4217, 0.3616, 0.4310, 0.6332, 0.5328], [0.5975, 0.3927, 0.8889, 0.5248, 0.4028, 0.5850, 0.6696, 0.5266], [0.6125, 0.3791, 0.8563, 0.5179, 0.4040, 0.5349, 0.5519, 0.5101], [0.5716, 0.3439, 0.9131, 0.4123, 0.3465, 0.3948, 0.5630, 0.5008], [0.6269, 0.3833, 0.7719, 0.2893, 0.4000, 0.3447, 0.5547, 0.5907]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6293, 0.4097, 0.8800, 0.2517, 0.5263, 0.2600, 0.7430, 0.5378], [0.6286, 0.4078, 0.8062, 0.2267, 0.4787, 0.1533, 0.5953, 0.4913], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6286, 0.4060, 0.9187, 0.4333, 0.3675, 0.4167, 0.7034, 0.5528], [0.6339, 0.4123, 0.8637, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100], [0.6196, 0.4094, 0.7563, 0.2817, 0.3938, 0.3183, 0.6012, 0.6183]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.12444062257418409 step: 67 running loss: 0.0018573227249878222 Train Steps: 67/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496], [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609], [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.6228, 0.4119, 0.7938, 0.2233, 0.4674, 0.1773, 0.6188, 0.5433], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879], [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6532, 0.3863, 0.8381, 0.2773, 0.4113, 0.2732, 0.5710, 0.5276], [0.6314, 0.3613, 0.8673, 0.3157, 0.4232, 0.2928, 0.6520, 0.5310], [0.5641, 0.3614, 0.8284, 0.5704, 0.3935, 0.4518, 0.5438, 0.5686], [0.6321, 0.3723, 0.8681, 0.2921, 0.5005, 0.2325, 0.6820, 0.5124], [0.6510, 0.3914, 0.7841, 0.2322, 0.4774, 0.1835, 0.5655, 0.5282], [0.5351, 0.3280, 0.8166, 0.3076, 0.3767, 0.2982, 0.5308, 0.5306], [0.5455, 0.3412, 0.8776, 0.4614, 0.3899, 0.5606, 0.5658, 0.4976], [0.5498, 0.3304, 0.8194, 0.2611, 0.3887, 0.2648, 0.5999, 0.4929]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496], [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609], [0.6212, 0.4159, 0.8675, 0.5783, 0.4087, 0.4317, 0.5612, 0.5917], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.6228, 0.4119, 0.7937, 0.2233, 0.4674, 0.1773, 0.6187, 0.5433], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879], [0.6257, 0.4034, 0.8288, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.1254878489417024 step: 68 running loss: 0.0018454095432603294 Train Steps: 68/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [ nan, nan, 0.8938, 0.2850, 0.4662, 0.3117, 0.7406, 0.5528], [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726], [0.6138, 0.4101, 0.8800, 0.5083, 0.4637, 0.5950, 0.5587, 0.5077], [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433], [ nan, nan, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933], [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6844, 0.4405, 0.7799, 0.3047, 0.3637, 0.3610, 0.6419, 0.5207], [0.7517, 0.4813, 0.7875, 0.3211, 0.4087, 0.2684, 0.5113, 0.5277], [0.3289, 0.1807, 0.9043, 0.3134, 0.4653, 0.2704, 0.7535, 0.5364], [0.6930, 0.4234, 0.7588, 0.2919, 0.3988, 0.2571, 0.5680, 0.4735], [0.6735, 0.4149, 0.8795, 0.5366, 0.4842, 0.5513, 0.5983, 0.5389], [0.6846, 0.4375, 0.8348, 0.3652, 0.3751, 0.2941, 0.5715, 0.5281], [0.2488, 0.1441, 0.7344, 0.2680, 0.4080, 0.2224, 0.5436, 0.5447], [0.7038, 0.4378, 0.8358, 0.3522, 0.3714, 0.3495, 0.5815, 0.5219]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6197, 0.4051, 0.7812, 0.2650, 0.3512, 0.4050, 0.6112, 0.5500], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.0000, 0.0000, 0.8938, 0.2850, 0.4663, 0.3117, 0.7406, 0.5528], [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726], [0.6138, 0.4101, 0.8800, 0.5083, 0.4638, 0.5950, 0.5587, 0.5077], [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5412, 0.5433], [0.0000, 0.0000, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933], [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667]]], device='cuda:0') loss_train_step before backward: tensor(0.0049, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0049, device='cuda:0', grad_fn=) loss_train: 0.13038400636287406 step: 69 running loss: 0.0018896232806213632 Train Steps: 69/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426], [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519], [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285], [0.6128, 0.4115, 0.7934, 0.3778, 0.3450, 0.4033, 0.5337, 0.5456]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5951, 0.3895, 0.8416, 0.5235, 0.4252, 0.4983, 0.6261, 0.4952], [0.5503, 0.3507, 0.8777, 0.4197, 0.3331, 0.4423, 0.6457, 0.5161], [0.6041, 0.3976, 0.7719, 0.2042, 0.4376, 0.1779, 0.6123, 0.5478], [0.5624, 0.3707, 0.8516, 0.5060, 0.4581, 0.5141, 0.5622, 0.5226], [0.5725, 0.3693, 0.8532, 0.4400, 0.3778, 0.4570, 0.5595, 0.5110], [0.5750, 0.3745, 0.7993, 0.2587, 0.4015, 0.2159, 0.6403, 0.5213], [0.6475, 0.4099, 0.8714, 0.3639, 0.4452, 0.2523, 0.6694, 0.5656], [0.6218, 0.3965, 0.7890, 0.3606, 0.3206, 0.4008, 0.5353, 0.5258]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6219, 0.3934, 0.8687, 0.5267, 0.4313, 0.4967, 0.5987, 0.4983], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426], [0.6100, 0.4016, 0.8600, 0.5067, 0.4613, 0.5233, 0.5086, 0.5519], [0.6185, 0.4067, 0.8838, 0.4450, 0.4038, 0.4733, 0.5213, 0.5142], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285], [0.6128, 0.4115, 0.7934, 0.3778, 0.3450, 0.4033, 0.5337, 0.5456]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.13097077485872433 step: 70 running loss: 0.0018710110694103476 Train Steps: 70/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411], [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246], [0.6069, 0.3975, 0.8625, 0.5083, 0.4388, 0.5483, 0.5650, 0.4967], [0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967], [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869], [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6149, 0.4330, 0.7990, 0.2836, 0.4278, 0.1979, 0.5801, 0.5593], [0.6108, 0.4046, 0.8090, 0.4209, 0.3469, 0.4819, 0.6012, 0.5438], [0.5945, 0.3862, 0.8863, 0.4450, 0.3350, 0.4499, 0.6335, 0.5160], [0.1991, 0.1424, 0.7229, 0.2248, 0.4367, 0.1982, 0.5435, 0.5702], [0.5651, 0.3826, 0.8614, 0.5356, 0.4345, 0.5389, 0.5768, 0.5182], [0.6056, 0.3988, 0.8797, 0.4604, 0.4088, 0.2651, 0.6134, 0.5259], [0.6272, 0.4085, 0.8028, 0.2248, 0.4821, 0.1933, 0.6165, 0.5272], [0.5926, 0.4070, 0.8533, 0.4843, 0.3864, 0.3605, 0.6208, 0.5338]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411], [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.0000, 0.0000, 0.6900, 0.1917, 0.3938, 0.2367, 0.5240, 0.5246], [0.6069, 0.3975, 0.8625, 0.5083, 0.4387, 0.5483, 0.5650, 0.4967], [0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967], [0.6276, 0.4095, 0.8238, 0.2250, 0.4663, 0.1783, 0.6171, 0.4869], [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.1323865795857273 step: 71 running loss: 0.0018645997124750325 Train Steps: 71/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983], [0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6188, 0.5400], [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633], [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467], [0.6109, 0.4015, 0.7668, 0.3639, 0.3513, 0.3667, 0.5200, 0.5641], [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5403, 0.3588, 0.8773, 0.5526, 0.4269, 0.4926, 0.5876, 0.5006], [0.5450, 0.3859, 0.8281, 0.3307, 0.3718, 0.3166, 0.6227, 0.5600], [0.6058, 0.4103, 0.8098, 0.2198, 0.4384, 0.1634, 0.6116, 0.5619], [0.5277, 0.3507, 0.8685, 0.5997, 0.3702, 0.4894, 0.5950, 0.4821], [0.5591, 0.3835, 0.9010, 0.4585, 0.4093, 0.5428, 0.6151, 0.5703], [0.5242, 0.3608, 0.7226, 0.2736, 0.3663, 0.2387, 0.5913, 0.5539], [0.5398, 0.3709, 0.8187, 0.3559, 0.3501, 0.3463, 0.5325, 0.5561], [0.5712, 0.3901, 0.7794, 0.2210, 0.4508, 0.1871, 0.6098, 0.5435]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6219, 0.3934, 0.8687, 0.5267, 0.4313, 0.4967, 0.5987, 0.4983], [0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6187, 0.5400], [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6189, 0.4049, 0.8888, 0.4417, 0.4212, 0.5200, 0.5987, 0.5633], [0.6162, 0.4134, 0.6700, 0.2467, 0.3963, 0.2533, 0.5738, 0.5467], [0.6109, 0.4015, 0.7668, 0.3639, 0.3512, 0.3667, 0.5200, 0.5641], [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.13352739083347842 step: 72 running loss: 0.0018545470949094226 Train Steps: 72/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183], [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822], [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285], [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862], [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5702, 0.3739, 0.7809, 0.2329, 0.4572, 0.1862, 0.6156, 0.5310], [0.6047, 0.4196, 0.8026, 0.3138, 0.4052, 0.2717, 0.6298, 0.5326], [0.5253, 0.3631, 0.8580, 0.4306, 0.3677, 0.3510, 0.5075, 0.5822], [0.5306, 0.3709, 0.7307, 0.2197, 0.3811, 0.2467, 0.5953, 0.5587], [0.6008, 0.4069, 0.7982, 0.4377, 0.3647, 0.4292, 0.5564, 0.5311], [0.5620, 0.3893, 0.8881, 0.4307, 0.3675, 0.4510, 0.6316, 0.5068], [0.4929, 0.3426, 0.8852, 0.4981, 0.4812, 0.4883, 0.5724, 0.5553], [0.5472, 0.3816, 0.8672, 0.4665, 0.3976, 0.4679, 0.5754, 0.5143]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.6201, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183], [0.6182, 0.4058, 0.8737, 0.4350, 0.3562, 0.3400, 0.5290, 0.5822], [0.6114, 0.4018, 0.7212, 0.1967, 0.3762, 0.2700, 0.5875, 0.5533], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285], [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862], [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.13450522517086938 step: 73 running loss: 0.0018425373311077998 Train Steps: 73/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6346, 0.4165, 0.9138, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471], [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6152, 0.4131, 0.6863, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [ nan, nan, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819], [0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249], [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733], [0.6307, 0.4029, 0.8650, 0.5200, 0.3763, 0.4017, 0.7311, 0.5366], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5932, 0.4021, 0.9224, 0.4201, 0.4089, 0.4184, 0.6681, 0.5319], [0.6426, 0.4420, 0.9131, 0.5361, 0.3849, 0.4679, 0.5603, 0.5626], [0.5596, 0.3877, 0.7069, 0.2391, 0.3801, 0.3218, 0.5559, 0.5568], [0.3145, 0.2176, 0.6766, 0.2093, 0.3874, 0.2276, 0.5255, 0.5453], [0.6297, 0.4334, 0.8823, 0.5122, 0.3979, 0.5125, 0.6349, 0.5204], [0.6396, 0.4406, 0.8905, 0.4983, 0.3858, 0.4697, 0.5679, 0.5379], [0.6697, 0.4546, 0.8755, 0.4976, 0.4065, 0.3941, 0.6260, 0.5066], [0.5101, 0.3365, 0.7944, 0.2642, 0.3986, 0.2635, 0.5119, 0.5375]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6346, 0.4165, 0.9137, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471], [0.6222, 0.4171, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6152, 0.4131, 0.6862, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [0.0000, 0.0000, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819], [0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249], [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733], [0.6307, 0.4029, 0.8650, 0.5200, 0.3762, 0.4017, 0.7311, 0.5366], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901]]], device='cuda:0') loss_train_step before backward: tensor(0.0034, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0034, device='cuda:0', grad_fn=) loss_train: 0.13787507318193093 step: 74 running loss: 0.0018631766646206882 Train Steps: 74/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6275, 0.4048, 0.8488, 0.2883, 0.4463, 0.2033, 0.6321, 0.5155], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507], [0.6048, 0.3928, 0.8538, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719], [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5797, 0.3975, 0.7268, 0.2070, 0.3868, 0.2786, 0.5860, 0.5435], [0.6807, 0.4556, 0.8901, 0.5002, 0.3596, 0.4272, 0.5110, 0.5236], [0.5820, 0.3928, 0.8670, 0.3012, 0.4777, 0.2253, 0.6494, 0.5298], [0.6709, 0.4309, 0.8455, 0.2722, 0.4336, 0.2452, 0.6256, 0.5038], [0.7241, 0.4788, 0.8761, 0.5397, 0.3933, 0.6080, 0.6220, 0.5062], [0.5731, 0.3881, 0.6840, 0.2172, 0.3951, 0.2544, 0.5523, 0.5419], [0.6635, 0.4368, 0.8471, 0.5739, 0.3895, 0.5456, 0.5621, 0.5044], [0.0911, 0.0716, 0.7280, 0.2380, 0.4164, 0.2594, 0.5111, 0.5556]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6275, 0.4048, 0.8487, 0.2883, 0.4462, 0.2033, 0.6321, 0.5155], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6072, 0.4029, 0.7038, 0.2150, 0.3913, 0.2267, 0.5516, 0.5507], [0.6048, 0.3928, 0.8537, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719], [0.0000, 0.0000, 0.6900, 0.1917, 0.3938, 0.2367, 0.5240, 0.5246]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.13911375048337504 step: 75 running loss: 0.0018548500064450005 Train Steps: 75/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750], [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540], [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507], [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899], [0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412], [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391], [0.6115, 0.4005, 0.8838, 0.3867, 0.3763, 0.4700, 0.5800, 0.5550], [0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6325, 0.4128, 0.8378, 0.4977, 0.4163, 0.5059, 0.5350, 0.5198], [0.4977, 0.3286, 0.8924, 0.4457, 0.3964, 0.3302, 0.6415, 0.5278], [0.5240, 0.3414, 0.6580, 0.1917, 0.3814, 0.2373, 0.5462, 0.5343], [0.6375, 0.4162, 0.8675, 0.4607, 0.3843, 0.4125, 0.6786, 0.5469], [0.6242, 0.3876, 0.8634, 0.3314, 0.3710, 0.2785, 0.5657, 0.5215], [0.5205, 0.3214, 0.7670, 0.2383, 0.4139, 0.2655, 0.5800, 0.5364], [0.6382, 0.4089, 0.8369, 0.3992, 0.3415, 0.5122, 0.5060, 0.5114], [0.6169, 0.4138, 0.8641, 0.3638, 0.3694, 0.2957, 0.6184, 0.4895]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750], [0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540], [0.6072, 0.4029, 0.7038, 0.2150, 0.3913, 0.2267, 0.5516, 0.5507], [0.6346, 0.4144, 0.9087, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899], [0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412], [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391], [0.6115, 0.4005, 0.8838, 0.3867, 0.3762, 0.4700, 0.5800, 0.5550], [0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.14055398636264727 step: 76 running loss: 0.0018493945574032534 Train Steps: 76/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [ nan, nan, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220], [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717], [0.6126, 0.4067, 0.8638, 0.5383, 0.4188, 0.4850, 0.5016, 0.5392], [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433], [0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7221, 0.4383, 0.8886, 0.4777, 0.3893, 0.4451, 0.5912, 0.5377], [0.5946, 0.3824, 0.8502, 0.2653, 0.4305, 0.2288, 0.7005, 0.5474], [0.0674, 0.0377, 0.7684, 0.2331, 0.3859, 0.2675, 0.5374, 0.5340], [0.6066, 0.3915, 0.7468, 0.2663, 0.3789, 0.2621, 0.5982, 0.5673], [0.7450, 0.4842, 0.8662, 0.5883, 0.4239, 0.4681, 0.5471, 0.5095], [0.6306, 0.4095, 0.8095, 0.2387, 0.3810, 0.2444, 0.6396, 0.4975], [0.6211, 0.4149, 0.8376, 0.3492, 0.3540, 0.3035, 0.5393, 0.5472], [0.6774, 0.4275, 0.8233, 0.3700, 0.3682, 0.5086, 0.5857, 0.5150]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6197, 0.4091, 0.8800, 0.4783, 0.3537, 0.4767, 0.5950, 0.5550], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.0000, 0.0000, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220], [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717], [0.6126, 0.4067, 0.8637, 0.5383, 0.4187, 0.4850, 0.5016, 0.5392], [0.6257, 0.4034, 0.8288, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5412, 0.5433], [0.6132, 0.4118, 0.8200, 0.3633, 0.3562, 0.5400, 0.5787, 0.5136]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.14160173939308152 step: 77 running loss: 0.0018389836284815782 Train Steps: 77/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843], [0.6200, 0.3993, 0.8519, 0.4923, 0.3962, 0.4717, 0.6013, 0.5433], [0.6222, 0.4169, 0.8638, 0.5650, 0.4313, 0.4783, 0.5637, 0.5633], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400], [0.6152, 0.4131, 0.6863, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6685, 0.4316, 0.9437, 0.4782, 0.3725, 0.5001, 0.6333, 0.4880], [0.6572, 0.4154, 0.9135, 0.4934, 0.3772, 0.4707, 0.6351, 0.5345], [0.6967, 0.4506, 0.9175, 0.5615, 0.4209, 0.4681, 0.6014, 0.5842], [0.5686, 0.3843, 0.8098, 0.2375, 0.4232, 0.1728, 0.5963, 0.5342], [0.6842, 0.4281, 0.8738, 0.3398, 0.3418, 0.4645, 0.6383, 0.5833], [0.5826, 0.3776, 0.7209, 0.2505, 0.3553, 0.3277, 0.5802, 0.5649], [0.4752, 0.2821, 0.7004, 0.1778, 0.3955, 0.1971, 0.5591, 0.5399], [0.4212, 0.2661, 0.7238, 0.2077, 0.4278, 0.1912, 0.5486, 0.5777]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6264, 0.4049, 0.8988, 0.4633, 0.3812, 0.4983, 0.6326, 0.4843], [0.6200, 0.3993, 0.8519, 0.4923, 0.3963, 0.4717, 0.6012, 0.5433], [0.6222, 0.4169, 0.8637, 0.5650, 0.4313, 0.4783, 0.5638, 0.5633], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400], [0.6152, 0.4131, 0.6862, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [0.6054, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5512, 0.5750]]], device='cuda:0') loss_train_step before backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train: 0.14373794820858166 step: 78 running loss: 0.001842794207802329 Train Steps: 78/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717], [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6005, 0.3883, 0.7376, 0.2639, 0.3651, 0.2618, 0.5934, 0.5770], [0.5534, 0.3800, 0.8344, 0.2513, 0.5063, 0.2552, 0.7289, 0.5471], [0.6056, 0.3878, 0.8910, 0.4077, 0.4137, 0.2377, 0.5669, 0.5622], [0.6107, 0.3752, 0.8548, 0.4358, 0.3673, 0.5117, 0.5738, 0.5285], [0.5490, 0.3539, 0.8691, 0.4588, 0.4263, 0.5274, 0.5895, 0.5428], [0.5000, 0.3380, 0.6937, 0.2551, 0.4053, 0.2099, 0.5916, 0.5737], [0.5854, 0.3727, 0.8872, 0.4041, 0.3371, 0.4301, 0.6151, 0.5077], [0.6438, 0.4108, 0.8618, 0.3766, 0.3392, 0.3494, 0.5652, 0.5354]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717], [0.6300, 0.4133, 0.8537, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.6226, 0.4098, 0.8913, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.14473954314598814 step: 79 running loss: 0.0018321461157720018 Train Steps: 79/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519], [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167], [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986], [0.6200, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583], [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400], [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5663, 0.3640, 0.8650, 0.4958, 0.4625, 0.4853, 0.5205, 0.5457], [0.6266, 0.4113, 0.8763, 0.3917, 0.3851, 0.3266, 0.6356, 0.5400], [0.6017, 0.4096, 0.7041, 0.2273, 0.3641, 0.3217, 0.6051, 0.5864], [0.6583, 0.4444, 0.8952, 0.3599, 0.4401, 0.2048, 0.6525, 0.5195], [0.5728, 0.3495, 0.8861, 0.4576, 0.3689, 0.3999, 0.5798, 0.5363], [0.6590, 0.4320, 0.6928, 0.2192, 0.3806, 0.2505, 0.6148, 0.5537], [0.6123, 0.3829, 0.8421, 0.3424, 0.3509, 0.4881, 0.6346, 0.5532], [0.5733, 0.3608, 0.9131, 0.4369, 0.4142, 0.5019, 0.5784, 0.5521]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6100, 0.4016, 0.8600, 0.5067, 0.4613, 0.5233, 0.5086, 0.5519], [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6263, 0.5167], [0.6132, 0.4066, 0.7259, 0.2402, 0.3587, 0.3300, 0.6000, 0.5600], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986], [0.6201, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6087, 0.5583], [0.6168, 0.4029, 0.8523, 0.3417, 0.3587, 0.5000, 0.6125, 0.5400], [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5838, 0.5500]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.14520067651756108 step: 80 running loss: 0.0018150084564695136 Train Steps: 80/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290], [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960], [0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6128, 0.4118, 0.8638, 0.5333, 0.4625, 0.5267, 0.5193, 0.5475], [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7075, 0.4653, 0.8551, 0.4899, 0.3809, 0.3419, 0.6343, 0.4962], [0.7156, 0.4710, 0.7829, 0.2770, 0.3385, 0.3715, 0.6403, 0.5497], [0.6334, 0.4071, 0.8531, 0.3993, 0.4389, 0.4503, 0.5759, 0.5201], [0.6304, 0.4207, 0.8709, 0.3307, 0.4067, 0.2475, 0.6356, 0.5532], [0.5920, 0.3828, 0.8642, 0.4004, 0.4326, 0.5304, 0.6126, 0.5525], [0.6249, 0.4180, 0.8541, 0.4851, 0.4458, 0.4441, 0.5536, 0.5530], [0.4778, 0.3300, 0.8495, 0.3993, 0.4294, 0.5057, 0.6012, 0.5509], [0.5924, 0.4014, 0.8258, 0.4693, 0.4127, 0.4612, 0.6556, 0.5125]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6272, 0.4071, 0.8737, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290], [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960], [0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6128, 0.4118, 0.8637, 0.5333, 0.4625, 0.5267, 0.5193, 0.5475], [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6299, 0.4008, 0.8450, 0.5350, 0.4212, 0.5000, 0.6350, 0.5100]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.1471182822715491 step: 81 running loss: 0.0018162750897722112 Train Steps: 81/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6117, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103], [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344], [ nan, nan, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983], [0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116], [0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650], [0.6166, 0.4008, 0.8563, 0.5667, 0.4388, 0.4933, 0.5575, 0.5567], [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6829, 0.4525, 0.6718, 0.2448, 0.4178, 0.2707, 0.5433, 0.5212], [0.6605, 0.4421, 0.9349, 0.4715, 0.4095, 0.4238, 0.5632, 0.5242], [0.1362, 0.0846, 0.7230, 0.2268, 0.4773, 0.2087, 0.5771, 0.5049], [0.7012, 0.4477, 0.6752, 0.2197, 0.4411, 0.2342, 0.5737, 0.5244], [0.7194, 0.4713, 0.7563, 0.2293, 0.4259, 0.2597, 0.6057, 0.4829], [0.5869, 0.3943, 0.9245, 0.4066, 0.3863, 0.4469, 0.6174, 0.5767], [0.6784, 0.4460, 0.8884, 0.5537, 0.4868, 0.5206, 0.5965, 0.5378], [0.6576, 0.4474, 0.7983, 0.4115, 0.3844, 0.3781, 0.5655, 0.5796]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6116, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103], [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344], [0.0000, 0.0000, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983], [0.6054, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6165, 0.4106, 0.7575, 0.1733, 0.3837, 0.2650, 0.5680, 0.5116], [0.6203, 0.4096, 0.8863, 0.4267, 0.3537, 0.4117, 0.6025, 0.5650], [0.6166, 0.4008, 0.8562, 0.5667, 0.4387, 0.4933, 0.5575, 0.5567], [0.6179, 0.4118, 0.7278, 0.4237, 0.3587, 0.3400, 0.5675, 0.5917]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.1490206648595631 step: 82 running loss: 0.0018173251812141843 Train Steps: 82/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6222, 0.4169, 0.8638, 0.5650, 0.4313, 0.4783, 0.5637, 0.5633], [0.6198, 0.4115, 0.7762, 0.2717, 0.3713, 0.3200, 0.5837, 0.5683], [0.6198, 0.4101, 0.8838, 0.5283, 0.3763, 0.5267, 0.5913, 0.5567], [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103], [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717], [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6113, 0.4148, 0.8368, 0.5495, 0.4525, 0.4887, 0.5821, 0.5630], [0.6829, 0.4633, 0.7749, 0.2959, 0.3903, 0.3291, 0.5708, 0.5146], [0.6550, 0.4410, 0.8355, 0.5153, 0.4186, 0.5387, 0.5706, 0.5386], [0.6351, 0.4176, 0.8367, 0.4638, 0.4074, 0.4470, 0.5879, 0.5185], [0.5865, 0.3998, 0.8404, 0.4334, 0.4042, 0.4874, 0.5415, 0.4974], [0.6618, 0.4470, 0.8834, 0.4661, 0.4353, 0.5668, 0.7026, 0.5230], [0.6311, 0.4185, 0.8328, 0.4574, 0.4471, 0.4798, 0.5398, 0.5308], [0.6300, 0.4266, 0.8661, 0.3974, 0.4017, 0.3521, 0.6201, 0.4980]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6222, 0.4169, 0.8637, 0.5650, 0.4313, 0.4783, 0.5638, 0.5633], [0.6198, 0.4115, 0.7763, 0.2717, 0.3713, 0.3200, 0.5838, 0.5683], [0.6198, 0.4101, 0.8838, 0.5283, 0.3762, 0.5267, 0.5913, 0.5567], [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103], [0.6339, 0.4102, 0.9087, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5412, 0.5717], [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.14975211105775088 step: 83 running loss: 0.001804242301900613 Train Steps: 83/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114], [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533], [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197], [0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471], [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967], [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6903, 0.4541, 0.8041, 0.3519, 0.4225, 0.2267, 0.5705, 0.4859], [0.5929, 0.3934, 0.8567, 0.5297, 0.4653, 0.5290, 0.6259, 0.4991], [0.6601, 0.4450, 0.6932, 0.2327, 0.3852, 0.2471, 0.6020, 0.5289], [0.5416, 0.3797, 0.8674, 0.5292, 0.4690, 0.5223, 0.5841, 0.5070], [0.5711, 0.3844, 0.8338, 0.5346, 0.4663, 0.5404, 0.5545, 0.5090], [0.6060, 0.4009, 0.8575, 0.4630, 0.3739, 0.4338, 0.4998, 0.5272], [0.6225, 0.3925, 0.8570, 0.4345, 0.3532, 0.4917, 0.6009, 0.5021], [0.6001, 0.4009, 0.8238, 0.3495, 0.3516, 0.3660, 0.5789, 0.5336]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6143, 0.4040, 0.8238, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114], [0.6114, 0.4018, 0.7212, 0.1967, 0.3762, 0.2700, 0.5875, 0.5533], [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197], [0.6075, 0.4000, 0.8512, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6127, 0.4084, 0.8700, 0.4467, 0.3988, 0.4317, 0.5013, 0.5471], [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967], [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.15045929490588605 step: 84 running loss: 0.0017911820822129293 Train Steps: 84/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550], [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507], [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800], [0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896], [0.6234, 0.4179, 0.7825, 0.3450, 0.3813, 0.2867, 0.5675, 0.5617], [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5609, 0.3478, 0.9110, 0.5086, 0.3812, 0.5075, 0.5569, 0.5241], [0.6739, 0.4449, 0.7094, 0.2639, 0.4078, 0.2612, 0.5407, 0.5280], [0.5018, 0.3333, 0.7566, 0.3187, 0.3691, 0.3055, 0.4783, 0.5424], [0.6060, 0.4071, 0.7797, 0.2923, 0.3817, 0.3736, 0.5510, 0.5038], [0.6098, 0.4134, 0.8727, 0.2888, 0.5322, 0.2752, 0.7309, 0.5040], [0.5934, 0.3906, 0.8474, 0.4316, 0.3445, 0.3416, 0.5021, 0.5575], [0.5262, 0.3429, 0.8057, 0.3826, 0.3886, 0.3141, 0.5591, 0.5266], [0.5721, 0.3836, 0.8637, 0.6086, 0.4031, 0.4786, 0.5398, 0.5554]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6197, 0.4091, 0.8800, 0.4783, 0.3537, 0.4767, 0.5950, 0.5550], [0.6072, 0.4029, 0.7038, 0.2150, 0.3913, 0.2267, 0.5516, 0.5507], [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5462, 0.5800], [0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896], [0.6234, 0.4179, 0.7825, 0.3450, 0.3812, 0.2867, 0.5675, 0.5617], [0.6212, 0.4159, 0.8675, 0.5783, 0.4087, 0.4317, 0.5612, 0.5917]]], device='cuda:0') loss_train_step before backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train: 0.15215556626208127 step: 85 running loss: 0.0017900654854362502 Train Steps: 85/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383], [0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767], [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283], [ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617], [ nan, nan, 0.7192, 0.2346, 0.4037, 0.2050, 0.5138, 0.5650], [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6643, 0.4457, 0.8618, 0.2849, 0.5180, 0.2876, 0.6996, 0.5425], [0.7646, 0.5065, 0.7449, 0.2939, 0.4197, 0.1968, 0.5505, 0.5345], [0.6854, 0.4418, 0.7921, 0.3072, 0.3834, 0.3308, 0.5712, 0.5596], [0.6527, 0.4247, 0.8917, 0.4491, 0.3591, 0.4004, 0.5751, 0.5346], [0.6775, 0.4342, 0.9259, 0.4055, 0.3780, 0.2888, 0.5830, 0.5323], [0.2793, 0.1801, 0.7538, 0.2756, 0.4184, 0.2544, 0.5126, 0.5533], [0.0762, 0.0398, 0.7043, 0.2500, 0.3781, 0.2918, 0.4472, 0.5579], [0.6697, 0.4160, 0.8804, 0.4967, 0.3482, 0.4281, 0.4500, 0.5749]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383], [0.6230, 0.4152, 0.7588, 0.2283, 0.4013, 0.2883, 0.6200, 0.5767], [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6263, 0.5167], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6187, 0.5283], [0.0000, 0.0000, 0.7513, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617], [0.0000, 0.0000, 0.7192, 0.2346, 0.4038, 0.2050, 0.5138, 0.5650], [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892]]], device='cuda:0') loss_train_step before backward: tensor(0.0037, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0037, device='cuda:0', grad_fn=) loss_train: 0.15586218307726085 step: 86 running loss: 0.0018123509660146612 Train Steps: 86/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.6120, 0.4014, 0.6863, 0.2817, 0.3700, 0.2783, 0.5513, 0.5667], [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400], [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333], [0.6083, 0.3957, 0.8638, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5563, 0.3623, 0.9090, 0.4497, 0.4262, 0.5491, 0.6082, 0.5700], [0.6014, 0.4059, 0.8575, 0.3241, 0.3529, 0.2868, 0.5669, 0.5697], [0.6375, 0.4184, 0.8368, 0.5489, 0.3802, 0.4560, 0.5802, 0.6651], [0.6058, 0.3839, 0.8531, 0.4630, 0.3414, 0.3593, 0.4970, 0.5782], [0.5963, 0.3940, 0.7217, 0.2645, 0.3463, 0.2958, 0.5310, 0.5844], [0.6387, 0.3952, 0.9072, 0.4854, 0.4515, 0.5408, 0.5929, 0.5492], [0.6590, 0.4444, 0.8863, 0.4673, 0.3424, 0.4635, 0.6213, 0.5645], [0.5478, 0.3320, 0.8871, 0.4691, 0.4188, 0.4898, 0.5487, 0.5225]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6151, 0.4125, 0.8737, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.6120, 0.4013, 0.6862, 0.2817, 0.3700, 0.2783, 0.5512, 0.5667], [0.6185, 0.4098, 0.8838, 0.4900, 0.4538, 0.5800, 0.6288, 0.5400], [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6137, 0.5333], [0.6083, 0.3957, 0.8637, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.15654120134422556 step: 87 running loss: 0.001799324153381903 Train Steps: 87/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6357, 0.4118, 0.8400, 0.2500, 0.5413, 0.1633, 0.6725, 0.5586], [0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200], [0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882], [0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123], [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567], [0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967], [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5790, 0.3507, 0.8852, 0.2404, 0.5145, 0.1579, 0.6487, 0.5750], [0.5448, 0.3536, 0.8686, 0.5279, 0.3656, 0.4509, 0.5883, 0.5766], [0.6526, 0.4043, 0.8698, 0.4957, 0.3553, 0.3628, 0.5310, 0.5696], [0.5740, 0.3852, 0.8894, 0.4132, 0.3652, 0.5292, 0.5823, 0.5577], [0.5861, 0.3909, 0.7450, 0.2332, 0.3718, 0.2609, 0.5354, 0.5544], [0.5748, 0.3649, 0.8456, 0.4180, 0.3466, 0.3991, 0.5565, 0.6004], [0.6497, 0.3969, 0.8797, 0.4306, 0.3741, 0.5402, 0.5959, 0.6054], [0.6204, 0.3975, 0.8895, 0.4839, 0.4497, 0.4924, 0.5877, 0.5582]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6357, 0.4118, 0.8400, 0.2500, 0.5412, 0.1633, 0.6725, 0.5586], [0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200], [0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882], [0.6161, 0.4099, 0.8737, 0.4383, 0.3787, 0.5483, 0.5605, 0.5019], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123], [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567], [0.6199, 0.4102, 0.8950, 0.4417, 0.4013, 0.5367, 0.6112, 0.5967], [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.1574872225173749 step: 88 running loss: 0.001789627528606533 Train Steps: 88/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683], [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400], [0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5887, 0.5017], [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204], [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852], [0.6218, 0.4098, 0.7238, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350], [0.6267, 0.4080, 0.8438, 0.2633, 0.4763, 0.1800, 0.6259, 0.5240], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6563, 0.4143, 0.8711, 0.5141, 0.3567, 0.3754, 0.5637, 0.5939], [0.4888, 0.3187, 0.8533, 0.4704, 0.4120, 0.5141, 0.5794, 0.5805], [0.5305, 0.3432, 0.8643, 0.4474, 0.3680, 0.4902, 0.5796, 0.5600], [0.5621, 0.3650, 0.8665, 0.5146, 0.3924, 0.5260, 0.6470, 0.5585], [0.5229, 0.3579, 0.8662, 0.4833, 0.4403, 0.4909, 0.5109, 0.5972], [0.6561, 0.4365, 0.7485, 0.2128, 0.4266, 0.2200, 0.5967, 0.5774], [0.6677, 0.4292, 0.8820, 0.2452, 0.4535, 0.1810, 0.6232, 0.5437], [0.5569, 0.3679, 0.7763, 0.1875, 0.4438, 0.1674, 0.5821, 0.5393]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5412, 0.5683], [0.6201, 0.3970, 0.8413, 0.4950, 0.4412, 0.5183, 0.6087, 0.5400], [0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5888, 0.5017], [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204], [0.6140, 0.4070, 0.8700, 0.5000, 0.4613, 0.4900, 0.5260, 0.5852], [0.6218, 0.4098, 0.7237, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350], [0.6267, 0.4080, 0.8438, 0.2633, 0.4762, 0.1800, 0.6259, 0.5240], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.15884605079190806 step: 89 running loss: 0.0017847870875495287 Train Steps: 89/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6248, 0.4032, 0.7738, 0.1900, 0.4813, 0.1400, 0.5941, 0.4904], [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683], [0.6171, 0.4127, 0.8900, 0.4800, 0.4325, 0.5783, 0.5769, 0.5090], [0.6284, 0.4093, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5701, 0.3899, 0.7749, 0.2235, 0.4891, 0.1285, 0.6125, 0.5358], [0.6624, 0.4344, 0.8536, 0.4378, 0.3787, 0.3752, 0.5085, 0.5879], [0.6781, 0.4539, 0.8667, 0.5125, 0.3853, 0.3703, 0.5882, 0.5832], [0.6412, 0.4408, 0.9126, 0.4809, 0.4415, 0.5653, 0.5623, 0.5423], [0.6393, 0.4321, 0.8859, 0.4558, 0.3700, 0.3787, 0.6376, 0.5289], [0.5884, 0.4076, 0.9018, 0.4764, 0.4055, 0.5284, 0.6674, 0.5022], [0.1344, 0.0866, 0.6973, 0.2091, 0.4333, 0.1891, 0.5127, 0.5675], [0.6251, 0.4193, 0.7968, 0.2227, 0.4558, 0.2134, 0.6566, 0.5755]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6248, 0.4032, 0.7738, 0.1900, 0.4812, 0.1400, 0.5941, 0.4904], [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5412, 0.5683], [0.6171, 0.4127, 0.8900, 0.4800, 0.4325, 0.5783, 0.5769, 0.5090], [0.6284, 0.4092, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.0000, 0.0000, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.1597334560356103 step: 90 running loss: 0.0017748161781734476 Valid Steps: 10/10 Loss: nan 1.0033 -------------------------------------------------- Epoch: 3 Train Loss: 0.0018 Valid Loss: nan -------------------------------------------------- size of train loader is: 90 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6069, 0.3975, 0.8625, 0.5083, 0.4388, 0.5483, 0.5650, 0.4967], [0.6164, 0.4119, 0.7913, 0.2650, 0.3538, 0.3500, 0.5614, 0.5038], [0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167], [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6063, 0.6083], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650], [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892], [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5968, 0.4003, 0.8611, 0.5263, 0.4863, 0.5078, 0.5804, 0.5099], [0.5379, 0.3639, 0.7909, 0.2545, 0.3858, 0.3378, 0.6020, 0.4907], [0.5206, 0.3413, 0.8314, 0.2572, 0.4827, 0.1848, 0.6728, 0.4873], [0.6144, 0.4080, 0.8864, 0.5306, 0.4392, 0.4873, 0.6470, 0.5047], [0.6072, 0.4264, 0.8403, 0.3939, 0.3851, 0.4020, 0.6346, 0.5666], [0.5697, 0.3890, 0.7807, 0.2813, 0.4378, 0.2358, 0.5794, 0.5491], [0.6462, 0.4159, 0.8569, 0.4667, 0.4047, 0.3509, 0.5302, 0.5521], [0.5676, 0.3707, 0.7575, 0.2766, 0.4180, 0.2593, 0.5911, 0.4781]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6069, 0.3975, 0.8625, 0.5083, 0.4387, 0.5483, 0.5650, 0.4967], [0.6164, 0.4119, 0.7912, 0.2650, 0.3537, 0.3500, 0.5614, 0.5038], [0.6277, 0.4056, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167], [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6062, 0.6083], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650], [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892], [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.0011329366825520992 step: 1 running loss: 0.0011329366825520992 Train Steps: 1/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6203, 0.4021, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235], [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500], [0.6214, 0.4112, 0.7838, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083], [0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683], [0.6289, 0.4081, 0.8720, 0.3487, 0.3900, 0.3183, 0.6703, 0.5376], [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6306, 0.3991, 0.8891, 0.5327, 0.4077, 0.3841, 0.5966, 0.4759], [0.5868, 0.3788, 0.8220, 0.5059, 0.4200, 0.4516, 0.5521, 0.5059], [0.5743, 0.3819, 0.8683, 0.4992, 0.4607, 0.5130, 0.6418, 0.4970], [0.5523, 0.3709, 0.7587, 0.2767, 0.3894, 0.3869, 0.5942, 0.5189], [0.5849, 0.3792, 0.7901, 0.2298, 0.3933, 0.2529, 0.5977, 0.4827], [0.6309, 0.4188, 0.8185, 0.3005, 0.4504, 0.2756, 0.6154, 0.5288], [0.6407, 0.4076, 0.8784, 0.3673, 0.4201, 0.2635, 0.6971, 0.5146], [0.5239, 0.3321, 0.7461, 0.2019, 0.4320, 0.2133, 0.5811, 0.4603]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6203, 0.4020, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235], [0.6197, 0.4051, 0.7812, 0.2650, 0.3512, 0.4050, 0.6112, 0.5500], [0.6214, 0.4112, 0.7837, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083], [0.6200, 0.4098, 0.8238, 0.2917, 0.4013, 0.2967, 0.6000, 0.5683], [0.6289, 0.4081, 0.8720, 0.3487, 0.3900, 0.3183, 0.6703, 0.5376], [0.6165, 0.4106, 0.7575, 0.1733, 0.3837, 0.2650, 0.5680, 0.5116]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.0023326061200350523 step: 2 running loss: 0.0011663030600175261 Train Steps: 2/90 Loss: 0.0012 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300], [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413], [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864], [0.6332, 0.4118, 0.9238, 0.4267, 0.4012, 0.4733, 0.7525, 0.5436], [0.6250, 0.4103, 0.8950, 0.4400, 0.3912, 0.5650, 0.6050, 0.5133], [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882], [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5860, 0.3922, 0.7268, 0.2220, 0.3817, 0.2936, 0.5812, 0.4967], [0.4988, 0.3249, 0.8917, 0.3148, 0.4922, 0.2506, 0.6934, 0.4801], [0.7497, 0.4933, 0.8419, 0.5241, 0.3644, 0.4834, 0.6657, 0.5144], [0.7188, 0.4813, 0.8878, 0.4690, 0.3864, 0.4804, 0.6865, 0.5008], [0.7292, 0.4875, 0.8690, 0.4671, 0.3839, 0.5779, 0.5837, 0.4822], [0.0061, 0.0024, 0.7012, 0.2367, 0.3962, 0.2293, 0.4864, 0.4885], [0.7251, 0.4817, 0.8466, 0.4170, 0.3446, 0.3731, 0.5056, 0.4686], [0.6953, 0.4570, 0.8304, 0.2385, 0.5326, 0.2588, 0.6773, 0.4800]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6207, 0.4081, 0.7663, 0.2067, 0.3963, 0.3200, 0.6313, 0.5300], [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413], [0.6314, 0.4107, 0.8750, 0.5100, 0.3787, 0.4900, 0.7121, 0.5864], [0.6332, 0.4118, 0.9237, 0.4267, 0.4013, 0.4733, 0.7525, 0.5436], [0.6250, 0.4103, 0.8950, 0.4400, 0.3913, 0.5650, 0.6050, 0.5133], [0.0000, 0.0000, 0.6900, 0.1917, 0.3938, 0.2367, 0.5240, 0.5246], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882], [0.6300, 0.4133, 0.8537, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413]]], device='cuda:0') loss_train_step before backward: tensor(0.0023, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0023, device='cuda:0', grad_fn=) loss_train: 0.004649204900488257 step: 3 running loss: 0.0015497349668294191 Train Steps: 3/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767], [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.6099, 0.4030, 0.8638, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388], [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350], [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6228, 0.4140, 0.8659, 0.4301, 0.3465, 0.3910, 0.6165, 0.5447], [0.6294, 0.4215, 0.8306, 0.2009, 0.5087, 0.2778, 0.7287, 0.4982], [0.5685, 0.3807, 0.8171, 0.4263, 0.3248, 0.3840, 0.5332, 0.5097], [0.6234, 0.4169, 0.8464, 0.4805, 0.4680, 0.4954, 0.5454, 0.4970], [0.6373, 0.4156, 0.8276, 0.4305, 0.3745, 0.5263, 0.6380, 0.4825], [0.5885, 0.3843, 0.8212, 0.4443, 0.3774, 0.5082, 0.5607, 0.4995], [0.5832, 0.3737, 0.8340, 0.4706, 0.3457, 0.4484, 0.5524, 0.4831], [0.6135, 0.4110, 0.8132, 0.1869, 0.4392, 0.1840, 0.6581, 0.4998]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767], [0.6300, 0.4133, 0.8537, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.6151, 0.4125, 0.8737, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.6098, 0.4030, 0.8637, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388], [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6143, 0.3982, 0.8650, 0.4883, 0.3913, 0.4317, 0.5315, 0.5350], [0.6266, 0.4065, 0.8313, 0.2467, 0.4787, 0.1733, 0.6313, 0.5133]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.00562284019542858 step: 4 running loss: 0.001405710048857145 Train Steps: 4/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148], [ nan, nan, 0.8488, 0.2300, 0.5563, 0.2100, 0.7390, 0.5679], [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092], [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285], [0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272], [ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933], [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667], [0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6993, 0.4755, 0.8851, 0.5005, 0.3740, 0.4976, 0.6179, 0.5072], [0.4512, 0.3040, 0.8451, 0.2089, 0.4832, 0.2711, 0.7245, 0.5286], [0.6676, 0.4357, 0.8666, 0.4902, 0.3854, 0.5505, 0.5973, 0.4928], [0.6404, 0.4130, 0.8976, 0.3328, 0.4247, 0.2651, 0.6338, 0.5441], [0.6596, 0.4259, 0.8557, 0.3324, 0.3837, 0.2897, 0.5354, 0.5428], [0.2002, 0.1357, 0.6766, 0.1868, 0.3779, 0.2244, 0.5412, 0.5682], [0.6832, 0.4638, 0.8442, 0.3196, 0.3664, 0.3608, 0.6115, 0.5717], [0.5639, 0.3835, 0.7373, 0.2500, 0.3418, 0.3510, 0.5388, 0.5312]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148], [0.0000, 0.0000, 0.8487, 0.2300, 0.5562, 0.2100, 0.7390, 0.5679], [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092], [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285], [0.6186, 0.4080, 0.8625, 0.3483, 0.3787, 0.2650, 0.5320, 0.5272], [0.0000, 0.0000, 0.6688, 0.2513, 0.4112, 0.2117, 0.5193, 0.5933], [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5987, 0.5667], [0.6109, 0.4041, 0.6975, 0.3167, 0.3512, 0.3383, 0.5153, 0.5319]]], device='cuda:0') loss_train_step before backward: tensor(0.0065, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0065, device='cuda:0', grad_fn=) loss_train: 0.01212452695472166 step: 5 running loss: 0.002424905390944332 Train Steps: 5/90 Loss: 0.0024 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835], [0.6188, 0.4099, 0.7400, 0.2433, 0.3962, 0.2750, 0.6162, 0.5467], [0.6353, 0.4128, 0.9138, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220], [0.6087, 0.3976, 0.8337, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300], [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [0.6075, 0.4007, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5911, 0.3556, 0.8988, 0.4392, 0.3467, 0.4394, 0.5725, 0.5445], [0.6665, 0.4088, 0.9202, 0.3296, 0.4442, 0.2362, 0.6043, 0.5236], [0.5699, 0.3660, 0.7474, 0.1740, 0.3935, 0.3022, 0.5830, 0.5746], [0.5511, 0.3537, 0.9361, 0.3323, 0.4631, 0.3199, 0.7435, 0.5976], [0.3874, 0.2319, 0.7176, 0.1733, 0.4439, 0.1785, 0.5358, 0.5427], [0.6752, 0.4473, 0.8615, 0.3541, 0.3713, 0.3553, 0.5935, 0.5718], [0.5413, 0.3587, 0.8534, 0.5498, 0.3760, 0.4432, 0.5977, 0.5819], [0.6556, 0.4265, 0.8246, 0.4529, 0.4112, 0.5356, 0.5225, 0.5649]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6236, 0.3965, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6273, 0.4110, 0.8900, 0.3817, 0.4187, 0.2167, 0.5858, 0.4835], [0.6188, 0.4099, 0.7400, 0.2433, 0.3963, 0.2750, 0.6162, 0.5467], [0.6353, 0.4128, 0.9137, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220], [0.6087, 0.3976, 0.8338, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300], [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [0.6075, 0.4006, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280]]], device='cuda:0') loss_train_step before backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train: 0.014573542110156268 step: 6 running loss: 0.0024289236850260445 Train Steps: 6/90 Loss: 0.0024 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [ nan, nan, 0.7648, 0.2722, 0.3962, 0.2183, 0.5060, 0.5422], [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320], [0.6252, 0.4158, 0.8988, 0.4083, 0.3788, 0.4783, 0.6225, 0.5633], [0.6230, 0.4113, 0.7213, 0.1983, 0.4325, 0.2367, 0.6262, 0.5400], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6546, 0.4125, 0.9096, 0.4107, 0.3861, 0.2479, 0.6083, 0.5739], [ 0.0055, -0.0085, 0.7559, 0.2674, 0.3948, 0.2246, 0.4803, 0.5537], [ 0.5819, 0.3834, 0.7958, 0.2445, 0.4335, 0.2698, 0.6186, 0.5927], [ 0.7106, 0.4729, 0.9110, 0.4499, 0.3817, 0.4728, 0.6339, 0.5947], [ 0.5566, 0.3425, 0.7486, 0.2117, 0.4368, 0.2164, 0.6001, 0.5948], [ 0.5981, 0.3679, 0.8987, 0.3423, 0.4487, 0.3167, 0.6891, 0.5739], [ 0.6130, 0.3963, 0.7683, 0.2571, 0.4715, 0.2283, 0.5340, 0.6134], [ 0.6651, 0.4216, 0.9075, 0.5245, 0.4099, 0.5539, 0.5701, 0.5658]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6277, 0.4117, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.0000, 0.0000, 0.7648, 0.2722, 0.3963, 0.2183, 0.5060, 0.5422], [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320], [0.6252, 0.4158, 0.8988, 0.4083, 0.3787, 0.4783, 0.6225, 0.5633], [0.6230, 0.4113, 0.7212, 0.1983, 0.4325, 0.2367, 0.6263, 0.5400], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6229, 0.4198, 0.7663, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.015636640891898423 step: 7 running loss: 0.0022338058416997747 Train Steps: 7/90 Loss: 0.0022 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6202, 0.4079, 0.8025, 0.2500, 0.3763, 0.3217, 0.6125, 0.5533], [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5787, 0.5550], [0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767], [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167], [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899], [0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5737, 0.5700]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5450, 0.3389, 0.9121, 0.4218, 0.4478, 0.3192, 0.7040, 0.5919], [0.4682, 0.2779, 0.7069, 0.2023, 0.4132, 0.1661, 0.5391, 0.5515], [0.5974, 0.3781, 0.8077, 0.2790, 0.4024, 0.2788, 0.5935, 0.5618], [0.5002, 0.3248, 0.8132, 0.3105, 0.3683, 0.3401, 0.5391, 0.5600], [0.5790, 0.3737, 0.7905, 0.2610, 0.4246, 0.2579, 0.5991, 0.5909], [0.5536, 0.3923, 0.8460, 0.3414, 0.3876, 0.2783, 0.5599, 0.5604], [0.5235, 0.3519, 0.9101, 0.4799, 0.4183, 0.3825, 0.6948, 0.6061], [0.6200, 0.4087, 0.8988, 0.5803, 0.4056, 0.4541, 0.5931, 0.5973]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6054, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6202, 0.4079, 0.8025, 0.2500, 0.3762, 0.3217, 0.6125, 0.5533], [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5788, 0.5550], [0.6230, 0.4152, 0.7588, 0.2283, 0.4013, 0.2883, 0.6200, 0.5767], [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167], [0.6346, 0.4144, 0.9087, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899], [0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5738, 0.5700]]], device='cuda:0') loss_train_step before backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train: 0.017651385918725282 step: 8 running loss: 0.0022064232398406602 Train Steps: 8/90 Loss: 0.0022 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246], [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197], [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124], [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6198, 0.4101, 0.8838, 0.5283, 0.3763, 0.5267, 0.5913, 0.5567], [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5988, 0.5283], [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783], [ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.0428, 0.0314, 0.7288, 0.2345, 0.4140, 0.2244, 0.5735, 0.5468], [0.7112, 0.4756, 0.9116, 0.5149, 0.4737, 0.5195, 0.6098, 0.5295], [0.6298, 0.4101, 0.8339, 0.2671, 0.4087, 0.2395, 0.6392, 0.5416], [0.6986, 0.4763, 0.8342, 0.3661, 0.4985, 0.2606, 0.5705, 0.6165], [0.7682, 0.4894, 0.9038, 0.5506, 0.3864, 0.5193, 0.6369, 0.5682], [0.7155, 0.4652, 0.8983, 0.5525, 0.3863, 0.4746, 0.6439, 0.5403], [0.6947, 0.4640, 0.9098, 0.4999, 0.3746, 0.4000, 0.5840, 0.5835], [0.1006, 0.0617, 0.7249, 0.2260, 0.4266, 0.1740, 0.5654, 0.5623]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.6900, 0.1917, 0.3938, 0.2367, 0.5240, 0.5246], [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197], [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124], [0.6299, 0.4303, 0.7962, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6198, 0.4101, 0.8838, 0.5283, 0.3762, 0.5267, 0.5913, 0.5567], [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5987, 0.5283], [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783], [0.0000, 0.0000, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729]]], device='cuda:0') loss_train_step before backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train: 0.019613695156294852 step: 9 running loss: 0.002179299461810539 Train Steps: 9/90 Loss: 0.0022 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400], [0.6286, 0.4055, 0.9000, 0.4717, 0.3763, 0.4683, 0.7018, 0.5494], [0.6160, 0.4093, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808], [0.6224, 0.4061, 0.8988, 0.4300, 0.3838, 0.4750, 0.6112, 0.5483], [0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927], [0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609], [0.6188, 0.4099, 0.7400, 0.2433, 0.3962, 0.2750, 0.6162, 0.5467], [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6030, 0.4097, 0.8930, 0.4494, 0.3813, 0.4447, 0.6150, 0.5515], [0.6140, 0.4132, 0.9168, 0.5096, 0.3778, 0.4366, 0.7009, 0.5404], [0.5959, 0.4015, 0.8532, 0.4779, 0.3983, 0.4302, 0.5673, 0.5675], [0.6657, 0.4229, 0.9223, 0.4793, 0.3905, 0.4876, 0.6136, 0.5228], [0.6222, 0.3936, 0.8991, 0.5774, 0.3917, 0.4128, 0.6241, 0.4926], [0.5483, 0.3770, 0.8813, 0.4745, 0.4251, 0.5456, 0.6047, 0.5475], [0.5623, 0.3769, 0.7501, 0.2297, 0.4064, 0.2503, 0.5999, 0.5592], [0.5206, 0.3534, 0.7160, 0.2273, 0.4247, 0.2127, 0.5603, 0.5618]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6137, 0.5400], [0.6286, 0.4055, 0.9000, 0.4717, 0.3762, 0.4683, 0.7018, 0.5494], [0.6160, 0.4092, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808], [0.6224, 0.4061, 0.8988, 0.4300, 0.3837, 0.4750, 0.6112, 0.5483], [0.6321, 0.4048, 0.8737, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927], [0.6064, 0.4019, 0.8650, 0.4517, 0.4038, 0.5367, 0.5703, 0.5609], [0.6188, 0.4099, 0.7400, 0.2433, 0.3963, 0.2750, 0.6162, 0.5467], [0.6072, 0.4029, 0.7038, 0.2150, 0.3913, 0.2267, 0.5516, 0.5507]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.020335583365522325 step: 10 running loss: 0.0020335583365522327 Train Steps: 10/90 Loss: 0.0020 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8850, 0.2817, 0.5112, 0.2183, 0.7184, 0.5436], [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6188, 0.5433], [0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471], [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123], [0.6222, 0.4169, 0.8638, 0.5650, 0.4313, 0.4783, 0.5637, 0.5633], [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050], [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.1967, 0.1461, 0.8668, 0.2789, 0.5008, 0.2300, 0.7540, 0.5135], [0.6227, 0.4107, 0.8813, 0.4764, 0.3725, 0.5147, 0.6305, 0.5193], [0.6610, 0.4424, 0.8588, 0.4559, 0.3658, 0.4232, 0.5012, 0.5153], [0.5926, 0.3841, 0.7065, 0.2488, 0.4126, 0.2771, 0.6147, 0.5956], [0.6309, 0.4225, 0.8329, 0.3700, 0.3376, 0.4131, 0.5710, 0.5053], [0.5498, 0.3652, 0.8582, 0.5761, 0.4092, 0.4807, 0.5568, 0.5532], [0.6210, 0.4160, 0.8617, 0.4565, 0.3626, 0.3940, 0.5778, 0.4699], [0.6542, 0.4385, 0.8640, 0.4484, 0.4057, 0.5174, 0.6278, 0.5312]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8850, 0.2817, 0.5113, 0.2183, 0.7184, 0.5436], [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6187, 0.5433], [0.6127, 0.4084, 0.8700, 0.4467, 0.3988, 0.4317, 0.5013, 0.5471], [0.6193, 0.4078, 0.7287, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6213, 0.4131, 0.8438, 0.3550, 0.3512, 0.4400, 0.5716, 0.5123], [0.6222, 0.4169, 0.8637, 0.5650, 0.4313, 0.4783, 0.5638, 0.5633], [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050], [0.6189, 0.4049, 0.8888, 0.4417, 0.4212, 0.5200, 0.5987, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.02176640962716192 step: 11 running loss: 0.0019787645115601745 Train Steps: 11/90 Loss: 0.0020 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285], [0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050], [0.6199, 0.4065, 0.7598, 0.2385, 0.4317, 0.1981, 0.5933, 0.5221], [0.6277, 0.4103, 0.8087, 0.5717, 0.4188, 0.4750, 0.5663, 0.6083], [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5811, 0.3768, 0.9099, 0.4876, 0.3926, 0.5734, 0.6679, 0.5131], [0.5699, 0.4015, 0.8979, 0.4165, 0.3709, 0.4824, 0.6132, 0.5498], [0.5431, 0.3632, 0.7523, 0.2119, 0.4363, 0.2313, 0.5866, 0.4968], [0.5687, 0.3855, 0.8256, 0.5647, 0.3947, 0.4864, 0.5870, 0.5849], [0.6221, 0.4016, 0.9093, 0.4778, 0.4513, 0.5638, 0.6455, 0.5084], [0.4819, 0.3230, 0.6971, 0.3072, 0.3550, 0.3079, 0.5256, 0.5407], [0.5666, 0.3805, 0.8646, 0.4635, 0.3537, 0.3870, 0.5214, 0.5420], [0.5962, 0.4013, 0.7175, 0.2607, 0.3857, 0.2391, 0.5381, 0.5099]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6254, 0.3993, 0.8988, 0.4767, 0.3988, 0.5517, 0.6955, 0.5285], [0.6200, 0.4112, 0.8863, 0.4100, 0.3638, 0.4917, 0.6087, 0.6050], [0.6199, 0.4065, 0.7598, 0.2385, 0.4317, 0.1981, 0.5933, 0.5221], [0.6277, 0.4103, 0.8087, 0.5717, 0.4187, 0.4750, 0.5663, 0.6083], [0.6251, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6087, 0.5183], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5362, 0.5550]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.02287836465984583 step: 12 running loss: 0.0019065303883204858 Train Steps: 12/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6214, 0.4112, 0.7838, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083], [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5988, 0.5283], [0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467], [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750], [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6317, 0.4038, 0.8287, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986], [0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5935, 0.3896, 0.7540, 0.1949, 0.3524, 0.3019, 0.5652, 0.5206], [0.5673, 0.3691, 0.8442, 0.5159, 0.3690, 0.4971, 0.5622, 0.5308], [0.5962, 0.3964, 0.8590, 0.4969, 0.3963, 0.5442, 0.5967, 0.5409], [0.4966, 0.3253, 0.7469, 0.2914, 0.4054, 0.2778, 0.5003, 0.5784], [0.5631, 0.3603, 0.8656, 0.3860, 0.4165, 0.3810, 0.6919, 0.5644], [0.5929, 0.3764, 0.8253, 0.5604, 0.3623, 0.4864, 0.5951, 0.5051], [0.5627, 0.3630, 0.8148, 0.3790, 0.3547, 0.4484, 0.5368, 0.5210], [0.5737, 0.4015, 0.8589, 0.4194, 0.4251, 0.5275, 0.5585, 0.5323]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6214, 0.4112, 0.7837, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083], [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5987, 0.5283], [0.6199, 0.4059, 0.8700, 0.4900, 0.4162, 0.5000, 0.6162, 0.5467], [0.6182, 0.4099, 0.7812, 0.3000, 0.3938, 0.2367, 0.5325, 0.5750], [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6317, 0.4038, 0.8288, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986], [0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.02397310920059681 step: 13 running loss: 0.0018440853231228315 Train Steps: 13/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6277, 0.4083, 0.8350, 0.2717, 0.4562, 0.1800, 0.5918, 0.4878], [0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433], [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5787, 0.5550], [0.6059, 0.4002, 0.7562, 0.2767, 0.3538, 0.3033, 0.5529, 0.5455], [0.6307, 0.4029, 0.8650, 0.5200, 0.3763, 0.4017, 0.7311, 0.5366], [0.6296, 0.4045, 0.9138, 0.4100, 0.4232, 0.4242, 0.7422, 0.5297], [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5830, 0.3639, 0.7772, 0.2662, 0.4585, 0.2378, 0.5365, 0.5227], [0.5380, 0.3593, 0.7519, 0.2464, 0.3452, 0.3739, 0.5676, 0.5757], [0.5398, 0.3570, 0.7709, 0.2866, 0.3395, 0.4024, 0.5126, 0.5589], [0.5468, 0.3776, 0.7313, 0.2644, 0.3622, 0.3308, 0.5612, 0.5544], [0.5890, 0.3941, 0.8587, 0.5155, 0.3783, 0.4066, 0.6172, 0.5367], [0.6245, 0.4113, 0.8901, 0.4214, 0.3851, 0.4393, 0.6568, 0.5655], [0.5801, 0.3806, 0.8156, 0.2517, 0.4710, 0.2306, 0.5710, 0.5322], [0.5394, 0.3569, 0.7571, 0.2928, 0.4019, 0.3020, 0.5815, 0.5393]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6277, 0.4083, 0.8350, 0.2717, 0.4563, 0.1800, 0.5918, 0.4878], [0.6175, 0.4091, 0.7862, 0.2800, 0.3638, 0.3583, 0.6187, 0.5433], [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5788, 0.5550], [0.6059, 0.4002, 0.7563, 0.2767, 0.3537, 0.3033, 0.5529, 0.5455], [0.6307, 0.4029, 0.8650, 0.5200, 0.3762, 0.4017, 0.7311, 0.5366], [0.6296, 0.4045, 0.9137, 0.4100, 0.4232, 0.4242, 0.7422, 0.5297], [0.6275, 0.4111, 0.8462, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6201, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.025551110040396452 step: 14 running loss: 0.0018250792885997466 Train Steps: 14/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6250, 0.4013, 0.8525, 0.5417, 0.4037, 0.5117, 0.6325, 0.5017], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6246, 0.4008, 0.8757, 0.5088, 0.4101, 0.5392, 0.6644, 0.5133], [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233], [0.6267, 0.4080, 0.8438, 0.2633, 0.4763, 0.1800, 0.6259, 0.5240], [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309], [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200], [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5966, 0.3827, 0.8279, 0.5126, 0.3845, 0.5103, 0.5687, 0.5114], [0.6408, 0.4424, 0.7315, 0.2028, 0.3657, 0.2999, 0.5820, 0.5476], [0.6043, 0.3967, 0.8429, 0.4752, 0.3932, 0.5304, 0.6078, 0.5437], [0.5796, 0.3996, 0.8783, 0.4260, 0.3826, 0.5202, 0.6104, 0.5483], [0.6333, 0.4238, 0.8142, 0.2279, 0.4601, 0.1983, 0.5953, 0.5282], [0.6257, 0.4307, 0.8559, 0.4417, 0.3964, 0.3713, 0.5981, 0.5385], [0.6219, 0.4237, 0.8560, 0.4450, 0.3656, 0.4629, 0.5451, 0.5470], [0.5676, 0.3878, 0.8360, 0.4887, 0.3789, 0.5326, 0.6601, 0.5957]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6250, 0.4013, 0.8525, 0.5417, 0.4038, 0.5117, 0.6325, 0.5017], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6246, 0.4008, 0.8757, 0.5088, 0.4101, 0.5392, 0.6644, 0.5133], [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233], [0.6267, 0.4080, 0.8438, 0.2633, 0.4762, 0.1800, 0.6259, 0.5240], [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309], [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200], [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.026409554469864815 step: 15 running loss: 0.0017606369646576544 Train Steps: 15/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6179, 0.4040, 0.7412, 0.1850, 0.3825, 0.2783, 0.5837, 0.5600], [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050], [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5237, 0.3421, 0.7651, 0.2634, 0.4441, 0.2637, 0.6147, 0.5645], [0.6785, 0.4316, 0.8797, 0.5774, 0.3911, 0.4217, 0.5852, 0.5596], [0.6267, 0.4133, 0.8920, 0.5273, 0.3849, 0.4389, 0.5403, 0.5284], [0.6243, 0.3974, 0.8944, 0.5076, 0.4023, 0.5420, 0.6750, 0.5255], [0.7272, 0.4969, 0.7238, 0.2236, 0.3934, 0.2854, 0.6170, 0.5553], [0.6593, 0.4337, 0.7901, 0.2948, 0.3699, 0.3122, 0.6406, 0.5055], [0.6604, 0.4426, 0.7870, 0.2455, 0.3581, 0.3741, 0.6380, 0.5378], [0.5822, 0.3795, 0.8651, 0.4301, 0.3749, 0.4536, 0.6211, 0.5225]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5688, 0.5567], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6179, 0.4040, 0.7412, 0.1850, 0.3825, 0.2783, 0.5838, 0.5600], [0.6135, 0.3994, 0.7912, 0.3050, 0.3625, 0.3050, 0.5838, 0.5050], [0.6197, 0.4051, 0.7812, 0.2650, 0.3512, 0.4050, 0.6112, 0.5500], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.02758348052157089 step: 16 running loss: 0.0017239675325981807 Train Steps: 16/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6304, 0.4024, 0.8925, 0.4800, 0.3937, 0.4817, 0.7485, 0.5297], [0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290], [0.6206, 0.4001, 0.8900, 0.3933, 0.3588, 0.3567, 0.5837, 0.5083], [0.6122, 0.3993, 0.8738, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927], [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567], [0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7156, 0.4425, 0.8823, 0.4837, 0.3652, 0.4849, 0.7299, 0.5256], [0.7126, 0.4818, 0.7093, 0.2302, 0.3714, 0.2479, 0.5777, 0.5423], [0.7234, 0.4645, 0.7586, 0.2182, 0.4175, 0.2038, 0.6208, 0.5105], [0.6508, 0.4159, 0.7700, 0.2354, 0.4291, 0.2011, 0.6831, 0.5272], [0.7000, 0.4389, 0.9155, 0.4157, 0.3596, 0.3509, 0.5941, 0.5133], [0.7093, 0.4505, 0.9017, 0.4934, 0.4225, 0.5042, 0.5203, 0.5092], [0.1840, 0.1324, 0.7520, 0.2711, 0.3754, 0.2328, 0.5478, 0.5738], [0.6669, 0.4338, 0.9112, 0.4935, 0.4624, 0.4776, 0.6183, 0.5521]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6304, 0.4024, 0.8925, 0.4800, 0.3938, 0.4817, 0.7485, 0.5297], [0.6187, 0.4103, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6202, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290], [0.6206, 0.4001, 0.8900, 0.3933, 0.3587, 0.3567, 0.5838, 0.5083], [0.6122, 0.3993, 0.8737, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927], [0.0000, 0.0000, 0.7515, 0.2708, 0.3988, 0.2267, 0.5163, 0.5567], [0.6112, 0.4029, 0.8637, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567]]], device='cuda:0') loss_train_step before backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train: 0.029665038164239377 step: 17 running loss: 0.0017450022449552575 Train Steps: 17/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [0.6289, 0.4032, 0.8419, 0.5446, 0.4075, 0.5017, 0.6312, 0.5117], [0.6118, 0.4052, 0.8463, 0.3917, 0.3538, 0.3450, 0.5053, 0.5593], [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917], [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378], [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689], [0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6093, 0.3796, 0.8609, 0.5016, 0.3703, 0.3870, 0.6474, 0.5185], [0.7249, 0.4409, 0.8565, 0.4559, 0.3879, 0.4894, 0.6929, 0.4906], [0.6259, 0.4103, 0.8489, 0.3080, 0.3685, 0.3349, 0.5577, 0.5404], [0.6718, 0.4446, 0.8509, 0.4925, 0.4141, 0.4078, 0.6081, 0.5764], [0.6810, 0.4354, 0.8846, 0.3431, 0.4289, 0.2577, 0.6137, 0.5327], [0.6487, 0.4090, 0.8828, 0.5071, 0.3807, 0.4111, 0.6443, 0.5090], [0.6511, 0.4149, 0.7971, 0.2856, 0.3502, 0.3732, 0.6101, 0.5407], [0.6062, 0.3869, 0.8544, 0.4510, 0.4083, 0.3646, 0.6009, 0.5720]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [0.6289, 0.4031, 0.8419, 0.5446, 0.4075, 0.5017, 0.6313, 0.5117], [0.6118, 0.4052, 0.8462, 0.3917, 0.3537, 0.3450, 0.5053, 0.5593], [0.6212, 0.4159, 0.8675, 0.5783, 0.4087, 0.4317, 0.5612, 0.5917], [0.6226, 0.4098, 0.8913, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378], [0.6102, 0.4001, 0.7738, 0.3583, 0.3462, 0.3800, 0.5524, 0.5689], [0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.03151935857022181 step: 18 running loss: 0.001751075476123434 Train Steps: 18/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650], [0.6081, 0.3950, 0.8538, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954], [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717], [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617], [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938], [0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6285, 0.4289, 0.8369, 0.3992, 0.3666, 0.3394, 0.5768, 0.5714], [0.5382, 0.3461, 0.9028, 0.4570, 0.3757, 0.4687, 0.5894, 0.5244], [0.6505, 0.4122, 0.6919, 0.2711, 0.3828, 0.2432, 0.5745, 0.5503], [0.6192, 0.3904, 0.8273, 0.2519, 0.4510, 0.1801, 0.6629, 0.5030], [0.6138, 0.3868, 0.8164, 0.3509, 0.3933, 0.3060, 0.6069, 0.5410], [0.6278, 0.3900, 0.8875, 0.5938, 0.4004, 0.4714, 0.6583, 0.5202], [0.6803, 0.4378, 0.8683, 0.3065, 0.4459, 0.2154, 0.6347, 0.5347], [0.6068, 0.3959, 0.8165, 0.3017, 0.3987, 0.2889, 0.6404, 0.5753]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650], [0.6081, 0.3950, 0.8537, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954], [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717], [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617], [0.6334, 0.4037, 0.8637, 0.5733, 0.4013, 0.4717, 0.6369, 0.4938], [0.6250, 0.4131, 0.8687, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6200, 0.4098, 0.8238, 0.2917, 0.4013, 0.2967, 0.6000, 0.5683]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.032034559291787446 step: 19 running loss: 0.0016860294364098656 Train Steps: 19/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204], [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166], [0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693], [0.6136, 0.4117, 0.8700, 0.5167, 0.4188, 0.5083, 0.5147, 0.5495], [0.6250, 0.4110, 0.7238, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6260, 0.4120, 0.8013, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895], [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5812, 0.3663, 0.8780, 0.5663, 0.4113, 0.5094, 0.6484, 0.5449], [0.6205, 0.3792, 0.8964, 0.5088, 0.3694, 0.5186, 0.6588, 0.5418], [0.5394, 0.3596, 0.8567, 0.5064, 0.3538, 0.4711, 0.5429, 0.5481], [0.5951, 0.3821, 0.9107, 0.5731, 0.4070, 0.5004, 0.5472, 0.5524], [0.6014, 0.3876, 0.7242, 0.2879, 0.4193, 0.1686, 0.5552, 0.5635], [0.6758, 0.4629, 0.8608, 0.2929, 0.4783, 0.1708, 0.6264, 0.5358], [0.7078, 0.4596, 0.8143, 0.2689, 0.4817, 0.1511, 0.6334, 0.5194], [0.6756, 0.4203, 0.8839, 0.5290, 0.4115, 0.5091, 0.5758, 0.5183]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204], [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166], [0.6038, 0.3946, 0.8413, 0.4883, 0.3562, 0.4550, 0.5266, 0.4693], [0.6136, 0.4117, 0.8700, 0.5167, 0.4187, 0.5083, 0.5147, 0.5495], [0.6250, 0.4110, 0.7237, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6259, 0.4120, 0.8012, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895], [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.03298321762122214 step: 20 running loss: 0.0016491608810611069 Train Steps: 20/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050], [0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626], [0.6152, 0.4131, 0.6863, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133], [0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133], [0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.1661, 0.0989, 0.8938, 0.3742, 0.4654, 0.1780, 0.6484, 0.5681], [0.6539, 0.4123, 0.8830, 0.5821, 0.4425, 0.5081, 0.5056, 0.5100], [0.6937, 0.4245, 0.8564, 0.6342, 0.4042, 0.4477, 0.6047, 0.4869], [0.7473, 0.4797, 0.6982, 0.3150, 0.3798, 0.3226, 0.5214, 0.5388], [0.6735, 0.4346, 0.8389, 0.2927, 0.4800, 0.1470, 0.5898, 0.5247], [0.6678, 0.4201, 0.8428, 0.2941, 0.4032, 0.2558, 0.5571, 0.5247], [0.6108, 0.3926, 0.8709, 0.4782, 0.4101, 0.5413, 0.5579, 0.5436], [0.6580, 0.4124, 0.8757, 0.4204, 0.3782, 0.2969, 0.5750, 0.5194]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6097, 0.3988, 0.8650, 0.5250, 0.4212, 0.5200, 0.5675, 0.5050], [0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626], [0.6152, 0.4131, 0.6862, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [0.6266, 0.4065, 0.8313, 0.2467, 0.4787, 0.1733, 0.6313, 0.5133], [0.6205, 0.4015, 0.8350, 0.2717, 0.3988, 0.2550, 0.5788, 0.5133], [0.6064, 0.4019, 0.8650, 0.4517, 0.4038, 0.5367, 0.5703, 0.5609], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374]]], device='cuda:0') loss_train_step before backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train: 0.034703588928095996 step: 21 running loss: 0.001652551853718857 Train Steps: 21/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767], [0.6250, 0.4103, 0.8950, 0.4400, 0.3912, 0.5650, 0.6050, 0.5133], [0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6097, 0.3810, 0.8444, 0.3163, 0.4960, 0.2205, 0.6758, 0.5090], [0.6230, 0.4154, 0.7445, 0.3101, 0.4256, 0.1573, 0.5326, 0.5013], [0.6116, 0.4013, 0.8533, 0.5269, 0.4231, 0.5081, 0.5280, 0.5487], [0.5224, 0.3296, 0.8872, 0.4637, 0.4107, 0.5685, 0.5906, 0.5198], [0.4906, 0.3321, 0.8419, 0.4613, 0.4652, 0.5174, 0.5737, 0.5340], [0.6110, 0.3884, 0.8540, 0.6034, 0.4019, 0.4349, 0.5250, 0.5502], [0.6076, 0.3926, 0.8540, 0.3914, 0.3840, 0.2692, 0.5650, 0.5152], [0.6196, 0.4102, 0.8748, 0.4050, 0.4474, 0.2056, 0.5809, 0.5021]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6148, 0.4076, 0.8666, 0.4820, 0.4137, 0.5067, 0.5250, 0.5767], [0.6250, 0.4103, 0.8950, 0.4400, 0.3913, 0.5650, 0.6050, 0.5133], [0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.03575298085343093 step: 22 running loss: 0.0016251354933377695 Train Steps: 22/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377], [0.6205, 0.4062, 0.8337, 0.2683, 0.3675, 0.4283, 0.6338, 0.5250], [0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006], [0.6179, 0.4008, 0.8600, 0.4015, 0.3932, 0.2515, 0.5711, 0.5438], [0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131], [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5948, 0.3777, 0.8598, 0.5559, 0.4323, 0.4669, 0.5633, 0.5099], [0.5748, 0.3623, 0.8276, 0.3119, 0.4020, 0.3961, 0.6031, 0.4939], [0.5953, 0.3702, 0.8583, 0.6192, 0.4189, 0.4924, 0.5780, 0.4961], [0.5731, 0.3803, 0.8354, 0.3893, 0.4161, 0.2656, 0.5426, 0.5256], [0.6213, 0.3960, 0.8607, 0.4340, 0.3896, 0.4437, 0.5424, 0.4852], [0.5919, 0.3985, 0.8903, 0.4316, 0.4365, 0.2533, 0.5274, 0.5285], [0.5917, 0.3857, 0.8968, 0.4565, 0.4929, 0.4819, 0.5501, 0.5295], [0.6593, 0.4371, 0.7370, 0.2946, 0.4135, 0.2788, 0.5746, 0.5380]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377], [0.6205, 0.4062, 0.8338, 0.2683, 0.3675, 0.4283, 0.6338, 0.5250], [0.6310, 0.4017, 0.8562, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006], [0.6179, 0.4008, 0.8600, 0.4015, 0.3932, 0.2515, 0.5711, 0.5438], [0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131], [0.6226, 0.4098, 0.8913, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.03661173034925014 step: 23 running loss: 0.0015918143630108755 Train Steps: 23/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6132, 0.4037, 0.6963, 0.2217, 0.4100, 0.1950, 0.5395, 0.5175], [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038], [0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983], [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [ nan, nan, 0.7725, 0.2611, 0.3675, 0.2733, 0.5413, 0.5167], [0.6200, 0.4070, 0.8938, 0.4183, 0.3538, 0.4567, 0.6175, 0.5400], [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332], [0.6274, 0.4003, 0.8638, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6434, 0.4394, 0.7563, 0.2616, 0.4180, 0.1771, 0.5199, 0.4975], [0.6028, 0.4039, 0.8736, 0.4948, 0.4675, 0.5166, 0.5356, 0.5194], [0.6994, 0.4571, 0.8777, 0.5358, 0.4021, 0.4242, 0.5314, 0.5979], [0.6870, 0.4445, 0.9097, 0.3819, 0.3974, 0.4629, 0.6863, 0.5208], [0.1170, 0.0916, 0.7732, 0.2960, 0.3928, 0.2935, 0.5456, 0.5585], [0.7068, 0.4634, 0.9245, 0.4283, 0.4178, 0.5128, 0.6311, 0.5296], [0.6364, 0.4310, 0.8491, 0.5356, 0.3984, 0.5179, 0.6621, 0.5221], [0.7051, 0.4599, 0.8635, 0.5810, 0.4164, 0.4868, 0.6125, 0.5172]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6132, 0.4037, 0.6963, 0.2217, 0.4100, 0.1950, 0.5395, 0.5175], [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038], [0.6260, 0.4214, 0.8537, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983], [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [0.0000, 0.0000, 0.7725, 0.2611, 0.3675, 0.2733, 0.5412, 0.5167], [0.6200, 0.4070, 0.8938, 0.4183, 0.3537, 0.4567, 0.6175, 0.5400], [0.6289, 0.4019, 0.8112, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332], [0.6274, 0.4003, 0.8637, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.03813450282905251 step: 24 running loss: 0.001588937617877188 Train Steps: 24/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6198, 0.4115, 0.7762, 0.2717, 0.3713, 0.3200, 0.5837, 0.5683], [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6353, 0.4128, 0.9138, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5181, 0.3658, 0.7687, 0.2471, 0.3332, 0.3298, 0.5577, 0.5300], [0.5627, 0.3805, 0.8305, 0.2223, 0.4413, 0.2407, 0.6896, 0.5307], [0.4681, 0.3387, 0.8765, 0.3318, 0.4165, 0.3352, 0.7017, 0.5461], [0.5176, 0.3823, 0.8142, 0.4260, 0.4169, 0.3081, 0.5115, 0.5719], [0.5179, 0.3504, 0.8527, 0.4173, 0.4445, 0.5811, 0.5924, 0.5427], [0.5762, 0.4029, 0.8374, 0.5142, 0.3543, 0.3983, 0.5364, 0.5255], [0.5836, 0.3958, 0.8593, 0.4602, 0.3841, 0.5372, 0.6346, 0.4635], [0.5452, 0.3705, 0.7265, 0.3065, 0.4365, 0.2370, 0.5239, 0.5725]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6198, 0.4115, 0.7763, 0.2717, 0.3713, 0.3200, 0.5838, 0.5683], [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6353, 0.4128, 0.9137, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6184, 0.4079, 0.8838, 0.4617, 0.4837, 0.5650, 0.6175, 0.5850], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5412, 0.5683], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167]]], device='cuda:0') loss_train_step before backward: tensor(0.0023, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0023, device='cuda:0', grad_fn=) loss_train: 0.040397314936853945 step: 25 running loss: 0.0016158925974741579 Train Steps: 25/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633], [0.6228, 0.4119, 0.7938, 0.2233, 0.4674, 0.1773, 0.6188, 0.5433], [0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136], [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533], [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400], [0.6202, 0.4064, 0.7879, 0.2179, 0.4567, 0.1725, 0.5955, 0.5478], [0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196], [0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5563, 0.5533]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.0116, 0.0246, 0.7393, 0.2538, 0.3489, 0.2734, 0.5412, 0.5697], [0.6360, 0.4286, 0.8073, 0.2053, 0.4622, 0.1909, 0.6346, 0.5329], [0.5926, 0.4112, 0.8407, 0.3210, 0.3518, 0.5539, 0.6553, 0.5225], [0.6690, 0.4513, 0.8514, 0.4062, 0.3517, 0.3713, 0.6100, 0.5537], [0.6008, 0.3862, 0.8725, 0.4734, 0.4421, 0.6024, 0.6467, 0.5385], [0.5480, 0.3798, 0.7652, 0.2324, 0.4457, 0.1914, 0.6148, 0.5599], [0.6214, 0.4346, 0.8467, 0.4490, 0.4235, 0.3341, 0.5924, 0.6016], [0.6507, 0.4407, 0.8519, 0.5101, 0.3846, 0.4925, 0.6274, 0.5505]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633], [0.6228, 0.4119, 0.7937, 0.2233, 0.4674, 0.1773, 0.6187, 0.5433], [0.6132, 0.4118, 0.8200, 0.3633, 0.3562, 0.5400, 0.5787, 0.5136], [0.6252, 0.4162, 0.8662, 0.4467, 0.3625, 0.3567, 0.6037, 0.5533], [0.6185, 0.4098, 0.8838, 0.4900, 0.4538, 0.5800, 0.6288, 0.5400], [0.6202, 0.4064, 0.7879, 0.2179, 0.4567, 0.1725, 0.5955, 0.5478], [0.6278, 0.4253, 0.8875, 0.5017, 0.4112, 0.2750, 0.5413, 0.6196], [0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5562, 0.5533]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.04119387164246291 step: 26 running loss: 0.0015843796785562658 Train Steps: 26/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131], [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500], [0.6277, 0.4083, 0.8350, 0.2717, 0.4562, 0.1800, 0.5918, 0.4878], [0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050], [ nan, nan, 0.8488, 0.2300, 0.5563, 0.2100, 0.7390, 0.5679], [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833], [ nan, nan, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583], [0.6263, 0.4030, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7681, 0.5108, 0.8592, 0.4223, 0.3623, 0.4802, 0.6189, 0.5313], [0.7198, 0.4836, 0.8040, 0.2694, 0.3555, 0.4219, 0.6543, 0.5635], [0.5966, 0.3962, 0.8460, 0.2799, 0.4625, 0.2163, 0.6013, 0.5294], [0.6678, 0.4680, 0.8998, 0.4368, 0.3742, 0.5103, 0.6602, 0.6020], [0.1118, 0.0834, 0.8835, 0.2769, 0.5051, 0.2371, 0.7420, 0.5724], [0.6195, 0.4208, 0.7383, 0.3169, 0.3674, 0.2995, 0.5883, 0.6082], [0.1565, 0.1337, 0.7329, 0.2517, 0.4242, 0.1904, 0.5244, 0.5901], [0.7288, 0.4908, 0.9129, 0.4972, 0.3733, 0.5528, 0.6372, 0.5112]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131], [0.6197, 0.4051, 0.7812, 0.2650, 0.3512, 0.4050, 0.6112, 0.5500], [0.6277, 0.4083, 0.8350, 0.2717, 0.4563, 0.1800, 0.5918, 0.4878], [0.6200, 0.4112, 0.8863, 0.4100, 0.3638, 0.4917, 0.6087, 0.6050], [0.0000, 0.0000, 0.8487, 0.2300, 0.5562, 0.2100, 0.7390, 0.5679], [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833], [0.0000, 0.0000, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583], [0.6263, 0.4029, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771]]], device='cuda:0') loss_train_step before backward: tensor(0.0027, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0027, device='cuda:0', grad_fn=) loss_train: 0.043881157762371004 step: 27 running loss: 0.0016252280652730002 Train Steps: 27/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6092, 0.4001, 0.8638, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064], [0.6304, 0.4024, 0.8925, 0.4800, 0.3937, 0.4817, 0.7485, 0.5297], [0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343], [0.6263, 0.4030, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771], [0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689], [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417], [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5978, 0.3811, 0.8802, 0.4775, 0.4360, 0.5232, 0.5414, 0.5480], [0.6242, 0.3893, 0.8940, 0.4716, 0.3845, 0.4669, 0.7091, 0.5600], [0.5777, 0.3764, 0.8253, 0.5264, 0.3796, 0.4584, 0.6991, 0.5487], [0.5385, 0.3593, 0.9121, 0.4667, 0.3836, 0.5002, 0.6042, 0.5307], [0.5272, 0.3532, 0.7427, 0.2166, 0.3878, 0.2256, 0.5639, 0.5856], [0.5912, 0.3912, 0.8115, 0.3456, 0.3531, 0.3754, 0.5573, 0.6048], [0.5342, 0.3543, 0.8327, 0.2611, 0.4037, 0.2601, 0.6445, 0.5905], [0.5617, 0.3802, 0.9341, 0.4721, 0.4119, 0.5319, 0.7273, 0.6080]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6092, 0.4001, 0.8637, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064], [0.6304, 0.4024, 0.8925, 0.4800, 0.3938, 0.4817, 0.7485, 0.5297], [0.6311, 0.3998, 0.7975, 0.5767, 0.3837, 0.4850, 0.7327, 0.5343], [0.6263, 0.4029, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771], [0.6187, 0.4103, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6102, 0.4001, 0.7738, 0.3583, 0.3462, 0.3800, 0.5524, 0.5689], [0.6243, 0.4128, 0.7763, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417], [0.6339, 0.4102, 0.9087, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.04513794963713735 step: 28 running loss: 0.0016120696298977627 Train Steps: 28/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823], [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983], [0.6267, 0.4080, 0.8438, 0.2633, 0.4763, 0.1800, 0.6259, 0.5240], [0.6198, 0.4130, 0.8762, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103], [0.6131, 0.4037, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680], [0.6075, 0.4007, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4372, 0.2758, 0.7354, 0.1883, 0.3909, 0.1579, 0.5633, 0.5465], [0.6577, 0.4155, 0.8830, 0.5415, 0.3977, 0.4813, 0.6043, 0.5893], [0.5946, 0.3737, 0.8991, 0.3180, 0.4474, 0.3089, 0.7670, 0.5524], [0.5719, 0.3405, 0.8579, 0.5194, 0.4275, 0.4948, 0.6411, 0.5257], [0.4898, 0.3002, 0.8607, 0.2459, 0.4668, 0.1549, 0.6598, 0.5464], [0.5073, 0.3276, 0.8746, 0.3994, 0.3560, 0.4804, 0.6034, 0.5596], [0.5150, 0.3232, 0.6938, 0.2471, 0.3633, 0.2815, 0.5660, 0.5930], [0.5969, 0.3752, 0.8305, 0.4599, 0.3981, 0.4965, 0.5806, 0.5713]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6055, 0.4015, 0.7425, 0.2033, 0.4112, 0.1883, 0.5217, 0.4823], [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6219, 0.3934, 0.8687, 0.5267, 0.4313, 0.4967, 0.5987, 0.4983], [0.6267, 0.4080, 0.8438, 0.2633, 0.4762, 0.1800, 0.6259, 0.5240], [0.6198, 0.4130, 0.8763, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103], [0.6131, 0.4036, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680], [0.6075, 0.4006, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280]]], device='cuda:0') loss_train_step before backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train: 0.04752955154981464 step: 29 running loss: 0.0016389500534418843 Train Steps: 29/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391], [0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5613, 0.5900], [0.6353, 0.4128, 0.8488, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6133, 0.4066, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358], [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395], [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5865, 0.3450, 0.7975, 0.2635, 0.4103, 0.2449, 0.6346, 0.5429], [0.6215, 0.3994, 0.8307, 0.4658, 0.3353, 0.3533, 0.5554, 0.5437], [0.2566, 0.1509, 0.8737, 0.2176, 0.5307, 0.1564, 0.7105, 0.5645], [0.6468, 0.3995, 0.8455, 0.4233, 0.3543, 0.4662, 0.5918, 0.5448], [0.5884, 0.3784, 0.8735, 0.4595, 0.4237, 0.4856, 0.5844, 0.5492], [0.5698, 0.3733, 0.7134, 0.2465, 0.3807, 0.2194, 0.5472, 0.5236], [0.6155, 0.3682, 0.7660, 0.2156, 0.3963, 0.2318, 0.6159, 0.5211], [0.6217, 0.3869, 0.8140, 0.5425, 0.3716, 0.5056, 0.6957, 0.5169]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391], [0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5612, 0.5900], [0.6353, 0.4128, 0.8487, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6133, 0.4065, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358], [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395], [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366]]], device='cuda:0') loss_train_step before backward: tensor(0.0038, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0038, device='cuda:0', grad_fn=) loss_train: 0.05135433666873723 step: 30 running loss: 0.0017118112222912411 Train Steps: 30/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6261, 0.4066, 0.8325, 0.2150, 0.4763, 0.2667, 0.7002, 0.5633], [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391], [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246], [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5668, 0.3629, 0.8842, 0.4258, 0.3682, 0.4059, 0.5781, 0.5503], [0.6187, 0.3936, 0.7304, 0.2942, 0.4803, 0.2003, 0.5441, 0.5974], [0.6314, 0.3861, 0.8064, 0.1883, 0.4717, 0.2359, 0.7089, 0.5526], [0.6939, 0.4356, 0.8328, 0.5578, 0.3852, 0.4608, 0.6231, 0.4841], [0.6649, 0.4234, 0.8089, 0.5228, 0.3513, 0.4833, 0.6610, 0.4989], [0.6308, 0.4149, 0.8182, 0.2647, 0.4233, 0.2208, 0.5827, 0.5212], [0.1004, 0.0594, 0.6898, 0.1966, 0.4079, 0.2105, 0.5226, 0.5340], [0.6396, 0.4302, 0.8272, 0.4877, 0.4743, 0.4958, 0.5450, 0.5044]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6212, 0.4033, 0.8938, 0.4167, 0.3812, 0.4267, 0.5612, 0.5583], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6261, 0.4066, 0.8325, 0.2150, 0.4762, 0.2667, 0.7002, 0.5633], [0.6334, 0.4037, 0.8637, 0.5733, 0.4013, 0.4717, 0.6369, 0.4938], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6199, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391], [0.0000, 0.0000, 0.6900, 0.1917, 0.3938, 0.2367, 0.5240, 0.5246], [0.6102, 0.4005, 0.8687, 0.5100, 0.4812, 0.5400, 0.5404, 0.5064]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.052187160414177924 step: 31 running loss: 0.0016834567875541266 Train Steps: 31/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862], [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917], [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [ nan, nan, 0.8300, 0.3150, 0.3588, 0.3383, 0.5208, 0.5194]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6223, 0.4037, 0.8574, 0.4101, 0.3940, 0.3909, 0.6264, 0.5640], [0.6208, 0.4085, 0.8440, 0.4685, 0.4030, 0.3002, 0.5461, 0.5567], [0.6917, 0.4606, 0.6627, 0.2580, 0.4542, 0.2239, 0.5561, 0.5403], [0.6006, 0.3673, 0.7958, 0.3297, 0.3836, 0.3588, 0.6420, 0.5311], [0.6163, 0.3988, 0.8610, 0.4075, 0.4042, 0.4429, 0.6372, 0.4826], [0.6346, 0.4224, 0.7407, 0.3825, 0.3826, 0.3340, 0.5764, 0.5770], [0.5733, 0.3838, 0.8238, 0.4076, 0.3889, 0.3602, 0.5383, 0.4948], [0.4540, 0.2950, 0.7634, 0.2964, 0.3890, 0.3205, 0.5472, 0.5035]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6203, 0.4096, 0.8863, 0.4267, 0.3537, 0.4117, 0.6025, 0.5650], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5362, 0.5550], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862], [0.6179, 0.4118, 0.7278, 0.4237, 0.3587, 0.3400, 0.5675, 0.5917], [0.6082, 0.4024, 0.8737, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.0000, 0.0000, 0.8300, 0.3150, 0.3587, 0.3383, 0.5208, 0.5194]]], device='cuda:0') loss_train_step before backward: tensor(0.0053, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0053, device='cuda:0', grad_fn=) loss_train: 0.05747255013557151 step: 32 running loss: 0.0017960171917366097 Train Steps: 32/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389], [0.6275, 0.4157, 0.8337, 0.5800, 0.3763, 0.4200, 0.5547, 0.6125], [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869], [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6110, 0.3984, 0.8750, 0.4933, 0.4625, 0.4950, 0.5578, 0.5676]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6144, 0.4203, 0.7985, 0.3863, 0.3475, 0.2884, 0.5326, 0.5545], [0.5494, 0.3584, 0.7314, 0.2205, 0.3959, 0.2670, 0.5991, 0.5164], [0.6711, 0.4399, 0.7896, 0.5625, 0.3788, 0.4291, 0.5623, 0.5686], [0.5554, 0.3771, 0.8378, 0.5346, 0.3878, 0.4062, 0.5128, 0.5303], [0.6204, 0.4052, 0.8259, 0.4745, 0.4359, 0.4610, 0.5449, 0.5183], [0.6332, 0.4169, 0.7866, 0.2161, 0.5003, 0.2074, 0.6308, 0.4744], [0.5956, 0.4142, 0.8529, 0.4707, 0.4455, 0.5582, 0.5868, 0.5015], [0.6420, 0.4306, 0.8381, 0.4839, 0.4555, 0.4854, 0.5397, 0.5353]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389], [0.6275, 0.4157, 0.8338, 0.5800, 0.3762, 0.4200, 0.5547, 0.6125], [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6276, 0.4095, 0.8238, 0.2250, 0.4663, 0.1783, 0.6171, 0.4869], [0.6155, 0.4124, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6110, 0.3984, 0.8750, 0.4933, 0.4625, 0.4950, 0.5578, 0.5676]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.058107397344429046 step: 33 running loss: 0.001760830222558456 Train Steps: 33/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5613, 0.5900], [0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567], [0.6200, 0.4049, 0.8638, 0.5617, 0.4125, 0.5100, 0.6013, 0.5317], [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578], [0.6030, 0.3969, 0.7988, 0.3917, 0.3450, 0.3667, 0.5266, 0.4700], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6827, 0.4496, 0.8256, 0.4827, 0.3709, 0.3676, 0.5322, 0.5571], [0.6665, 0.4444, 0.8460, 0.5793, 0.4669, 0.4921, 0.5802, 0.5466], [0.1378, 0.0869, 0.7551, 0.2594, 0.4420, 0.2424, 0.4932, 0.5484], [0.6830, 0.4586, 0.8622, 0.5962, 0.4376, 0.4989, 0.5673, 0.5239], [0.6869, 0.4526, 0.6950, 0.2222, 0.4388, 0.2058, 0.5477, 0.5366], [0.6921, 0.4603, 0.8187, 0.4140, 0.3610, 0.3888, 0.5339, 0.5119], [0.6835, 0.4319, 0.8662, 0.4696, 0.3985, 0.5319, 0.5821, 0.5164], [0.6697, 0.4426, 0.7883, 0.3365, 0.3718, 0.3618, 0.5934, 0.5166]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5612, 0.5900], [0.6229, 0.4066, 0.8512, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.0000, 0.0000, 0.7515, 0.2708, 0.3988, 0.2267, 0.5163, 0.5567], [0.6199, 0.4049, 0.8637, 0.5617, 0.4125, 0.5100, 0.6012, 0.5317], [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578], [0.6030, 0.3969, 0.7987, 0.3917, 0.3450, 0.3667, 0.5266, 0.4700], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.059539014648180455 step: 34 running loss: 0.0017511474896523664 Train Steps: 34/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083], [0.6307, 0.4029, 0.8650, 0.5200, 0.3763, 0.4017, 0.7311, 0.5366], [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5988, 0.5283], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6177, 0.4086, 0.8738, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6581, 0.4109, 0.8762, 0.4965, 0.3920, 0.4964, 0.5997, 0.5183], [0.5823, 0.3867, 0.8585, 0.4944, 0.4278, 0.4738, 0.5168, 0.5337], [0.6553, 0.4376, 0.8685, 0.4777, 0.3748, 0.3175, 0.5004, 0.5822], [0.6329, 0.4304, 0.8425, 0.5011, 0.4032, 0.4497, 0.5394, 0.5825], [0.6444, 0.4426, 0.8537, 0.5210, 0.4073, 0.3890, 0.6372, 0.5359], [0.6312, 0.4139, 0.8603, 0.5559, 0.4056, 0.4757, 0.5384, 0.5498], [0.6495, 0.4212, 0.8286, 0.5655, 0.3784, 0.5000, 0.6258, 0.5397], [0.6462, 0.4144, 0.8579, 0.4170, 0.3991, 0.5355, 0.5896, 0.5531]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6186, 0.4060, 0.8750, 0.5050, 0.3537, 0.4367, 0.5813, 0.6083], [0.6307, 0.4029, 0.8650, 0.5200, 0.3762, 0.4017, 0.7311, 0.5366], [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5987, 0.5283], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6177, 0.4085, 0.8737, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.06021460593910888 step: 35 running loss: 0.001720417312545968 Train Steps: 35/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517], [0.6261, 0.4066, 0.8325, 0.2150, 0.4763, 0.2667, 0.7002, 0.5633], [0.6092, 0.4001, 0.8638, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064], [0.6230, 0.4113, 0.7213, 0.1983, 0.4325, 0.2367, 0.6262, 0.5400], [0.6126, 0.4039, 0.8237, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138], [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750], [0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131], [0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6218, 0.4113, 0.7499, 0.2620, 0.3945, 0.2552, 0.5913, 0.5554], [0.6403, 0.4063, 0.8377, 0.2635, 0.4469, 0.2516, 0.6621, 0.5564], [0.5643, 0.3632, 0.8689, 0.5466, 0.3856, 0.5353, 0.5056, 0.4891], [0.6050, 0.4061, 0.7295, 0.2546, 0.4117, 0.2465, 0.6089, 0.5497], [0.5580, 0.3757, 0.8355, 0.4600, 0.3196, 0.3810, 0.5580, 0.6032], [0.5661, 0.3794, 0.7924, 0.3434, 0.3821, 0.2560, 0.4986, 0.5691], [0.5372, 0.3464, 0.8553, 0.4697, 0.3184, 0.4471, 0.5288, 0.5201], [0.5403, 0.3731, 0.8659, 0.6326, 0.4140, 0.4998, 0.5535, 0.5452]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6200, 0.4071, 0.7337, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517], [0.6261, 0.4066, 0.8325, 0.2150, 0.4762, 0.2667, 0.7002, 0.5633], [0.6092, 0.4001, 0.8637, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064], [0.6230, 0.4113, 0.7212, 0.1983, 0.4325, 0.2367, 0.6263, 0.5400], [0.6126, 0.4038, 0.8238, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138], [0.6182, 0.4099, 0.7812, 0.3000, 0.3938, 0.2367, 0.5325, 0.5750], [0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131], [0.6229, 0.4066, 0.8512, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.06144377455348149 step: 36 running loss: 0.0017067715153744859 Train Steps: 36/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767], [ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385], [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166], [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5900, 0.3929, 0.9066, 0.5424, 0.3696, 0.5060, 0.5424, 0.5662], [0.2800, 0.1911, 0.7943, 0.3393, 0.3765, 0.2683, 0.5316, 0.5769], [0.5682, 0.3823, 0.7468, 0.2356, 0.3920, 0.2099, 0.5797, 0.5402], [0.5947, 0.3937, 0.8851, 0.5474, 0.4331, 0.5075, 0.5386, 0.5139], [0.6052, 0.4075, 0.9037, 0.5253, 0.4177, 0.5480, 0.5937, 0.5712], [0.6961, 0.4666, 0.8221, 0.2895, 0.4008, 0.2437, 0.6741, 0.5568], [0.6450, 0.4200, 0.8834, 0.5708, 0.3868, 0.5175, 0.6314, 0.5497], [0.6174, 0.4091, 0.7355, 0.2822, 0.3222, 0.3458, 0.5709, 0.5829]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6148, 0.4076, 0.8666, 0.4820, 0.4137, 0.5067, 0.5250, 0.5767], [0.0000, 0.0000, 0.7613, 0.3250, 0.4038, 0.2533, 0.5437, 0.5767], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.6102, 0.4005, 0.8687, 0.5100, 0.4812, 0.5400, 0.5404, 0.5064], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385], [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166], [0.6132, 0.4066, 0.7259, 0.2402, 0.3587, 0.3300, 0.6000, 0.5600]]], device='cuda:0') loss_train_step before backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train: 0.0640558628947474 step: 37 running loss: 0.0017312395376958758 Train Steps: 37/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550], [0.6058, 0.3986, 0.8324, 0.4626, 0.3838, 0.4983, 0.5147, 0.5466], [0.6339, 0.4149, 0.8800, 0.5000, 0.3900, 0.5283, 0.7541, 0.5424], [0.6250, 0.4103, 0.8950, 0.4400, 0.3912, 0.5650, 0.6050, 0.5133], [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5787, 0.5600], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285], [0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [ nan, nan, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6463, 0.4486, 0.7205, 0.2611, 0.3622, 0.2387, 0.5730, 0.5696], [0.5679, 0.3821, 0.8604, 0.4857, 0.3614, 0.4933, 0.5540, 0.5588], [0.5808, 0.3955, 0.9359, 0.5488, 0.3764, 0.5652, 0.7324, 0.5749], [0.6251, 0.4041, 0.9236, 0.4773, 0.3702, 0.5895, 0.6485, 0.5471], [0.6711, 0.4540, 0.9037, 0.5153, 0.4514, 0.5124, 0.6330, 0.5645], [0.5877, 0.3802, 0.8120, 0.4371, 0.3248, 0.4588, 0.5510, 0.5464], [0.5369, 0.3578, 0.8401, 0.3343, 0.3902, 0.2322, 0.5598, 0.5353], [0.2642, 0.1942, 0.7214, 0.2071, 0.4413, 0.2090, 0.5914, 0.5881]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5362, 0.5550], [0.6058, 0.3986, 0.8324, 0.4626, 0.3837, 0.4983, 0.5147, 0.5466], [0.6339, 0.4149, 0.8800, 0.5000, 0.3900, 0.5283, 0.7541, 0.5424], [0.6250, 0.4103, 0.8950, 0.4400, 0.3913, 0.5650, 0.6050, 0.5133], [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5788, 0.5600], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285], [0.6143, 0.4040, 0.8238, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.0000, 0.0000, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733]]], device='cuda:0') loss_train_step before backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train: 0.0668318530661054 step: 38 running loss: 0.0017587329754238262 Train Steps: 38/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540], [0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183], [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320], [0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954], [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400], [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166], [0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412], [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4808, 0.3253, 0.8920, 0.4184, 0.4204, 0.3435, 0.6587, 0.5698], [0.5491, 0.3737, 0.8658, 0.3256, 0.4150, 0.3050, 0.6280, 0.5679], [0.5523, 0.3764, 0.8521, 0.4818, 0.3708, 0.5154, 0.6665, 0.5470], [0.5079, 0.3312, 0.8279, 0.3715, 0.3946, 0.2913, 0.5220, 0.5341], [0.6193, 0.3941, 0.8742, 0.4529, 0.4648, 0.5880, 0.6097, 0.5398], [0.6312, 0.4041, 0.8724, 0.4444, 0.4001, 0.5414, 0.6760, 0.5446], [0.5933, 0.3861, 0.8602, 0.3223, 0.3764, 0.2990, 0.6152, 0.5545], [0.5005, 0.3396, 0.8607, 0.4584, 0.4423, 0.4931, 0.5061, 0.5280]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540], [0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183], [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320], [0.6137, 0.4038, 0.8562, 0.4050, 0.3812, 0.2550, 0.5106, 0.4954], [0.6185, 0.4098, 0.8838, 0.4900, 0.4538, 0.5800, 0.6288, 0.5400], [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166], [0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412], [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431]]], device='cuda:0') loss_train_step before backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train: 0.06857802328886464 step: 39 running loss: 0.0017584108535606319 Train Steps: 39/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6183, 0.4076, 0.8838, 0.4517, 0.3813, 0.4483, 0.5775, 0.5633], [0.6218, 0.4098, 0.7238, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350], [0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650], [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689], [0.6350, 0.4043, 0.8738, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950], [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5737, 0.5350]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5491, 0.3530, 0.8468, 0.5286, 0.3845, 0.5185, 0.7065, 0.5212], [0.6097, 0.3884, 0.9029, 0.4055, 0.4121, 0.4778, 0.6226, 0.5578], [0.6146, 0.4019, 0.7406, 0.1859, 0.4536, 0.2654, 0.6394, 0.5445], [0.6019, 0.3787, 0.8673, 0.2497, 0.4574, 0.2452, 0.6368, 0.5177], [0.5598, 0.3475, 0.8746, 0.4580, 0.4476, 0.5110, 0.5642, 0.5375], [0.5357, 0.3467, 0.7885, 0.3384, 0.3596, 0.4130, 0.5556, 0.5617], [0.5348, 0.3464, 0.8776, 0.5457, 0.4168, 0.5147, 0.6626, 0.5053], [0.5515, 0.3683, 0.8907, 0.4461, 0.4028, 0.3392, 0.6119, 0.5656]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6183, 0.4076, 0.8838, 0.4517, 0.3812, 0.4483, 0.5775, 0.5633], [0.6218, 0.4098, 0.7237, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350], [0.6250, 0.4131, 0.8687, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6098, 0.3991, 0.8637, 0.4717, 0.4263, 0.4967, 0.5213, 0.5650], [0.6102, 0.4001, 0.7738, 0.3583, 0.3462, 0.3800, 0.5524, 0.5689], [0.6350, 0.4043, 0.8737, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950], [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5738, 0.5350]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.06985682185040787 step: 40 running loss: 0.0017464205462601966 Train Steps: 40/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6272, 0.4045, 0.8538, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6104, 0.4029, 0.8738, 0.4900, 0.4088, 0.4533, 0.5070, 0.5510], [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170], [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6201, 0.4102, 0.7288, 0.2417, 0.4150, 0.2383, 0.6100, 0.5500], [0.6120, 0.4014, 0.6863, 0.2817, 0.3700, 0.2783, 0.5513, 0.5667], [ nan, nan, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6196, 0.3877, 0.8643, 0.5624, 0.3899, 0.4373, 0.6491, 0.4795], [0.5921, 0.3826, 0.9056, 0.4820, 0.4337, 0.4751, 0.5843, 0.5186], [0.6419, 0.3869, 0.8472, 0.2046, 0.5146, 0.2001, 0.6809, 0.5151], [0.6691, 0.4310, 0.8642, 0.5447, 0.4198, 0.4953, 0.6364, 0.5944], [0.7202, 0.4597, 0.9144, 0.4096, 0.4746, 0.5625, 0.6505, 0.5424], [0.6505, 0.4100, 0.7622, 0.2280, 0.4200, 0.2682, 0.6104, 0.5503], [0.5771, 0.3638, 0.7290, 0.2275, 0.3697, 0.3242, 0.5733, 0.5296], [0.0531, 0.0196, 0.7404, 0.2252, 0.4170, 0.2659, 0.5337, 0.5396]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6271, 0.4045, 0.8537, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6104, 0.4029, 0.8737, 0.4900, 0.4087, 0.4533, 0.5070, 0.5510], [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170], [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6201, 0.4102, 0.7287, 0.2417, 0.4150, 0.2383, 0.6100, 0.5500], [0.6120, 0.4013, 0.6862, 0.2817, 0.3700, 0.2783, 0.5512, 0.5667], [0.0000, 0.0000, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.07106089597800747 step: 41 running loss: 0.0017331925848294504 Train Steps: 41/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400], [0.6256, 0.4199, 0.8638, 0.5800, 0.3987, 0.4383, 0.5600, 0.5950], [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377], [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6090, 0.4010, 0.7838, 0.3483, 0.3538, 0.3783, 0.5462, 0.5077], [0.6139, 0.4019, 0.7137, 0.2150, 0.4375, 0.1533, 0.5293, 0.5006], [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6513, 0.4053, 0.8174, 0.2385, 0.4802, 0.2250, 0.6510, 0.4989], [0.6342, 0.3939, 0.8911, 0.4020, 0.3887, 0.5048, 0.6225, 0.5364], [0.6598, 0.4216, 0.8528, 0.5737, 0.4386, 0.4707, 0.5988, 0.5928], [0.6260, 0.4001, 0.8529, 0.5291, 0.4325, 0.5023, 0.6193, 0.5269], [0.5934, 0.3801, 0.8919, 0.4087, 0.4370, 0.2679, 0.5986, 0.5490], [0.5857, 0.3637, 0.7795, 0.3259, 0.3542, 0.3958, 0.5541, 0.5063], [0.3791, 0.2412, 0.6990, 0.1641, 0.4398, 0.1908, 0.5585, 0.4823], [0.5369, 0.3439, 0.6833, 0.2231, 0.4152, 0.2109, 0.5087, 0.5430]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6277, 0.4056, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6137, 0.5400], [0.6256, 0.4199, 0.8637, 0.5800, 0.3988, 0.4383, 0.5600, 0.5950], [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377], [0.6226, 0.4098, 0.8913, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6090, 0.4010, 0.7837, 0.3483, 0.3537, 0.3783, 0.5462, 0.5077], [0.6139, 0.4019, 0.7138, 0.2150, 0.4375, 0.1533, 0.5293, 0.5006], [0.6159, 0.4085, 0.6900, 0.2283, 0.4087, 0.1950, 0.5123, 0.5397]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.07297768600983545 step: 42 running loss: 0.0017375639526151297 Train Steps: 42/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850], [0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6218, 0.4137, 0.7263, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064], [0.6274, 0.4003, 0.8638, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5348, 0.3344, 0.8918, 0.5236, 0.3754, 0.4113, 0.5386, 0.5193], [0.6320, 0.3968, 0.8734, 0.4376, 0.4709, 0.5201, 0.6116, 0.5432], [0.6034, 0.3844, 0.7373, 0.2353, 0.4592, 0.1258, 0.5958, 0.5342], [0.5810, 0.3819, 0.7541, 0.2476, 0.4223, 0.1637, 0.5635, 0.5006], [0.6223, 0.3896, 0.7139, 0.2423, 0.4031, 0.2472, 0.5769, 0.5434], [0.5928, 0.3790, 0.8403, 0.4260, 0.3608, 0.4231, 0.5573, 0.5454], [0.6233, 0.3942, 0.8505, 0.5002, 0.4849, 0.4924, 0.5449, 0.4821], [0.6497, 0.4057, 0.8433, 0.5769, 0.3865, 0.4554, 0.6365, 0.4721]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6184, 0.4079, 0.8838, 0.4617, 0.4837, 0.5650, 0.6175, 0.5850], [0.6245, 0.4100, 0.7763, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6218, 0.4137, 0.7262, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6102, 0.4005, 0.8687, 0.5100, 0.4812, 0.5400, 0.5404, 0.5064], [0.6274, 0.4003, 0.8637, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.07371883763698861 step: 43 running loss: 0.0017143915729532235 Train Steps: 43/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367], [0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320], [0.6151, 0.4058, 0.7068, 0.2680, 0.3400, 0.4083, 0.5775, 0.5733], [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869], [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6772, 0.4490, 0.8220, 0.3402, 0.4000, 0.2432, 0.6008, 0.5333], [0.6473, 0.4303, 0.8864, 0.4997, 0.3895, 0.5120, 0.6969, 0.5476], [0.5202, 0.3567, 0.6952, 0.2878, 0.3395, 0.3716, 0.5097, 0.5642], [0.5861, 0.4021, 0.8474, 0.6113, 0.3800, 0.3585, 0.5587, 0.4985], [0.6097, 0.4207, 0.7052, 0.2364, 0.4152, 0.2074, 0.5507, 0.5604], [0.5311, 0.3470, 0.7862, 0.2609, 0.4629, 0.1725, 0.5829, 0.5163], [0.6151, 0.4042, 0.7916, 0.2439, 0.4920, 0.1672, 0.5747, 0.4947], [0.6339, 0.4210, 0.8235, 0.2968, 0.4063, 0.2324, 0.5447, 0.5461]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367], [0.6250, 0.3993, 0.9137, 0.4333, 0.3762, 0.5217, 0.6995, 0.5320], [0.6151, 0.4058, 0.7068, 0.2680, 0.3400, 0.4083, 0.5775, 0.5733], [0.6272, 0.4071, 0.8737, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6276, 0.4095, 0.8238, 0.2250, 0.4663, 0.1783, 0.6171, 0.4869], [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.07493773492751643 step: 44 running loss: 0.0017031303392617372 Train Steps: 44/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667], [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633], [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563], [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672], [0.6110, 0.4047, 0.8700, 0.4483, 0.3713, 0.3967, 0.5088, 0.5517], [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.1702, 0.1298, 0.7268, 0.2563, 0.3765, 0.2613, 0.5716, 0.5468], [0.7073, 0.4773, 0.8548, 0.5149, 0.3481, 0.3462, 0.5311, 0.5441], [0.7714, 0.5124, 0.7089, 0.2594, 0.4090, 0.2234, 0.5770, 0.6174], [0.7612, 0.5006, 0.8099, 0.2706, 0.4542, 0.1748, 0.6107, 0.5040], [0.2474, 0.1812, 0.7453, 0.3121, 0.3465, 0.2823, 0.4903, 0.5219], [0.8275, 0.5585, 0.8661, 0.5414, 0.3792, 0.4369, 0.6221, 0.5096], [0.6594, 0.4703, 0.8696, 0.4741, 0.3748, 0.3802, 0.4978, 0.5200], [0.7095, 0.4710, 0.8922, 0.4212, 0.3628, 0.2994, 0.5741, 0.5112]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.7525, 0.2291, 0.3837, 0.3017, 0.6050, 0.5667], [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5638, 0.5633], [0.6193, 0.4078, 0.7287, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6277, 0.4056, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.0000, 0.0000, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563], [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672], [0.6110, 0.4047, 0.8700, 0.4483, 0.3713, 0.3967, 0.5088, 0.5517], [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250]]], device='cuda:0') loss_train_step before backward: tensor(0.0050, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0050, device='cuda:0', grad_fn=) loss_train: 0.07991487270919606 step: 45 running loss: 0.0017758860602043569 Train Steps: 45/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6259, 0.4133, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817], [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773], [0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6370, 0.4299, 0.8304, 0.5609, 0.3799, 0.4245, 0.5558, 0.6125], [0.6131, 0.4136, 0.7523, 0.2322, 0.3399, 0.2686, 0.5923, 0.5328], [0.6087, 0.4224, 0.8043, 0.2357, 0.4771, 0.1322, 0.6005, 0.4889], [0.5696, 0.3804, 0.7569, 0.2403, 0.3683, 0.2328, 0.5821, 0.5382], [0.5883, 0.3892, 0.8575, 0.5295, 0.3315, 0.3194, 0.5310, 0.5377], [0.6072, 0.3977, 0.8384, 0.5736, 0.4364, 0.4315, 0.4989, 0.5056], [0.6448, 0.4398, 0.8627, 0.4838, 0.4578, 0.4607, 0.4873, 0.5472], [0.6714, 0.4410, 0.9013, 0.4527, 0.3524, 0.4750, 0.7102, 0.5378]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6259, 0.4132, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817], [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773], [0.6250, 0.3993, 0.9137, 0.4333, 0.3762, 0.5217, 0.6995, 0.5320]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.08066413854248822 step: 46 running loss: 0.0017535682291845264 Train Steps: 46/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.6276, 0.4120, 0.8738, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892], [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170], [0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5563, 0.5533], [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869], [0.6058, 0.3986, 0.8324, 0.4626, 0.3838, 0.4983, 0.5147, 0.5466], [0.6353, 0.4128, 0.9138, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5871, 0.3827, 0.7595, 0.2599, 0.4304, 0.1861, 0.5773, 0.5302], [0.5781, 0.3907, 0.8509, 0.3644, 0.4057, 0.2110, 0.5864, 0.4940], [0.5893, 0.3861, 0.8078, 0.2766, 0.4622, 0.1781, 0.6135, 0.5363], [0.5620, 0.3663, 0.8583, 0.5443, 0.3683, 0.4701, 0.5473, 0.5656], [0.6003, 0.3920, 0.8042, 0.2455, 0.4554, 0.1913, 0.5968, 0.4997], [0.6014, 0.4023, 0.8215, 0.4718, 0.3669, 0.4867, 0.5102, 0.5505], [0.5840, 0.3870, 0.8857, 0.3702, 0.4207, 0.2885, 0.6763, 0.5843], [0.5322, 0.3514, 0.7657, 0.3172, 0.3484, 0.2597, 0.5061, 0.5328]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.6276, 0.4120, 0.8737, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892], [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170], [0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5562, 0.5533], [0.6276, 0.4095, 0.8238, 0.2250, 0.4663, 0.1783, 0.6171, 0.4869], [0.6058, 0.3986, 0.8324, 0.4626, 0.3837, 0.4983, 0.5147, 0.5466], [0.6353, 0.4128, 0.9137, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.08150381379527971 step: 47 running loss: 0.0017341236977719088 Train Steps: 47/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633], [0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283], [0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183], [ nan, nan, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933], [0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123], [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6507, 0.4312, 0.8895, 0.5101, 0.3482, 0.3958, 0.5736, 0.5568], [ 0.6158, 0.4013, 0.8695, 0.5192, 0.4944, 0.5012, 0.5342, 0.5290], [ 0.7051, 0.4321, 0.7647, 0.2217, 0.4441, 0.1759, 0.5999, 0.5217], [-0.0247, 0.0039, 0.7466, 0.2767, 0.3981, 0.2568, 0.5355, 0.5709], [ 0.7045, 0.4758, 0.8790, 0.3624, 0.3280, 0.4345, 0.5924, 0.5202], [ 0.6245, 0.4351, 0.8443, 0.3931, 0.4787, 0.2461, 0.5647, 0.6040], [ 0.7088, 0.4377, 0.9097, 0.4705, 0.3590, 0.4088, 0.6427, 0.5270], [ 0.6248, 0.4281, 0.7528, 0.2074, 0.4034, 0.2537, 0.6343, 0.5507]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5638, 0.5633], [0.6084, 0.4008, 0.8587, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283], [0.6213, 0.4001, 0.7713, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183], [0.0000, 0.0000, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933], [0.6213, 0.4131, 0.8438, 0.3550, 0.3512, 0.4400, 0.5716, 0.5123], [0.6299, 0.4303, 0.7962, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.08213754411553964 step: 48 running loss: 0.001711198835740409 Train Steps: 48/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6193, 0.4108, 0.7438, 0.2700, 0.3650, 0.3683, 0.6238, 0.5717], [0.6128, 0.4118, 0.8638, 0.5333, 0.4625, 0.5267, 0.5193, 0.5475], [0.6234, 0.4179, 0.7825, 0.3450, 0.3813, 0.2867, 0.5675, 0.5617], [0.6099, 0.4030, 0.8638, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388], [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833], [0.6131, 0.4064, 0.8638, 0.5200, 0.4788, 0.4783, 0.5258, 0.5867], [0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5613, 0.5900], [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5959, 0.4000, 0.7621, 0.2510, 0.3706, 0.3683, 0.6783, 0.5539], [0.6303, 0.4155, 0.8891, 0.5091, 0.4621, 0.4924, 0.5939, 0.5456], [0.5692, 0.3708, 0.8184, 0.3110, 0.3984, 0.2670, 0.6072, 0.5645], [0.5968, 0.3888, 0.8988, 0.4949, 0.5172, 0.4731, 0.5595, 0.5253], [0.6241, 0.3917, 0.8970, 0.4466, 0.3830, 0.4809, 0.6115, 0.4860], [0.6153, 0.4076, 0.8837, 0.4904, 0.4676, 0.4800, 0.5875, 0.5728], [0.5677, 0.3693, 0.8548, 0.4486, 0.3512, 0.3619, 0.5806, 0.5470], [0.5376, 0.3586, 0.8915, 0.3900, 0.3824, 0.2972, 0.5508, 0.5358]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6193, 0.4108, 0.7437, 0.2700, 0.3650, 0.3683, 0.6237, 0.5717], [0.6128, 0.4118, 0.8637, 0.5333, 0.4625, 0.5267, 0.5193, 0.5475], [0.6234, 0.4179, 0.7825, 0.3450, 0.3812, 0.2867, 0.5675, 0.5617], [0.6098, 0.4030, 0.8637, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388], [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6037, 0.4833], [0.6132, 0.4063, 0.8637, 0.5200, 0.4787, 0.4783, 0.5258, 0.5867], [0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5612, 0.5900], [0.6133, 0.4094, 0.8495, 0.4028, 0.3587, 0.3200, 0.5003, 0.5407]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.08310104586416855 step: 49 running loss: 0.001695939711513644 Train Steps: 49/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817], [0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233], [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783], [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550], [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5848, 0.3844, 0.8986, 0.4746, 0.4875, 0.5192, 0.5734, 0.5595], [0.5635, 0.3480, 0.8617, 0.5607, 0.4685, 0.5106, 0.5479, 0.5005], [0.6011, 0.3629, 0.8678, 0.5411, 0.4214, 0.4758, 0.5868, 0.5364], [0.5875, 0.3733, 0.8663, 0.3486, 0.3731, 0.3224, 0.5392, 0.5639], [0.6006, 0.3898, 0.7459, 0.2103, 0.4047, 0.2789, 0.5768, 0.5538], [0.5433, 0.3653, 0.7747, 0.2095, 0.4477, 0.2071, 0.6094, 0.5367], [0.5403, 0.3561, 0.8304, 0.2777, 0.3716, 0.4511, 0.6204, 0.5646], [0.6354, 0.4143, 0.8928, 0.3292, 0.4869, 0.2185, 0.6488, 0.5354]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817], [0.6222, 0.3937, 0.8350, 0.5617, 0.4137, 0.4600, 0.5800, 0.5233], [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783], [0.6108, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550], [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.08384644222678617 step: 50 running loss: 0.0016769288445357233 Train Steps: 50/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926], [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204], [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350], [0.6160, 0.4093, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283], [0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6177, 0.4086, 0.8738, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6018, 0.3981, 0.8972, 0.4834, 0.4861, 0.6041, 0.6385, 0.5484], [0.5998, 0.4012, 0.7800, 0.2538, 0.4224, 0.2867, 0.6259, 0.5099], [0.6428, 0.4122, 0.8682, 0.5459, 0.4503, 0.5400, 0.6588, 0.5392], [0.6533, 0.3981, 0.8786, 0.5066, 0.4169, 0.4677, 0.5248, 0.5279], [0.5899, 0.3816, 0.8542, 0.4554, 0.3995, 0.4642, 0.5371, 0.5860], [0.7126, 0.4692, 0.8975, 0.4193, 0.3818, 0.3942, 0.5930, 0.5641], [0.6259, 0.4239, 0.7293, 0.2223, 0.4088, 0.2643, 0.5765, 0.5782], [0.5646, 0.3629, 0.8836, 0.3953, 0.4145, 0.5627, 0.6421, 0.5542]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926], [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204], [0.6143, 0.3982, 0.8650, 0.4883, 0.3913, 0.4317, 0.5315, 0.5350], [0.6160, 0.4092, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283], [0.6187, 0.4103, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6177, 0.4085, 0.8737, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.08451496018096805 step: 51 running loss: 0.0016571560819797655 Train Steps: 51/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6339, 0.4102, 0.8588, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5963, 0.6217], [0.6199, 0.4065, 0.7598, 0.2385, 0.4317, 0.1981, 0.5933, 0.5221], [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378], [ nan, nan, 0.8463, 0.2550, 0.5850, 0.2133, 0.7129, 0.6072], [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6809, 0.4253, 0.8568, 0.2991, 0.4494, 0.2359, 0.6058, 0.5163], [0.5803, 0.3767, 0.8677, 0.4200, 0.4125, 0.6332, 0.5489, 0.5271], [0.6340, 0.3932, 0.8501, 0.5158, 0.3730, 0.5077, 0.5502, 0.5099], [0.5400, 0.3482, 0.6996, 0.2484, 0.4044, 0.2768, 0.5484, 0.6001], [0.6741, 0.4330, 0.7321, 0.2190, 0.4196, 0.2534, 0.5463, 0.5307], [0.6390, 0.4043, 0.8925, 0.4650, 0.3780, 0.3900, 0.6473, 0.5288], [0.4883, 0.3077, 0.8510, 0.2756, 0.5270, 0.3071, 0.6809, 0.5730], [0.5779, 0.3868, 0.8395, 0.5546, 0.4571, 0.4585, 0.5455, 0.6070]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6339, 0.4102, 0.8587, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5962, 0.6217], [0.6199, 0.4065, 0.7598, 0.2385, 0.4317, 0.1981, 0.5933, 0.5221], [0.6307, 0.4029, 0.8988, 0.4817, 0.3938, 0.3500, 0.7311, 0.5378], [0.0000, 0.0000, 0.8462, 0.2550, 0.5850, 0.2133, 0.7129, 0.6072], [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933]]], device='cuda:0') loss_train_step before backward: tensor(0.0064, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0064, device='cuda:0', grad_fn=) loss_train: 0.09093694994226098 step: 52 running loss: 0.0017487874988896342 Train Steps: 52/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633], [0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683], [0.6198, 0.4130, 0.8762, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6138, 0.5450], [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750], [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5538, 0.3485, 0.8356, 0.5000, 0.4175, 0.6157, 0.6252, 0.5638], [0.5602, 0.3685, 0.7444, 0.2204, 0.4513, 0.2355, 0.6059, 0.5304], [0.5542, 0.3638, 0.8654, 0.5291, 0.3889, 0.3825, 0.5489, 0.5775], [0.6536, 0.4437, 0.8621, 0.4250, 0.3704, 0.5169, 0.5454, 0.5253], [0.5912, 0.4080, 0.8529, 0.2716, 0.5181, 0.1879, 0.6456, 0.5222], [0.6205, 0.4104, 0.8665, 0.4365, 0.4210, 0.5852, 0.6185, 0.5612], [0.4969, 0.3367, 0.7763, 0.3154, 0.4192, 0.2654, 0.5452, 0.5896], [0.6335, 0.4119, 0.8561, 0.4200, 0.3686, 0.3731, 0.5136, 0.5897]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6293, 0.4024, 0.8750, 0.5000, 0.4013, 0.5733, 0.7121, 0.5633], [0.6202, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5412, 0.5683], [0.6198, 0.4130, 0.8763, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6137, 0.5450], [0.6182, 0.4099, 0.7812, 0.3000, 0.3938, 0.2367, 0.5325, 0.5750], [0.6182, 0.4058, 0.8737, 0.4350, 0.3562, 0.3400, 0.5290, 0.5822]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.0920837412122637 step: 53 running loss: 0.0017374290794766737 Train Steps: 53/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6263, 0.4233, 0.7924, 0.4626, 0.3788, 0.2883, 0.5573, 0.6047], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200], [0.6179, 0.4040, 0.7412, 0.1850, 0.3825, 0.2783, 0.5837, 0.5600], [0.6111, 0.4033, 0.8300, 0.3267, 0.3588, 0.3333, 0.5444, 0.5637], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6361, 0.4294, 0.7938, 0.4831, 0.4019, 0.3325, 0.5618, 0.6023], [0.6089, 0.4054, 0.8951, 0.3480, 0.4153, 0.2785, 0.6387, 0.5254], [0.6143, 0.3938, 0.8661, 0.5296, 0.3820, 0.4893, 0.6041, 0.4902], [0.6072, 0.4022, 0.8795, 0.4778, 0.3956, 0.4922, 0.5996, 0.5346], [0.5502, 0.3681, 0.7329, 0.2432, 0.4047, 0.3000, 0.6031, 0.5652], [0.5602, 0.3857, 0.8316, 0.3321, 0.3668, 0.3287, 0.5372, 0.5738], [0.5965, 0.4166, 0.8792, 0.4442, 0.4421, 0.5145, 0.5707, 0.5443], [0.5677, 0.3775, 0.7882, 0.3297, 0.3547, 0.3545, 0.6025, 0.5210]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6263, 0.4232, 0.7924, 0.4626, 0.3787, 0.2883, 0.5573, 0.6047], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6187, 0.5283], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200], [0.6179, 0.4040, 0.7412, 0.1850, 0.3825, 0.2783, 0.5838, 0.5600], [0.6111, 0.4033, 0.8300, 0.3267, 0.3587, 0.3333, 0.5444, 0.5637], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.0925499573640991 step: 54 running loss: 0.0017138880993351686 Train Steps: 54/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575], [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6118, 0.4052, 0.8463, 0.3917, 0.3538, 0.3450, 0.5053, 0.5593], [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123], [0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6453, 0.4232, 0.8670, 0.4669, 0.3818, 0.4092, 0.4712, 0.5853], [0.5555, 0.3425, 0.8555, 0.3448, 0.4047, 0.3914, 0.6735, 0.5412], [0.5939, 0.3750, 0.8760, 0.3566, 0.3731, 0.2931, 0.6047, 0.5714], [0.5895, 0.4000, 0.8453, 0.4075, 0.3547, 0.3354, 0.4997, 0.5532], [0.5426, 0.3417, 0.9176, 0.4015, 0.4547, 0.2686, 0.7097, 0.5357], [0.6337, 0.4331, 0.7227, 0.2762, 0.3751, 0.2437, 0.5495, 0.5227], [0.5736, 0.3552, 0.8865, 0.4961, 0.3587, 0.5078, 0.6134, 0.4761], [0.5949, 0.4000, 0.7421, 0.2326, 0.4343, 0.1824, 0.5836, 0.5334]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575], [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6118, 0.4052, 0.8462, 0.3917, 0.3537, 0.3450, 0.5053, 0.5593], [0.6271, 0.4040, 0.9137, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123], [0.6264, 0.4049, 0.8988, 0.4633, 0.3812, 0.4983, 0.6326, 0.4843], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.0932905454246793 step: 55 running loss: 0.001696191734994169 Train Steps: 55/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656], [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5960, 0.3916, 0.8328, 0.3137, 0.3090, 0.3448, 0.6383, 0.5323], [0.5227, 0.3610, 0.7821, 0.2958, 0.3755, 0.2210, 0.5449, 0.5759], [0.5617, 0.3655, 0.8606, 0.2810, 0.4852, 0.2073, 0.7063, 0.5246], [0.6541, 0.4277, 0.8747, 0.4585, 0.3404, 0.4017, 0.5486, 0.5071], [0.6761, 0.4527, 0.8541, 0.5255, 0.3467, 0.4580, 0.6511, 0.4972], [0.5504, 0.3616, 0.7424, 0.3120, 0.4642, 0.1758, 0.5704, 0.5991], [0.6524, 0.4272, 0.8397, 0.4339, 0.3420, 0.4341, 0.5560, 0.5430], [0.6357, 0.4008, 0.8463, 0.5149, 0.4319, 0.4953, 0.5789, 0.5273]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6182, 0.4099, 0.7812, 0.3000, 0.3938, 0.2367, 0.5325, 0.5750], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.6108, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6091, 0.3997, 0.8314, 0.4334, 0.3787, 0.4550, 0.5213, 0.5656], [0.6202, 0.4053, 0.8637, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.09418759748223238 step: 56 running loss: 0.0016819213836112925 Train Steps: 56/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309], [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317], [0.6201, 0.4064, 0.8688, 0.5050, 0.4225, 0.5100, 0.6138, 0.5500], [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167], [0.6250, 0.4106, 0.8700, 0.3717, 0.3588, 0.4967, 0.6038, 0.5167], [0.6224, 0.4179, 0.8700, 0.5683, 0.4037, 0.4683, 0.5650, 0.5600], [0.6127, 0.4115, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6676, 0.4368, 0.8900, 0.4799, 0.3935, 0.3156, 0.6512, 0.5184], [0.7548, 0.4830, 0.8803, 0.5309, 0.3699, 0.3366, 0.6227, 0.5090], [0.6117, 0.3932, 0.8814, 0.4646, 0.4048, 0.4646, 0.6164, 0.5311], [0.5857, 0.3969, 0.8048, 0.2933, 0.3470, 0.3339, 0.6369, 0.5901], [0.6177, 0.4172, 0.8760, 0.3511, 0.3644, 0.4374, 0.6449, 0.5348], [0.6561, 0.4186, 0.8780, 0.5503, 0.3889, 0.4274, 0.5882, 0.5325], [0.6282, 0.4230, 0.7384, 0.2648, 0.3612, 0.2285, 0.5473, 0.5478], [0.6170, 0.4053, 0.7809, 0.3479, 0.3499, 0.3602, 0.5553, 0.5295]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309], [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317], [0.6201, 0.4064, 0.8687, 0.5050, 0.4225, 0.5100, 0.6137, 0.5500], [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6037, 0.6167], [0.6250, 0.4105, 0.8700, 0.3717, 0.3587, 0.4967, 0.6037, 0.5167], [0.6224, 0.4179, 0.8700, 0.5683, 0.4038, 0.4683, 0.5650, 0.5600], [0.6127, 0.4114, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.09527627841453068 step: 57 running loss: 0.001671513656395275 Train Steps: 57/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810], [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879], [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378], [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263], [0.6256, 0.4199, 0.8638, 0.5800, 0.3987, 0.4383, 0.5600, 0.5950], [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5923, 0.3882, 0.7885, 0.2782, 0.3581, 0.2975, 0.6219, 0.5589], [0.6705, 0.4474, 0.8350, 0.2856, 0.4068, 0.2243, 0.6166, 0.4929], [0.6653, 0.4542, 0.8764, 0.4276, 0.3875, 0.5155, 0.6295, 0.5115], [0.6140, 0.4167, 0.8751, 0.5256, 0.3806, 0.4165, 0.5473, 0.5524], [0.6482, 0.4520, 0.8684, 0.3917, 0.3614, 0.3510, 0.5551, 0.5411], [0.7469, 0.4962, 0.7565, 0.2374, 0.4311, 0.1452, 0.5950, 0.5401], [0.6686, 0.4340, 0.8478, 0.5763, 0.3875, 0.4165, 0.5866, 0.6031], [0.5929, 0.3802, 0.9192, 0.4869, 0.3647, 0.4593, 0.6552, 0.5192]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810], [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879], [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378], [0.6082, 0.4024, 0.8737, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263], [0.6256, 0.4199, 0.8637, 0.5800, 0.3988, 0.4383, 0.5600, 0.5950], [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.09628399272332899 step: 58 running loss: 0.0016600688400573965 Train Steps: 58/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148], [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550], [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461], [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6063, 0.6083], [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133], [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533], [0.6148, 0.3996, 0.8488, 0.3867, 0.3488, 0.4067, 0.5863, 0.5000]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6687, 0.4266, 0.8866, 0.5246, 0.3931, 0.4508, 0.5781, 0.4987], [0.5809, 0.3803, 0.8802, 0.4642, 0.4092, 0.5018, 0.6329, 0.5567], [0.6141, 0.4125, 0.8633, 0.4633, 0.4655, 0.4415, 0.5980, 0.5629], [0.6745, 0.4543, 0.8192, 0.3643, 0.3462, 0.3749, 0.5754, 0.5205], [0.6390, 0.4397, 0.8338, 0.3683, 0.3432, 0.4207, 0.5815, 0.5846], [0.6567, 0.4287, 0.8254, 0.2548, 0.4760, 0.1247, 0.6478, 0.5078], [0.6230, 0.4060, 0.8469, 0.4417, 0.3595, 0.3145, 0.5798, 0.5405], [0.7041, 0.4520, 0.8424, 0.3891, 0.3473, 0.3727, 0.5912, 0.5173]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148], [0.6241, 0.4142, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550], [0.6058, 0.3978, 0.8288, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461], [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6062, 0.6083], [0.6266, 0.4065, 0.8313, 0.2467, 0.4787, 0.1733, 0.6313, 0.5133], [0.6252, 0.4162, 0.8662, 0.4467, 0.3625, 0.3567, 0.6037, 0.5533], [0.6148, 0.3996, 0.8487, 0.3867, 0.3487, 0.4067, 0.5863, 0.5000]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.09704984255949967 step: 59 running loss: 0.0016449125857542317 Train Steps: 59/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6099, 0.4030, 0.8638, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388], [0.6147, 0.4107, 0.8137, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412], [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467], [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633], [0.6129, 0.4063, 0.8738, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117], [0.6198, 0.4115, 0.7762, 0.2717, 0.3713, 0.3200, 0.5837, 0.5683], [0.6131, 0.4037, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5524, 0.3684, 0.8810, 0.5297, 0.5100, 0.4835, 0.5320, 0.5300], [0.5466, 0.3586, 0.8001, 0.3506, 0.3732, 0.2584, 0.5257, 0.5388], [0.6086, 0.3828, 0.8570, 0.3213, 0.3544, 0.3979, 0.5820, 0.5188], [0.6513, 0.4311, 0.8822, 0.5697, 0.4222, 0.5225, 0.6060, 0.5511], [0.6884, 0.4478, 0.8747, 0.5551, 0.4310, 0.4643, 0.5607, 0.5479], [0.8111, 0.5159, 0.7555, 0.2227, 0.4257, 0.2197, 0.5950, 0.4980], [0.6200, 0.4054, 0.7956, 0.2775, 0.3638, 0.3102, 0.5975, 0.5639], [0.5884, 0.3865, 0.7129, 0.2693, 0.3853, 0.2834, 0.5373, 0.5499]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6098, 0.4030, 0.8637, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388], [0.6147, 0.4107, 0.8138, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412], [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467], [0.6192, 0.4128, 0.8512, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633], [0.6130, 0.4063, 0.8737, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5788, 0.5117], [0.6198, 0.4115, 0.7763, 0.2717, 0.3713, 0.3200, 0.5838, 0.5683], [0.6131, 0.4036, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.0983880981511902 step: 60 running loss: 0.00163980163585317 Train Steps: 60/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413], [ nan, nan, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999], [0.6125, 0.4076, 0.8488, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505], [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567], [0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467], [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400], [0.6128, 0.4115, 0.7934, 0.3778, 0.3450, 0.4033, 0.5337, 0.5456], [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5544, 0.3590, 0.8481, 0.2554, 0.5409, 0.2350, 0.6758, 0.5367], [0.2139, 0.1506, 0.7774, 0.2641, 0.4452, 0.2224, 0.5098, 0.5522], [0.7379, 0.4704, 0.8540, 0.4256, 0.3654, 0.3925, 0.4794, 0.5282], [0.5526, 0.3593, 0.7588, 0.2905, 0.4331, 0.2762, 0.5581, 0.5591], [0.6221, 0.4242, 0.8964, 0.5445, 0.4330, 0.5784, 0.5897, 0.5532], [0.5752, 0.3930, 0.8614, 0.3583, 0.3594, 0.4973, 0.6189, 0.5515], [0.7502, 0.4973, 0.8014, 0.4046, 0.3410, 0.4333, 0.5064, 0.5422], [0.7757, 0.5141, 0.7650, 0.2587, 0.4693, 0.2058, 0.5800, 0.5271]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413], [0.0000, 0.0000, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999], [0.6125, 0.4076, 0.8487, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505], [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5688, 0.5567], [0.6199, 0.4059, 0.8700, 0.4900, 0.4162, 0.5000, 0.6162, 0.5467], [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400], [0.6128, 0.4115, 0.7934, 0.3778, 0.3450, 0.4033, 0.5337, 0.5456], [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284]]], device='cuda:0') loss_train_step before backward: tensor(0.0032, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0032, device='cuda:0', grad_fn=) loss_train: 0.10155432714964263 step: 61 running loss: 0.001664825035240043 Train Steps: 61/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954], [ nan, nan, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283], [ nan, nan, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650], [0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804], [ nan, nan, 0.8938, 0.2850, 0.4662, 0.3117, 0.7406, 0.5528]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7333, 0.5004, 0.7735, 0.3322, 0.3780, 0.3378, 0.4660, 0.5727], [0.7130, 0.4630, 0.7946, 0.3756, 0.3519, 0.4460, 0.5214, 0.5315], [0.3009, 0.2057, 0.8649, 0.3455, 0.5145, 0.2872, 0.6320, 0.5746], [0.7436, 0.4856, 0.8686, 0.3659, 0.4083, 0.3257, 0.5701, 0.5233], [0.2093, 0.1384, 0.6915, 0.2328, 0.4616, 0.2953, 0.5221, 0.5782], [0.6982, 0.4350, 0.7919, 0.3409, 0.4342, 0.2717, 0.4762, 0.5167], [0.8061, 0.5239, 0.8521, 0.3967, 0.3773, 0.4158, 0.5907, 0.5020], [0.1672, 0.1115, 0.8612, 0.2978, 0.4808, 0.3562, 0.6760, 0.5824]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6076, 0.3952, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954], [0.0000, 0.0000, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6187, 0.5283], [0.0000, 0.0000, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650], [0.6143, 0.4040, 0.8238, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804], [0.0000, 0.0000, 0.8938, 0.2850, 0.4663, 0.3117, 0.7406, 0.5528]]], device='cuda:0') loss_train_step before backward: tensor(0.0064, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0064, device='cuda:0', grad_fn=) loss_train: 0.10794588641147129 step: 62 running loss: 0.0017410626840559885 Train Steps: 62/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100], [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917], [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650], [ nan, nan, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533], [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6058, 0.3986, 0.8324, 0.4626, 0.3838, 0.4983, 0.5147, 0.5466], [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6443, 0.3973, 0.9186, 0.4363, 0.3844, 0.4203, 0.6107, 0.5025], [0.5870, 0.3576, 0.9018, 0.4988, 0.3745, 0.4480, 0.5780, 0.5176], [0.6250, 0.4147, 0.7870, 0.4136, 0.3721, 0.3787, 0.5733, 0.5922], [0.5771, 0.3501, 0.8886, 0.4988, 0.4560, 0.5279, 0.5326, 0.5574], [0.1619, 0.0922, 0.7203, 0.2126, 0.4333, 0.2041, 0.5526, 0.5699], [0.4896, 0.3245, 0.9115, 0.4716, 0.4796, 0.5833, 0.5782, 0.5692], [0.5880, 0.3797, 0.8613, 0.4694, 0.4195, 0.5210, 0.5192, 0.5485], [0.6626, 0.4191, 0.7250, 0.2456, 0.4148, 0.2477, 0.5574, 0.5377]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100], [0.6236, 0.3965, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6179, 0.4118, 0.7278, 0.4237, 0.3587, 0.3400, 0.5675, 0.5917], [0.6098, 0.3991, 0.8637, 0.4717, 0.4263, 0.4967, 0.5213, 0.5650], [0.0000, 0.0000, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533], [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6058, 0.3986, 0.8324, 0.4626, 0.3837, 0.4983, 0.5147, 0.5466], [0.6072, 0.4029, 0.7038, 0.2150, 0.3913, 0.2267, 0.5516, 0.5507]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.1094864918559324 step: 63 running loss: 0.001737880823110038 Train Steps: 63/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165], [0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200], [0.6200, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367], [0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167], [0.6246, 0.4028, 0.8738, 0.4867, 0.4088, 0.5667, 0.6362, 0.5200], [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5513, 0.3592, 0.8662, 0.3485, 0.3772, 0.3205, 0.5762, 0.5387], [0.4817, 0.3048, 0.8697, 0.5463, 0.3880, 0.4739, 0.5632, 0.5283], [0.5399, 0.3348, 0.9029, 0.4546, 0.4078, 0.5403, 0.5872, 0.5457], [0.4647, 0.2731, 0.7782, 0.2394, 0.4507, 0.1822, 0.5812, 0.5226], [0.5025, 0.3091, 0.8910, 0.4677, 0.4300, 0.5890, 0.6320, 0.5428], [0.5606, 0.3719, 0.8676, 0.5222, 0.4119, 0.4951, 0.6652, 0.5893], [0.4988, 0.3030, 0.7705, 0.3064, 0.3635, 0.3410, 0.5255, 0.5430], [0.4922, 0.3373, 0.8392, 0.3999, 0.3699, 0.3585, 0.5458, 0.5868]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165], [0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200], [0.6199, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367], [0.6204, 0.4007, 0.7837, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167], [0.6246, 0.4028, 0.8737, 0.4867, 0.4087, 0.5667, 0.6363, 0.5200], [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650]]], device='cuda:0') loss_train_step before backward: tensor(0.0029, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0029, device='cuda:0', grad_fn=) loss_train: 0.11233921625535004 step: 64 running loss: 0.0017553002539898444 Train Steps: 64/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617], [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033], [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6151, 0.4058, 0.7068, 0.2680, 0.3400, 0.4083, 0.5775, 0.5733], [0.6276, 0.4120, 0.8738, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892], [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4944, 0.3168, 0.8474, 0.5816, 0.4411, 0.4373, 0.5721, 0.5677], [0.5039, 0.3218, 0.8871, 0.5070, 0.4029, 0.5259, 0.5760, 0.5276], [0.4978, 0.3210, 0.7504, 0.2609, 0.4081, 0.2607, 0.6218, 0.6058], [0.5300, 0.3422, 0.7354, 0.2737, 0.3297, 0.3932, 0.5870, 0.5635], [0.4631, 0.3005, 0.8866, 0.3447, 0.4234, 0.2134, 0.6178, 0.4854], [0.5129, 0.3247, 0.8528, 0.3257, 0.3440, 0.3944, 0.5765, 0.5320], [0.4362, 0.2714, 0.8773, 0.4883, 0.3986, 0.4727, 0.5806, 0.5211], [0.4654, 0.2932, 0.8854, 0.3447, 0.3446, 0.4048, 0.6510, 0.5646]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617], [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033], [0.6193, 0.4078, 0.7287, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6151, 0.4058, 0.7068, 0.2680, 0.3400, 0.4083, 0.5775, 0.5733], [0.6276, 0.4120, 0.8737, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892], [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6257, 0.4167, 0.8775, 0.3433, 0.3562, 0.4133, 0.6200, 0.5667]]], device='cuda:0') loss_train_step before backward: tensor(0.0036, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0036, device='cuda:0', grad_fn=) loss_train: 0.11592077181558125 step: 65 running loss: 0.0017833964894704807 Train Steps: 65/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006], [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552], [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373], [0.6115, 0.3998, 0.7063, 0.2383, 0.4037, 0.1950, 0.5320, 0.4993], [0.6157, 0.4102, 0.8513, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890], [0.6304, 0.4024, 0.8925, 0.4800, 0.3937, 0.4817, 0.7485, 0.5297], [0.6173, 0.4114, 0.7325, 0.2500, 0.4213, 0.1917, 0.5338, 0.5700], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.5804, 0.3857, 0.8771, 0.5844, 0.3679, 0.4868, 0.6360, 0.5012], [-0.0425, -0.0367, 0.8551, 0.2400, 0.4732, 0.2295, 0.6997, 0.5585], [ 0.5929, 0.4005, 0.8401, 0.5809, 0.3690, 0.4718, 0.5810, 0.5505], [ 0.5663, 0.3888, 0.7067, 0.2198, 0.3865, 0.2174, 0.5131, 0.5074], [ 0.5684, 0.3669, 0.8644, 0.3607, 0.3334, 0.3551, 0.5328, 0.5657], [ 0.5612, 0.3696, 0.8814, 0.4700, 0.3682, 0.4813, 0.6840, 0.5249], [ 0.4214, 0.2801, 0.7307, 0.2213, 0.4080, 0.1993, 0.5412, 0.5526], [ 0.5440, 0.3590, 0.7677, 0.2257, 0.3728, 0.2917, 0.5940, 0.5315]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6310, 0.4017, 0.8562, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006], [0.0000, 0.0000, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552], [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373], [0.6115, 0.3998, 0.7063, 0.2383, 0.4038, 0.1950, 0.5320, 0.4993], [0.6157, 0.4102, 0.8512, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890], [0.6304, 0.4024, 0.8925, 0.4800, 0.3938, 0.4817, 0.7485, 0.5297], [0.6173, 0.4114, 0.7325, 0.2500, 0.4212, 0.1917, 0.5337, 0.5700], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.11747965749236755 step: 66 running loss: 0.0017799948104904174 Train Steps: 66/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397], [0.6129, 0.4063, 0.8738, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874], [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162], [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823], [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.6205, 0.4062, 0.8337, 0.2683, 0.3675, 0.4283, 0.6338, 0.5250], [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6138, 0.5450], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6251, 0.4033, 0.8557, 0.5007, 0.3239, 0.3519, 0.5428, 0.5372], [0.6137, 0.4180, 0.8449, 0.5570, 0.3967, 0.4510, 0.5643, 0.5701], [0.5641, 0.3861, 0.8715, 0.4379, 0.3183, 0.4774, 0.6737, 0.5093], [0.5425, 0.3741, 0.7356, 0.3609, 0.3198, 0.3816, 0.5545, 0.5847], [0.3630, 0.2469, 0.8479, 0.2203, 0.5101, 0.2250, 0.7429, 0.5562], [0.5277, 0.3549, 0.7876, 0.2757, 0.3350, 0.4166, 0.6158, 0.5284], [0.5695, 0.3940, 0.8755, 0.4750, 0.3733, 0.5183, 0.6559, 0.5532], [0.4964, 0.3130, 0.7487, 0.1767, 0.4193, 0.1578, 0.6286, 0.4901]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6127, 0.4084, 0.8737, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397], [0.6130, 0.4063, 0.8737, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874], [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162], [0.6170, 0.4102, 0.7468, 0.3695, 0.3462, 0.3767, 0.5238, 0.5823], [0.6300, 0.4133, 0.8537, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.6205, 0.4062, 0.8338, 0.2683, 0.3675, 0.4283, 0.6338, 0.5250], [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6137, 0.5450], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895]]], device='cuda:0') loss_train_step before backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train: 0.1202581440738868 step: 67 running loss: 0.001794897672744579 Train Steps: 67/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183], [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050], [0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6100, 0.4071, 0.7601, 0.3444, 0.3400, 0.4117, 0.5625, 0.5617], [0.6339, 0.4118, 0.7988, 0.5800, 0.3912, 0.4583, 0.7343, 0.5760]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6612, 0.4562, 0.8680, 0.4165, 0.4319, 0.5259, 0.6222, 0.5346], [0.5574, 0.3880, 0.8802, 0.4529, 0.4421, 0.5177, 0.6578, 0.5421], [0.5582, 0.3715, 0.8900, 0.4436, 0.3643, 0.3686, 0.6026, 0.5147], [0.5936, 0.4037, 0.8495, 0.4358, 0.3438, 0.3282, 0.5475, 0.5556], [0.5840, 0.4030, 0.7509, 0.2079, 0.3725, 0.2356, 0.6260, 0.4932], [0.4665, 0.3168, 0.7085, 0.2098, 0.3752, 0.2044, 0.5738, 0.5676], [0.6158, 0.4249, 0.7574, 0.3089, 0.3399, 0.4049, 0.5947, 0.5401], [0.5303, 0.3628, 0.7998, 0.4980, 0.3657, 0.4049, 0.6914, 0.5614]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6251, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6087, 0.5183], [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050], [0.6151, 0.4125, 0.8737, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6100, 0.4071, 0.7601, 0.3444, 0.3400, 0.4117, 0.5625, 0.5617], [0.6339, 0.4118, 0.7987, 0.5800, 0.3913, 0.4583, 0.7343, 0.5760]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.12180890384479426 step: 68 running loss: 0.0017913074094822684 Train Steps: 68/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567], [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598], [0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378], [0.6272, 0.4045, 0.8538, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6138, 0.5883]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6241, 0.4234, 0.8645, 0.3934, 0.3759, 0.5098, 0.5802, 0.5197], [0.6142, 0.4060, 0.7109, 0.2683, 0.4481, 0.1873, 0.5449, 0.5926], [0.5609, 0.3858, 0.8440, 0.4253, 0.4511, 0.4539, 0.5817, 0.5489], [0.6234, 0.4384, 0.8018, 0.4867, 0.4361, 0.4341, 0.5862, 0.5596], [0.5419, 0.3619, 0.8450, 0.2001, 0.4939, 0.2217, 0.7119, 0.5358], [0.6548, 0.4339, 0.8147, 0.5305, 0.3407, 0.3900, 0.5933, 0.4804], [0.6723, 0.4529, 0.8368, 0.3916, 0.3305, 0.3355, 0.6153, 0.4993], [0.6250, 0.4366, 0.8396, 0.3740, 0.4185, 0.5401, 0.6053, 0.5594]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6186, 0.4129, 0.8900, 0.4567, 0.3938, 0.5417, 0.5734, 0.5110], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6112, 0.4029, 0.8637, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567], [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598], [0.6293, 0.4097, 0.8800, 0.2517, 0.5263, 0.2600, 0.7430, 0.5378], [0.6271, 0.4045, 0.8537, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6137, 0.5883]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.12344274375936948 step: 69 running loss: 0.00178902527187492 Train Steps: 69/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6177, 0.4086, 0.8738, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700], [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413], [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864], [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852], [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5988, 0.5700], [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235], [0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7387, 0.4992, 0.8326, 0.3715, 0.3934, 0.5256, 0.6269, 0.5265], [0.6438, 0.4278, 0.9021, 0.3559, 0.4733, 0.2499, 0.6971, 0.5084], [0.6759, 0.4547, 0.8371, 0.5015, 0.4044, 0.4597, 0.6859, 0.5400], [0.6668, 0.4625, 0.8443, 0.4836, 0.4699, 0.4555, 0.5151, 0.5617], [0.5322, 0.3633, 0.7571, 0.2769, 0.3974, 0.3349, 0.5683, 0.5541], [0.6474, 0.4479, 0.8151, 0.3181, 0.3610, 0.4445, 0.6198, 0.5369], [0.6664, 0.4622, 0.8580, 0.4708, 0.4462, 0.5022, 0.6261, 0.5201], [0.7322, 0.4684, 0.8168, 0.2213, 0.4268, 0.2554, 0.6238, 0.4816]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6177, 0.4085, 0.8737, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700], [0.6271, 0.4040, 0.9137, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413], [0.6314, 0.4107, 0.8750, 0.5100, 0.3787, 0.4900, 0.7121, 0.5864], [0.6140, 0.4070, 0.8700, 0.5000, 0.4613, 0.4900, 0.5260, 0.5852], [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5987, 0.5700], [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235], [0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.12500527323572896 step: 70 running loss: 0.0017857896176532708 Train Steps: 70/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123], [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6147, 0.4107, 0.8137, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412], [0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917], [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261], [ nan, nan, 0.8850, 0.3000, 0.5363, 0.2250, 0.7343, 0.5771], [0.6222, 0.4169, 0.8638, 0.5650, 0.4313, 0.4783, 0.5637, 0.5633], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6740, 0.4580, 0.7269, 0.2416, 0.3939, 0.2809, 0.5639, 0.5228], [0.6931, 0.4447, 0.8656, 0.4627, 0.3803, 0.4282, 0.6171, 0.5161], [0.7225, 0.4748, 0.7947, 0.3181, 0.3930, 0.2757, 0.5183, 0.5355], [0.7884, 0.5138, 0.8540, 0.5704, 0.4757, 0.4530, 0.5667, 0.5913], [0.7516, 0.4976, 0.8647, 0.4521, 0.4303, 0.4952, 0.5614, 0.5281], [0.2397, 0.1529, 0.8816, 0.2423, 0.5373, 0.2346, 0.7495, 0.5470], [0.7616, 0.5046, 0.8529, 0.5398, 0.4444, 0.4879, 0.6154, 0.5691], [0.7566, 0.5095, 0.8891, 0.4354, 0.4529, 0.5162, 0.6032, 0.5257]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123], [0.6236, 0.3965, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6147, 0.4107, 0.8138, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412], [0.6248, 0.4185, 0.8500, 0.5767, 0.4462, 0.4550, 0.5612, 0.5917], [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261], [0.0000, 0.0000, 0.8850, 0.3000, 0.5362, 0.2250, 0.7343, 0.5771], [0.6222, 0.4169, 0.8637, 0.5650, 0.4313, 0.4783, 0.5638, 0.5633], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533]]], device='cuda:0') loss_train_step before backward: tensor(0.0039, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0039, device='cuda:0', grad_fn=) loss_train: 0.1288923279789742 step: 71 running loss: 0.0018153849011123128 Train Steps: 71/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153], [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364], [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200], [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617], [0.6336, 0.4154, 0.8900, 0.2767, 0.4988, 0.2867, 0.7422, 0.5540], [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250], [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6854, 0.4290, 0.8553, 0.5348, 0.4771, 0.5437, 0.5710, 0.5228], [0.7214, 0.4744, 0.7999, 0.2458, 0.4623, 0.2501, 0.6241, 0.5392], [0.6655, 0.4104, 0.8681, 0.3181, 0.4673, 0.2397, 0.6046, 0.5253], [0.7321, 0.4662, 0.8887, 0.5671, 0.4110, 0.4492, 0.5422, 0.5732], [0.5737, 0.3805, 0.8622, 0.3099, 0.5229, 0.3049, 0.6988, 0.5457], [0.6894, 0.4463, 0.8710, 0.4764, 0.4769, 0.5825, 0.5848, 0.5603], [0.7157, 0.4586, 0.8985, 0.4550, 0.4095, 0.4748, 0.6335, 0.5365], [0.6618, 0.4263, 0.8856, 0.5448, 0.3901, 0.4869, 0.5635, 0.5772]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153], [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364], [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200], [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617], [0.6336, 0.4154, 0.8900, 0.2767, 0.4988, 0.2867, 0.7422, 0.5540], [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6263, 0.4065, 0.9038, 0.4317, 0.3587, 0.4550, 0.6325, 0.5250], [0.6222, 0.4171, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.1301426654390525 step: 72 running loss: 0.0018075370199868404 Train Steps: 72/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6263, 0.4233, 0.7924, 0.4626, 0.3788, 0.2883, 0.5573, 0.6047], [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058], [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500], [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350], [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6125, 0.4076, 0.8488, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6557, 0.4289, 0.8310, 0.4977, 0.4275, 0.3516, 0.5835, 0.6059], [0.6078, 0.3892, 0.8621, 0.5863, 0.4557, 0.5347, 0.6241, 0.6111], [0.6815, 0.4376, 0.8850, 0.4496, 0.4109, 0.5026, 0.5828, 0.5285], [0.6125, 0.3737, 0.9020, 0.4991, 0.4120, 0.4265, 0.5849, 0.5473], [0.7450, 0.4814, 0.8346, 0.2803, 0.4986, 0.2320, 0.6264, 0.5490], [0.6894, 0.4417, 0.8345, 0.4276, 0.3791, 0.4259, 0.5487, 0.5544], [0.6251, 0.3859, 0.8485, 0.2816, 0.4423, 0.3012, 0.6581, 0.5162], [0.6389, 0.3790, 0.8842, 0.4048, 0.3998, 0.4133, 0.5169, 0.5496]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6263, 0.4232, 0.7924, 0.4626, 0.3787, 0.2883, 0.5573, 0.6047], [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6084, 0.4005, 0.8400, 0.4317, 0.3762, 0.4750, 0.5476, 0.5058], [0.6262, 0.4163, 0.8850, 0.5183, 0.3762, 0.4150, 0.6025, 0.5500], [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350], [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6125, 0.4076, 0.8487, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.13164239979232661 step: 73 running loss: 0.0018033205451003646 Train Steps: 73/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [ nan, nan, 0.8850, 0.3000, 0.5363, 0.2250, 0.7343, 0.5771], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767], [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050], [ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767], [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7152, 0.4496, 0.8840, 0.5291, 0.4344, 0.5146, 0.5446, 0.5550], [0.2075, 0.1350, 0.8997, 0.3122, 0.5219, 0.2621, 0.7280, 0.5755], [0.7042, 0.4605, 0.8933, 0.3257, 0.4916, 0.2425, 0.6603, 0.5587], [0.7824, 0.4959, 0.8963, 0.5029, 0.3762, 0.4046, 0.5804, 0.6031], [0.7340, 0.4657, 0.8780, 0.5537, 0.4509, 0.5468, 0.5906, 0.5369], [0.2747, 0.1706, 0.8041, 0.3679, 0.4138, 0.2918, 0.5423, 0.6006], [0.7430, 0.4759, 0.8953, 0.5835, 0.3869, 0.4296, 0.5988, 0.5045], [0.7167, 0.4848, 0.8927, 0.4600, 0.3611, 0.4525, 0.5368, 0.5691]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.0000, 0.0000, 0.8850, 0.3000, 0.5362, 0.2250, 0.7343, 0.5771], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767], [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050], [0.0000, 0.0000, 0.7613, 0.3250, 0.4038, 0.2533, 0.5437, 0.5767], [0.6272, 0.4071, 0.8737, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350]]], device='cuda:0') loss_train_step before backward: tensor(0.0047, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0047, device='cuda:0', grad_fn=) loss_train: 0.13630416334490292 step: 74 running loss: 0.001841948153309499 Train Steps: 74/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552], [0.6250, 0.4103, 0.8950, 0.4400, 0.3912, 0.5650, 0.6050, 0.5133], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6147, 0.4112, 0.7988, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550], [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6109, 0.4015, 0.7668, 0.3639, 0.3513, 0.3667, 0.5200, 0.5641], [0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.0903, 0.0535, 0.9101, 0.3129, 0.5345, 0.2583, 0.7578, 0.5770], [0.6861, 0.4237, 0.9318, 0.5016, 0.4171, 0.6002, 0.6311, 0.5333], [0.6454, 0.4002, 0.8520, 0.3570, 0.4135, 0.2628, 0.6123, 0.5308], [0.6206, 0.4030, 0.8520, 0.3886, 0.3837, 0.2987, 0.5037, 0.5705], [0.6418, 0.3951, 0.8160, 0.2760, 0.4705, 0.2130, 0.6188, 0.5424], [0.6373, 0.3980, 0.8801, 0.6439, 0.4178, 0.4880, 0.5721, 0.6193], [0.6372, 0.4049, 0.8112, 0.4223, 0.3590, 0.3866, 0.5287, 0.5580], [0.6187, 0.3832, 0.9024, 0.5265, 0.3736, 0.4140, 0.5123, 0.5496]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552], [0.6250, 0.4103, 0.8950, 0.4400, 0.3913, 0.5650, 0.6050, 0.5133], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6147, 0.4112, 0.7987, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550], [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6109, 0.4015, 0.7668, 0.3639, 0.3512, 0.3667, 0.5200, 0.5641], [0.6151, 0.4125, 0.8737, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.13764630918740295 step: 75 running loss: 0.001835284122498706 Train Steps: 75/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921], [ nan, nan, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411], [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767], [ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6749, 0.4133, 0.8857, 0.2980, 0.4145, 0.2898, 0.6248, 0.5178], [0.1693, 0.0882, 0.9349, 0.3785, 0.5117, 0.2394, 0.6746, 0.5773], [0.7113, 0.4677, 0.8635, 0.3449, 0.4212, 0.2230, 0.5482, 0.5491], [0.6812, 0.4329, 0.9257, 0.5342, 0.3713, 0.3869, 0.5587, 0.6029], [0.1745, 0.1099, 0.7082, 0.3025, 0.4149, 0.2373, 0.5410, 0.5802], [0.6236, 0.4131, 0.7854, 0.2501, 0.4213, 0.2076, 0.5841, 0.5242], [0.0469, 0.0247, 0.7511, 0.2891, 0.4237, 0.2496, 0.5318, 0.5650], [0.7042, 0.4386, 0.9340, 0.5136, 0.3466, 0.3730, 0.6089, 0.5196]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921], [0.0000, 0.0000, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411], [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767], [0.0000, 0.0000, 0.6688, 0.2513, 0.4112, 0.2117, 0.5193, 0.5933], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.0000, 0.0000, 0.6900, 0.1917, 0.3938, 0.2367, 0.5240, 0.5246], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672]]], device='cuda:0') loss_train_step before backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train: 0.14028748925193213 step: 76 running loss: 0.0018458880164727912 Train Steps: 76/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633], [0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [ nan, nan, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702], [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6346, 0.4092, 0.7712, 0.5917, 0.4037, 0.4767, 0.7343, 0.5725]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5165, 0.3344, 0.8865, 0.3937, 0.3729, 0.2577, 0.5661, 0.4983], [0.5008, 0.3059, 0.8650, 0.5040, 0.4121, 0.4633, 0.5061, 0.5278], [0.5801, 0.3869, 0.8603, 0.5037, 0.3318, 0.3774, 0.5080, 0.5630], [0.4800, 0.3126, 0.8583, 0.4781, 0.3460, 0.3829, 0.5058, 0.5076], [0.0449, 0.0263, 0.9105, 0.3653, 0.4914, 0.2295, 0.6939, 0.5560], [0.5305, 0.3265, 0.8573, 0.3543, 0.3323, 0.3917, 0.5863, 0.5516], [0.5581, 0.3568, 0.8780, 0.3115, 0.4651, 0.2026, 0.6345, 0.5329], [0.5063, 0.3323, 0.8051, 0.5204, 0.3592, 0.4526, 0.6360, 0.5617]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6277, 0.4117, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5638, 0.5633], [0.6108, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.0000, 0.0000, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702], [0.6257, 0.4167, 0.8775, 0.3433, 0.3562, 0.4133, 0.6200, 0.5667], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6346, 0.4092, 0.7713, 0.5917, 0.4038, 0.4767, 0.7343, 0.5725]]], device='cuda:0') loss_train_step before backward: tensor(0.0025, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0025, device='cuda:0', grad_fn=) loss_train: 0.14280669260188006 step: 77 running loss: 0.0018546323714529879 Train Steps: 77/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6224, 0.4097, 0.7438, 0.2267, 0.3850, 0.2850, 0.5988, 0.5250], [0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585], [0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583], [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378], [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332], [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320], [0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433], [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5059, 0.3345, 0.7602, 0.2278, 0.3854, 0.2539, 0.6262, 0.5230], [0.4361, 0.2765, 0.8633, 0.4468, 0.4724, 0.5165, 0.5810, 0.5478], [0.4907, 0.3082, 0.9063, 0.4370, 0.3787, 0.3822, 0.5449, 0.5404], [0.4585, 0.2929, 0.8790, 0.5238, 0.3940, 0.4032, 0.4904, 0.5269], [0.4985, 0.3049, 0.9053, 0.4899, 0.3699, 0.3515, 0.6380, 0.5238], [0.4820, 0.3069, 0.8524, 0.5386, 0.3745, 0.4547, 0.6722, 0.5328], [0.5487, 0.3470, 0.7873, 0.2789, 0.3601, 0.3334, 0.6035, 0.5362], [0.4773, 0.2953, 0.8434, 0.4209, 0.3548, 0.3967, 0.5461, 0.5606]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6224, 0.4097, 0.7437, 0.2267, 0.3850, 0.2850, 0.5987, 0.5250], [0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585], [0.6212, 0.4033, 0.8938, 0.4167, 0.3812, 0.4267, 0.5612, 0.5583], [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378], [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332], [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320], [0.6175, 0.4091, 0.7862, 0.2800, 0.3638, 0.3583, 0.6187, 0.5433], [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719]]], device='cuda:0') loss_train_step before backward: tensor(0.0039, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0039, device='cuda:0', grad_fn=) loss_train: 0.14673250037594698 step: 78 running loss: 0.0018811859022557305 Train Steps: 78/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250], [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433], [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823], [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116], [0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4087, 0.2881, 0.8793, 0.3682, 0.3442, 0.2797, 0.6288, 0.5055], [0.4526, 0.2877, 0.8537, 0.5116, 0.3869, 0.4292, 0.5877, 0.5334], [0.4888, 0.3502, 0.7298, 0.2433, 0.3949, 0.1769, 0.5642, 0.5608], [0.4653, 0.2900, 0.8206, 0.5220, 0.3491, 0.4446, 0.6688, 0.4878], [0.4368, 0.2940, 0.8106, 0.4830, 0.4568, 0.4552, 0.5322, 0.5297], [0.4606, 0.2997, 0.7351, 0.3265, 0.3292, 0.3455, 0.5507, 0.5565], [0.4449, 0.2896, 0.8538, 0.3646, 0.3796, 0.5156, 0.5988, 0.5139], [0.4937, 0.3109, 0.8220, 0.1960, 0.3931, 0.2451, 0.6621, 0.4891]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250], [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433], [0.6170, 0.4102, 0.7468, 0.3695, 0.3462, 0.3767, 0.5238, 0.5823], [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116], [0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921]]], device='cuda:0') loss_train_step before backward: tensor(0.0059, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0059, device='cuda:0', grad_fn=) loss_train: 0.15266501865698956 step: 79 running loss: 0.0019324685905948045 Train Steps: 79/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650], [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517], [ nan, nan, 0.8300, 0.3150, 0.3588, 0.3383, 0.5208, 0.5194], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6179, 0.3993, 0.8925, 0.4789, 0.3879, 0.4900, 0.6041, 0.5279], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5307, 0.3548, 0.7096, 0.2187, 0.4055, 0.2188, 0.5669, 0.5435], [0.4671, 0.3458, 0.8251, 0.3786, 0.3618, 0.3229, 0.5673, 0.5475], [0.4901, 0.3563, 0.7372, 0.1861, 0.4123, 0.2281, 0.6333, 0.5263], [0.0950, 0.0853, 0.7791, 0.2953, 0.3558, 0.3173, 0.5706, 0.5129], [0.5282, 0.3683, 0.6496, 0.2782, 0.3489, 0.2879, 0.5350, 0.5610], [0.5229, 0.3438, 0.8743, 0.4487, 0.3905, 0.4842, 0.6071, 0.5113], [0.4942, 0.3315, 0.7959, 0.2371, 0.3885, 0.2666, 0.6473, 0.4877], [0.5228, 0.3476, 0.7379, 0.1952, 0.3836, 0.2282, 0.6192, 0.5012]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6193, 0.4050, 0.7312, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650], [0.6200, 0.4071, 0.7337, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517], [0.0000, 0.0000, 0.8300, 0.3150, 0.3587, 0.3383, 0.5208, 0.5194], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6179, 0.3993, 0.8925, 0.4789, 0.3879, 0.4900, 0.6041, 0.5279], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395]]], device='cuda:0') loss_train_step before backward: tensor(0.0023, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0023, device='cuda:0', grad_fn=) loss_train: 0.15501353502622806 step: 80 running loss: 0.0019376691878278506 Train Steps: 80/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667], [0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700], [ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426], [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494], [ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6282, 0.4381, 0.8355, 0.2734, 0.3580, 0.4126, 0.6383, 0.5545], [0.6837, 0.4846, 0.7717, 0.3039, 0.3803, 0.3083, 0.5922, 0.5750], [0.0733, 0.0855, 0.7373, 0.2484, 0.3598, 0.3237, 0.5362, 0.5699], [0.6103, 0.4397, 0.8611, 0.3480, 0.3714, 0.3243, 0.5854, 0.5111], [0.6272, 0.4369, 0.8275, 0.4628, 0.3831, 0.4366, 0.6994, 0.5166], [0.0427, 0.0495, 0.7091, 0.1944, 0.4089, 0.2482, 0.5838, 0.5537], [0.6725, 0.4609, 0.7970, 0.5225, 0.3743, 0.5087, 0.5892, 0.4707], [0.5609, 0.4096, 0.6612, 0.1675, 0.4200, 0.2040, 0.5145, 0.5588]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6257, 0.4167, 0.8775, 0.3433, 0.3562, 0.4133, 0.6200, 0.5667], [0.6212, 0.4171, 0.7875, 0.3633, 0.3812, 0.2933, 0.5675, 0.5700], [0.0000, 0.0000, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426], [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494], [0.0000, 0.0000, 0.7425, 0.2117, 0.3938, 0.2433, 0.5437, 0.5567], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6149, 0.4054, 0.6712, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.15654646544135176 step: 81 running loss: 0.0019326724128561946 Train Steps: 81/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320], [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261], [0.6224, 0.4097, 0.7438, 0.2267, 0.3850, 0.2850, 0.5988, 0.5250], [0.6203, 0.4021, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563], [0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391], [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6120, 0.4161, 0.8576, 0.3980, 0.3587, 0.5396, 0.7251, 0.5467], [0.5780, 0.4041, 0.8139, 0.4110, 0.3942, 0.5006, 0.5479, 0.5276], [0.6369, 0.4528, 0.6879, 0.1798, 0.3727, 0.2992, 0.6297, 0.5245], [0.7181, 0.4893, 0.8373, 0.4649, 0.3389, 0.4007, 0.5916, 0.5105], [0.6550, 0.4485, 0.8356, 0.3766, 0.3475, 0.3610, 0.6218, 0.5490], [0.0318, 0.0481, 0.7165, 0.2537, 0.3442, 0.3365, 0.5431, 0.5739], [0.6474, 0.4496, 0.7977, 0.2374, 0.4135, 0.2600, 0.5980, 0.5348], [0.5745, 0.4167, 0.8119, 0.4426, 0.4552, 0.4981, 0.5370, 0.5871]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6250, 0.3993, 0.9137, 0.4333, 0.3762, 0.5217, 0.6995, 0.5320], [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261], [0.6224, 0.4097, 0.7437, 0.2267, 0.3850, 0.2850, 0.5987, 0.5250], [0.6203, 0.4020, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.0000, 0.0000, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563], [0.6199, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391], [0.6140, 0.4070, 0.8700, 0.5000, 0.4613, 0.4900, 0.5260, 0.5852]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.1579744432528969 step: 82 running loss: 0.001926517600645084 Train Steps: 82/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6188, 0.5400], [0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517], [ nan, nan, 0.8888, 0.3100, 0.5262, 0.2817, 0.7145, 0.6003], [0.6250, 0.4103, 0.8950, 0.4400, 0.3912, 0.5650, 0.6050, 0.5133], [0.6182, 0.3930, 0.8841, 0.3892, 0.3556, 0.4967, 0.6222, 0.5279], [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881], [0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6289, 0.4397, 0.7994, 0.3143, 0.3484, 0.3620, 0.6211, 0.5480], [0.7012, 0.4679, 0.6978, 0.2448, 0.4033, 0.2456, 0.5379, 0.5728], [0.6532, 0.4595, 0.7230, 0.2023, 0.4137, 0.2560, 0.6047, 0.5566], [0.0559, 0.0454, 0.8784, 0.3179, 0.4937, 0.2871, 0.7252, 0.5912], [0.7048, 0.4770, 0.8636, 0.4523, 0.3856, 0.6014, 0.5997, 0.5220], [0.5908, 0.3908, 0.8337, 0.4236, 0.3456, 0.4994, 0.6183, 0.5246], [0.6762, 0.4809, 0.8649, 0.5304, 0.3847, 0.4387, 0.4774, 0.5786], [0.6734, 0.4679, 0.7195, 0.1926, 0.3988, 0.2341, 0.5801, 0.5138]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6187, 0.5400], [0.6193, 0.4050, 0.7312, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6200, 0.4071, 0.7337, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517], [0.0000, 0.0000, 0.8888, 0.3100, 0.5263, 0.2817, 0.7145, 0.6003], [0.6250, 0.4103, 0.8950, 0.4400, 0.3913, 0.5650, 0.6050, 0.5133], [0.6182, 0.3930, 0.8841, 0.3892, 0.3556, 0.4967, 0.6222, 0.5279], [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881], [0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.15900916929240339 step: 83 running loss: 0.0019157731240048602 Train Steps: 83/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097], [0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334], [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183], [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6063, 0.6083], [0.6202, 0.4064, 0.7879, 0.2179, 0.4567, 0.1725, 0.5955, 0.5478], [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285], [0.6167, 0.4048, 0.6831, 0.3639, 0.3763, 0.3017, 0.5700, 0.5883]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6739, 0.4526, 0.8910, 0.4254, 0.4190, 0.5681, 0.5595, 0.5604], [0.6833, 0.4371, 0.8709, 0.4826, 0.3629, 0.4658, 0.6200, 0.5308], [0.6546, 0.4241, 0.8739, 0.4820, 0.4487, 0.5598, 0.6135, 0.5418], [0.6296, 0.4391, 0.8380, 0.3504, 0.3509, 0.4107, 0.5820, 0.6039], [0.5579, 0.3854, 0.7547, 0.2289, 0.4464, 0.1491, 0.5852, 0.5591], [0.6199, 0.3966, 0.8858, 0.4370, 0.4615, 0.5330, 0.6153, 0.5827], [0.6631, 0.4274, 0.7721, 0.4041, 0.3427, 0.4250, 0.5122, 0.5284], [0.5237, 0.3468, 0.7270, 0.3245, 0.3668, 0.2981, 0.5689, 0.5847]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097], [0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334], [0.6251, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6087, 0.5183], [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6062, 0.6083], [0.6202, 0.4064, 0.7879, 0.2179, 0.4567, 0.1725, 0.5955, 0.5478], [0.6184, 0.4079, 0.8838, 0.4617, 0.4837, 0.5650, 0.6175, 0.5850], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285], [0.6167, 0.4048, 0.6831, 0.3639, 0.3762, 0.3017, 0.5700, 0.5883]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.15980508815846406 step: 84 running loss: 0.0019024415256960008 Train Steps: 84/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6124, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402], [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6171, 0.3926, 0.8730, 0.5601, 0.3915, 0.4549, 0.5773, 0.5414], [0.6341, 0.4047, 0.7808, 0.2310, 0.4821, 0.1496, 0.5899, 0.5704], [0.7446, 0.4741, 0.8923, 0.5481, 0.3716, 0.4588, 0.5961, 0.5184], [0.6517, 0.4116, 0.7889, 0.2948, 0.3571, 0.4012, 0.5962, 0.5773], [0.6654, 0.4209, 0.9023, 0.4934, 0.4563, 0.5767, 0.6043, 0.5427], [0.6486, 0.4106, 0.8826, 0.4646, 0.3893, 0.5222, 0.5684, 0.5229], [0.7261, 0.4736, 0.8319, 0.5206, 0.3985, 0.4609, 0.5244, 0.5686], [0.6244, 0.3855, 0.9119, 0.4555, 0.4923, 0.5583, 0.5995, 0.5956]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6245, 0.4100, 0.7763, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.6197, 0.4051, 0.7812, 0.2650, 0.3512, 0.4050, 0.6112, 0.5500], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6123, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402], [0.6184, 0.4079, 0.8838, 0.4617, 0.4837, 0.5650, 0.6175, 0.5850]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.16067833415581845 step: 85 running loss: 0.0018903333430096288 Train Steps: 85/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6120, 0.4014, 0.6863, 0.2817, 0.3700, 0.2783, 0.5513, 0.5667], [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317], [0.6138, 0.4101, 0.8800, 0.5083, 0.4637, 0.5950, 0.5587, 0.5077], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117], [0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133], [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350], [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207], [0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6802, 0.4376, 0.7374, 0.3089, 0.3743, 0.2980, 0.5208, 0.5690], [0.4057, 0.2461, 0.7853, 0.2706, 0.4651, 0.2423, 0.5779, 0.5631], [0.6088, 0.3779, 0.9175, 0.5766, 0.4961, 0.5800, 0.5669, 0.5666], [0.7017, 0.4283, 0.7751, 0.2334, 0.4296, 0.2204, 0.5703, 0.5378], [0.6362, 0.3832, 0.8968, 0.5974, 0.4216, 0.5279, 0.6449, 0.5289], [0.6274, 0.3839, 0.9022, 0.5466, 0.4141, 0.4581, 0.5066, 0.5398], [0.6415, 0.3878, 0.8510, 0.3102, 0.4527, 0.2484, 0.6664, 0.5180], [0.7252, 0.4803, 0.9016, 0.4021, 0.3710, 0.4009, 0.5595, 0.5757]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6120, 0.4013, 0.6862, 0.2817, 0.3700, 0.2783, 0.5512, 0.5667], [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317], [0.6138, 0.4101, 0.8800, 0.5083, 0.4638, 0.5950, 0.5587, 0.5077], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5788, 0.5117], [0.6308, 0.3990, 0.8687, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133], [0.6143, 0.3982, 0.8650, 0.4883, 0.3913, 0.4317, 0.5315, 0.5350], [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207], [0.6219, 0.4097, 0.8737, 0.3400, 0.3562, 0.4117, 0.5975, 0.5683]]], device='cuda:0') loss_train_step before backward: tensor(0.0025, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0025, device='cuda:0', grad_fn=) loss_train: 0.1631579902896192 step: 86 running loss: 0.0018971859336002232 Train Steps: 86/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567], [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869], [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617], [0.6188, 0.4099, 0.7400, 0.2433, 0.3962, 0.2750, 0.6162, 0.5467], [0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967], [ nan, nan, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999], [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 6.5615e-01, 4.3238e-01, 8.7637e-01, 3.0122e-01, 4.6985e-01, 2.7266e-01, 6.8688e-01, 5.6284e-01], [ 5.5656e-02, -1.0218e-04, 7.6715e-01, 2.8529e-01, 4.1903e-01, 2.4294e-01, 5.3250e-01, 5.5664e-01], [ 7.3798e-01, 4.6379e-01, 8.9042e-01, 3.5780e-01, 4.6348e-01, 2.0265e-01, 5.6376e-01, 5.0279e-01], [ 8.2466e-01, 5.1510e-01, 8.9817e-01, 5.4252e-01, 4.8617e-01, 4.8849e-01, 5.3407e-01, 5.4021e-01], [ 7.7672e-01, 4.9967e-01, 7.5344e-01, 2.7726e-01, 3.9371e-01, 2.8717e-01, 6.0870e-01, 5.4520e-01], [ 7.9712e-01, 4.9606e-01, 9.1677e-01, 5.0082e-01, 4.2626e-01, 5.5632e-01, 5.9605e-01, 5.7758e-01], [ 6.8887e-02, 1.3605e-02, 8.0388e-01, 2.7323e-01, 4.5281e-01, 1.8833e-01, 5.1119e-01, 5.3303e-01], [ 7.6316e-01, 4.6732e-01, 8.8208e-01, 5.5342e-01, 4.2469e-01, 5.1457e-01, 6.1878e-01, 5.2243e-01]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.0000, 0.0000, 0.7425, 0.2117, 0.3938, 0.2433, 0.5437, 0.5567], [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869], [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617], [0.6188, 0.4099, 0.7400, 0.2433, 0.3963, 0.2750, 0.6162, 0.5467], [0.6199, 0.4102, 0.8950, 0.4417, 0.4013, 0.5367, 0.6112, 0.5967], [0.0000, 0.0000, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999], [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301]]], device='cuda:0') loss_train_step before backward: tensor(0.0033, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0033, device='cuda:0', grad_fn=) loss_train: 0.16645036087720655 step: 87 running loss: 0.001913222538818466 Train Steps: 87/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417], [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283], [0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6109, 0.3798, 0.9016, 0.5194, 0.4766, 0.5684, 0.6182, 0.5540], [0.6222, 0.3921, 0.7879, 0.2932, 0.4149, 0.2974, 0.6207, 0.5324], [0.5676, 0.3614, 0.8883, 0.5625, 0.4245, 0.5565, 0.7536, 0.5352], [0.5486, 0.3695, 0.8816, 0.4359, 0.3645, 0.3880, 0.5915, 0.5291], [0.5466, 0.3531, 0.8790, 0.4196, 0.4184, 0.3233, 0.5451, 0.5796], [0.5856, 0.3652, 0.7095, 0.2472, 0.4402, 0.1400, 0.5354, 0.5184], [0.6644, 0.4115, 0.8798, 0.4273, 0.3535, 0.3874, 0.6053, 0.5026], [0.5934, 0.3878, 0.8912, 0.4718, 0.4126, 0.4666, 0.5440, 0.5000]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6243, 0.4128, 0.7763, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417], [0.6339, 0.4123, 0.8637, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283], [0.6250, 0.4236, 0.8637, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.1671832787396852 step: 88 running loss: 0.0018998099856782408 Train Steps: 88/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6265, 0.4088, 0.8025, 0.1850, 0.4163, 0.2500, 0.6290, 0.4947], [0.6268, 0.4029, 0.8500, 0.2683, 0.3937, 0.3500, 0.6860, 0.5297], [0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6197, 0.4118, 0.8688, 0.5517, 0.4037, 0.5233, 0.5875, 0.5600], [0.6267, 0.4080, 0.8438, 0.2633, 0.4763, 0.1800, 0.6259, 0.5240], [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750], [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5255, 0.3113, 0.8251, 0.2187, 0.4319, 0.2542, 0.6413, 0.5076], [0.6474, 0.4166, 0.8514, 0.3105, 0.3775, 0.3558, 0.6948, 0.5177], [0.6135, 0.3969, 0.8811, 0.5613, 0.4470, 0.5547, 0.5503, 0.5267], [0.5737, 0.3469, 0.7954, 0.2748, 0.4297, 0.2041, 0.5773, 0.5026], [0.5767, 0.3556, 0.9012, 0.5930, 0.4184, 0.5899, 0.6142, 0.5412], [0.5345, 0.3243, 0.8682, 0.2854, 0.4664, 0.2013, 0.6259, 0.5182], [0.4622, 0.2721, 0.7097, 0.2735, 0.4363, 0.2165, 0.5343, 0.5638], [0.5916, 0.3473, 0.7170, 0.2494, 0.4071, 0.2011, 0.5076, 0.5404]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6265, 0.4088, 0.8025, 0.1850, 0.4162, 0.2500, 0.6290, 0.4947], [0.6268, 0.4029, 0.8500, 0.2683, 0.3938, 0.3500, 0.6860, 0.5297], [0.6075, 0.4000, 0.8512, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6197, 0.4118, 0.8687, 0.5517, 0.4038, 0.5233, 0.5875, 0.5600], [0.6267, 0.4080, 0.8438, 0.2633, 0.4762, 0.1800, 0.6259, 0.5240], [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5512, 0.5750], [0.6159, 0.4085, 0.6900, 0.2283, 0.4087, 0.1950, 0.5123, 0.5397]]], device='cuda:0') loss_train_step before backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train: 0.16899637822643854 step: 89 running loss: 0.0018988357104094218 Train Steps: 89/90 Loss: 0.0019 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6118, 0.4052, 0.8463, 0.3917, 0.3538, 0.3450, 0.5053, 0.5593], [0.6226, 0.4103, 0.8575, 0.3450, 0.4388, 0.2067, 0.5787, 0.5383], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [ nan, nan, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819], [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578], [0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116], [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4714, 0.3092, 0.8766, 0.3644, 0.3797, 0.3765, 0.5317, 0.5521], [0.7109, 0.4638, 0.8667, 0.3510, 0.4728, 0.2814, 0.5916, 0.5561], [0.5956, 0.3719, 0.9270, 0.5358, 0.3928, 0.4968, 0.6874, 0.4935], [0.2441, 0.1478, 0.7006, 0.2561, 0.4080, 0.2624, 0.5646, 0.5555], [0.6287, 0.4122, 0.7137, 0.2069, 0.4231, 0.2120, 0.5734, 0.5442], [0.6276, 0.4006, 0.9260, 0.3551, 0.4172, 0.2875, 0.6585, 0.5085], [0.6717, 0.4245, 0.7733, 0.1955, 0.4045, 0.2666, 0.6203, 0.5008], [0.5436, 0.3620, 0.9240, 0.5174, 0.4223, 0.4559, 0.5638, 0.5646]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6118, 0.4052, 0.8462, 0.3917, 0.3537, 0.3450, 0.5053, 0.5593], [0.6226, 0.4103, 0.8575, 0.3450, 0.4387, 0.2067, 0.5788, 0.5383], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.0000, 0.0000, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819], [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578], [0.6277, 0.4117, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6165, 0.4106, 0.7575, 0.1733, 0.3837, 0.2650, 0.5680, 0.5116], [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881]]], device='cuda:0') loss_train_step before backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train: 0.17162802946404554 step: 90 running loss: 0.0019069781051560615 Valid Steps: 10/10 Loss: nan 19 -------------------------------------------------- Epoch: 4 Train Loss: 0.0019 Valid Loss: nan -------------------------------------------------- size of train loader is: 90 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833], [ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533], [0.6353, 0.4128, 0.8488, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133], [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567], [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533], [0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5069, 0.3172, 0.9035, 0.4719, 0.3578, 0.5011, 0.6231, 0.4727], [0.2427, 0.1428, 0.7823, 0.2831, 0.3590, 0.3039, 0.5319, 0.5369], [0.4679, 0.2951, 0.8326, 0.2327, 0.5136, 0.2198, 0.6668, 0.5393], [0.6167, 0.3997, 0.7749, 0.2314, 0.4379, 0.2040, 0.6198, 0.5140], [0.6602, 0.4364, 0.8328, 0.2305, 0.4575, 0.1908, 0.6440, 0.5086], [0.6117, 0.3958, 0.7204, 0.2080, 0.3732, 0.2822, 0.5813, 0.5445], [0.6157, 0.3828, 0.7254, 0.2383, 0.3530, 0.3197, 0.6175, 0.5472], [0.6747, 0.4466, 0.8350, 0.2806, 0.4019, 0.2544, 0.5892, 0.5197]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6037, 0.4833], [0.0000, 0.0000, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533], [0.6353, 0.4128, 0.8487, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.6266, 0.4065, 0.8313, 0.2467, 0.4787, 0.1733, 0.6313, 0.5133], [0.6108, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567], [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533], [0.6199, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391]]], device='cuda:0') loss_train_step before backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train: 0.002571904333308339 step: 1 running loss: 0.002571904333308339 Train Steps: 1/90 Loss: 0.0026 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717], [0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006], [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533], [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6312, 0.4185, 0.8692, 0.4429, 0.4182, 0.4553, 0.5748, 0.5425], [0.5685, 0.3769, 0.8712, 0.5344, 0.3641, 0.4595, 0.6565, 0.5075], [0.4979, 0.3211, 0.7142, 0.2441, 0.3522, 0.2886, 0.5398, 0.5386], [0.6197, 0.4256, 0.8434, 0.2249, 0.4742, 0.1780, 0.6291, 0.4914], [0.6333, 0.4119, 0.7307, 0.1883, 0.3516, 0.2513, 0.6118, 0.5479], [0.5472, 0.3613, 0.8589, 0.4923, 0.3900, 0.4429, 0.6165, 0.5297], [0.6158, 0.4280, 0.8171, 0.2819, 0.3469, 0.3697, 0.5905, 0.5223], [0.6046, 0.4057, 0.7983, 0.3157, 0.3680, 0.2756, 0.5851, 0.5720]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5412, 0.5717], [0.6310, 0.4017, 0.8562, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006], [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6114, 0.4018, 0.7212, 0.1967, 0.3762, 0.2700, 0.5875, 0.5533], [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6212, 0.4171, 0.7875, 0.3633, 0.3812, 0.2933, 0.5675, 0.5700]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.003422100911848247 step: 2 running loss: 0.0017110504559241235 Train Steps: 2/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656], [0.6128, 0.4115, 0.7934, 0.3778, 0.3450, 0.4033, 0.5337, 0.5456], [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467], [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374], [0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393], [0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5767, 0.4007, 0.8143, 0.3931, 0.3484, 0.3931, 0.5638, 0.5402], [0.5164, 0.3594, 0.7774, 0.3241, 0.3213, 0.3582, 0.5171, 0.5565], [0.5055, 0.3489, 0.8258, 0.2838, 0.3339, 0.3657, 0.5966, 0.5359], [0.5999, 0.4185, 0.8611, 0.4580, 0.3542, 0.3977, 0.5300, 0.5404], [0.5769, 0.4194, 0.8771, 0.4247, 0.4371, 0.5407, 0.5959, 0.5125], [0.5649, 0.3986, 0.8371, 0.3267, 0.3532, 0.2654, 0.6139, 0.5156], [0.6567, 0.4417, 0.8134, 0.2386, 0.4005, 0.2341, 0.6993, 0.5478], [0.6901, 0.4905, 0.7182, 0.2260, 0.4397, 0.1728, 0.5948, 0.5641]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6091, 0.3997, 0.8314, 0.4334, 0.3787, 0.4550, 0.5213, 0.5656], [0.6128, 0.4115, 0.7934, 0.3778, 0.3450, 0.4033, 0.5337, 0.5456], [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467], [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6155, 0.4124, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374], [0.6286, 0.4086, 0.8408, 0.2801, 0.4162, 0.2800, 0.6725, 0.5393], [0.6229, 0.4198, 0.7663, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.004888763302005827 step: 3 running loss: 0.0016295877673352759 Train Steps: 3/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400], [ nan, nan, 0.7648, 0.2722, 0.3962, 0.2183, 0.5060, 0.5422], [0.6048, 0.3928, 0.8538, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719], [0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980], [0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263], [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097], [0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6569, 0.4591, 0.8035, 0.2211, 0.4456, 0.1612, 0.6344, 0.5325], [0.0221, 0.0467, 0.7281, 0.2313, 0.3881, 0.1999, 0.4928, 0.5521], [0.5861, 0.4172, 0.8259, 0.5173, 0.3827, 0.4669, 0.5640, 0.5321], [0.6071, 0.4254, 0.7368, 0.2558, 0.3392, 0.2520, 0.5336, 0.5193], [0.6452, 0.4404, 0.7274, 0.2249, 0.4207, 0.1520, 0.5773, 0.5537], [0.6508, 0.4609, 0.8622, 0.3731, 0.3328, 0.3513, 0.6068, 0.5783], [0.5856, 0.4143, 0.8582, 0.4284, 0.4329, 0.4699, 0.5170, 0.5379], [0.6478, 0.4482, 0.7278, 0.1706, 0.4110, 0.1783, 0.5851, 0.5236]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400], [0.0000, 0.0000, 0.7648, 0.2722, 0.3963, 0.2183, 0.5060, 0.5422], [0.6048, 0.3928, 0.8537, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719], [0.6079, 0.3964, 0.7420, 0.2958, 0.3562, 0.2917, 0.5351, 0.4980], [0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263], [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097], [0.6202, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.005824644234962761 step: 4 running loss: 0.0014561610587406904 Train Steps: 4/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6286, 0.4055, 0.9000, 0.4717, 0.3763, 0.4683, 0.7018, 0.5494], [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533], [0.6058, 0.3986, 0.8324, 0.4626, 0.3838, 0.4983, 0.5147, 0.5466], [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6226, 0.4463, 0.8285, 0.5319, 0.3688, 0.4619, 0.6373, 0.5457], [0.6219, 0.4211, 0.8754, 0.4207, 0.3663, 0.4380, 0.6535, 0.5222], [0.6355, 0.4175, 0.7015, 0.2082, 0.3690, 0.2846, 0.5753, 0.5700], [0.5749, 0.4067, 0.8070, 0.4229, 0.3914, 0.4444, 0.5041, 0.5446], [0.5692, 0.3908, 0.7660, 0.3264, 0.3533, 0.3449, 0.5118, 0.5689], [0.6845, 0.4591, 0.8447, 0.4739, 0.4148, 0.4918, 0.5787, 0.5144], [0.6625, 0.4511, 0.9037, 0.4784, 0.3745, 0.3178, 0.5695, 0.4884], [0.5560, 0.3884, 0.8544, 0.3927, 0.4004, 0.3863, 0.4718, 0.5870]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6286, 0.4055, 0.9000, 0.4717, 0.3762, 0.4683, 0.7018, 0.5494], [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533], [0.6058, 0.3986, 0.8324, 0.4626, 0.3837, 0.4983, 0.5147, 0.5466], [0.6102, 0.4001, 0.7738, 0.3583, 0.3462, 0.3800, 0.5524, 0.5689], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.0068395029520615935 step: 5 running loss: 0.0013679005904123187 Train Steps: 5/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6223, 0.4028, 0.8988, 0.4200, 0.3763, 0.5733, 0.6375, 0.5167], [0.6179, 0.3993, 0.8925, 0.4789, 0.3879, 0.4900, 0.6041, 0.5279], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431], [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250], [0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133], [0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6524, 0.4446, 0.8739, 0.4292, 0.3901, 0.5684, 0.5870, 0.5248], [0.7536, 0.5074, 0.8733, 0.4629, 0.3983, 0.4563, 0.5661, 0.5410], [0.6150, 0.4130, 0.8696, 0.4876, 0.4082, 0.5064, 0.6075, 0.4961], [0.1934, 0.1508, 0.6788, 0.2390, 0.4383, 0.1802, 0.4801, 0.5844], [0.6349, 0.4411, 0.8529, 0.4943, 0.4403, 0.4505, 0.4819, 0.5348], [0.6732, 0.4605, 0.8986, 0.4346, 0.3841, 0.4376, 0.6269, 0.5354], [0.6724, 0.4510, 0.8377, 0.5237, 0.4102, 0.4649, 0.6140, 0.5327], [0.6730, 0.4634, 0.6818, 0.2965, 0.3441, 0.3061, 0.5348, 0.5776]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6223, 0.4028, 0.8988, 0.4200, 0.3762, 0.5733, 0.6375, 0.5167], [0.6179, 0.3993, 0.8925, 0.4789, 0.3879, 0.4900, 0.6041, 0.5279], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.0000, 0.0000, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431], [0.6263, 0.4065, 0.9038, 0.4317, 0.3587, 0.4550, 0.6325, 0.5250], [0.6308, 0.3990, 0.8687, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133], [0.6134, 0.4090, 0.6926, 0.2819, 0.3537, 0.3233, 0.5562, 0.5667]]], device='cuda:0') loss_train_step before backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train: 0.008879333385266364 step: 6 running loss: 0.001479888897544394 Train Steps: 6/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717], [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167], [0.6058, 0.3986, 0.8324, 0.4626, 0.3838, 0.4983, 0.5147, 0.5466], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882], [0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183], [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6325, 0.4042, 0.8374, 0.5384, 0.4076, 0.5202, 0.6245, 0.5165], [0.5927, 0.3926, 0.8575, 0.4811, 0.4553, 0.5659, 0.5778, 0.5579], [0.5796, 0.3891, 0.6593, 0.2784, 0.3838, 0.2641, 0.5286, 0.5727], [0.5931, 0.3895, 0.8627, 0.5164, 0.4052, 0.5231, 0.5720, 0.5328], [0.5840, 0.3913, 0.8053, 0.4573, 0.3862, 0.4881, 0.5126, 0.5421], [0.5666, 0.4013, 0.8484, 0.4060, 0.3668, 0.3963, 0.5045, 0.5135], [0.5671, 0.3753, 0.8731, 0.3701, 0.4138, 0.2802, 0.5981, 0.5359], [0.6675, 0.4155, 0.8711, 0.4727, 0.3791, 0.4025, 0.6250, 0.5075]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717], [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167], [0.6058, 0.3986, 0.8324, 0.4626, 0.3837, 0.4983, 0.5147, 0.5466], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882], [0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183], [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.009341607365058735 step: 7 running loss: 0.0013345153378655336 Train Steps: 7/90 Loss: 0.0013 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6336, 0.4154, 0.8900, 0.2767, 0.4988, 0.2867, 0.7422, 0.5540], [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850], [0.6248, 0.4032, 0.7738, 0.1900, 0.4813, 0.1400, 0.5941, 0.4904], [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6206, 0.4001, 0.8900, 0.3933, 0.3588, 0.3567, 0.5837, 0.5083], [0.6286, 0.4055, 0.9000, 0.4717, 0.3763, 0.4683, 0.7018, 0.5494], [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5837, 0.5167]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5499, 0.3594, 0.8369, 0.3185, 0.4948, 0.3402, 0.6740, 0.5420], [0.5998, 0.3705, 0.8692, 0.4855, 0.4617, 0.6227, 0.5686, 0.5542], [0.5499, 0.3415, 0.7362, 0.2629, 0.4629, 0.1953, 0.5573, 0.5133], [0.4837, 0.2881, 0.7961, 0.2749, 0.4318, 0.2609, 0.5862, 0.5048], [0.6002, 0.3903, 0.7413, 0.2989, 0.4616, 0.2055, 0.5607, 0.5631], [0.5945, 0.3634, 0.8931, 0.4460, 0.3458, 0.4164, 0.5333, 0.5032], [0.5868, 0.3631, 0.8820, 0.5066, 0.3499, 0.5411, 0.6449, 0.5156], [0.5235, 0.3172, 0.8431, 0.4431, 0.3745, 0.3671, 0.5454, 0.5285]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6336, 0.4154, 0.8900, 0.2767, 0.4988, 0.2867, 0.7422, 0.5540], [0.6184, 0.4079, 0.8838, 0.4617, 0.4837, 0.5650, 0.6175, 0.5850], [0.6248, 0.4032, 0.7738, 0.1900, 0.4812, 0.1400, 0.5941, 0.4904], [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6245, 0.4100, 0.7763, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6206, 0.4001, 0.8900, 0.3933, 0.3587, 0.3567, 0.5838, 0.5083], [0.6286, 0.4055, 0.9000, 0.4717, 0.3762, 0.4683, 0.7018, 0.5494], [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5838, 0.5167]]], device='cuda:0') loss_train_step before backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train: 0.011782382178353146 step: 8 running loss: 0.0014727977722941432 Train Steps: 8/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950], [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6362, 0.5200], [0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.6200, 0.3993, 0.8519, 0.4923, 0.3962, 0.4717, 0.6013, 0.5433], [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200], [0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050], [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317], [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6115, 0.3676, 0.8520, 0.5439, 0.4247, 0.5460, 0.5940, 0.4784], [0.6112, 0.3801, 0.8665, 0.4029, 0.3777, 0.5389, 0.6466, 0.5161], [0.5991, 0.3805, 0.8630, 0.4775, 0.3677, 0.4075, 0.5308, 0.5507], [0.5744, 0.3545, 0.8599, 0.5122, 0.3941, 0.4950, 0.6332, 0.5469], [0.5568, 0.3427, 0.8693, 0.3457, 0.4553, 0.2653, 0.6197, 0.5071], [0.6292, 0.4066, 0.8632, 0.4395, 0.3943, 0.5198, 0.5969, 0.5634], [0.6212, 0.3967, 0.8706, 0.5548, 0.4042, 0.3961, 0.6012, 0.5386], [0.5783, 0.3662, 0.8935, 0.4913, 0.3843, 0.4621, 0.6692, 0.5343]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950], [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6363, 0.5200], [0.6151, 0.4125, 0.8737, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.6200, 0.3993, 0.8519, 0.4923, 0.3963, 0.4717, 0.6012, 0.5433], [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200], [0.6200, 0.4112, 0.8863, 0.4100, 0.3638, 0.4917, 0.6087, 0.6050], [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317], [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.012471756344893947 step: 9 running loss: 0.0013857507049882163 Train Steps: 9/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617], [ nan, nan, 0.7335, 0.2569, 0.3788, 0.2667, 0.5066, 0.5578], [0.6110, 0.4047, 0.8700, 0.4483, 0.3713, 0.3967, 0.5088, 0.5517], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6371, 0.4092, 0.8337, 0.5850, 0.3950, 0.5117, 0.6559, 0.5262], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6793, 0.4314, 0.9187, 0.5516, 0.3941, 0.4325, 0.6246, 0.5432], [ 0.0100, -0.0118, 0.7616, 0.2766, 0.4076, 0.2870, 0.5129, 0.5589], [ 0.6446, 0.4051, 0.9034, 0.4762, 0.3717, 0.4353, 0.5629, 0.5327], [ 0.7062, 0.4283, 0.8897, 0.4961, 0.4185, 0.5016, 0.5951, 0.5012], [ 0.7383, 0.4840, 0.8713, 0.5639, 0.3891, 0.5187, 0.7153, 0.5261], [ 0.6300, 0.3864, 0.8894, 0.5457, 0.4119, 0.4570, 0.5903, 0.4735], [ 0.6149, 0.3706, 0.9511, 0.4093, 0.4241, 0.3094, 0.7230, 0.5335], [ 0.6544, 0.4049, 0.9248, 0.4669, 0.4432, 0.5979, 0.6524, 0.5007]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617], [0.0000, 0.0000, 0.7335, 0.2569, 0.3787, 0.2667, 0.5066, 0.5578], [0.6110, 0.4047, 0.8700, 0.4483, 0.3713, 0.3967, 0.5088, 0.5517], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6371, 0.4092, 0.8338, 0.5850, 0.3950, 0.5117, 0.6559, 0.5262], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.013627727428684011 step: 10 running loss: 0.001362772742868401 Train Steps: 10/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528], [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038], [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5988, 0.5700], [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350], [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967], [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6525, 0.4029, 0.8657, 0.3708, 0.3643, 0.3567, 0.6704, 0.5558], [0.4966, 0.3262, 0.9405, 0.4151, 0.4271, 0.3549, 0.7339, 0.5399], [0.6338, 0.3980, 0.8844, 0.5107, 0.4377, 0.5056, 0.5811, 0.4925], [0.5620, 0.3533, 0.8747, 0.3214, 0.4716, 0.2212, 0.6513, 0.4920], [0.4764, 0.3062, 0.8148, 0.3349, 0.3829, 0.3372, 0.6139, 0.5654], [0.5407, 0.3467, 0.8335, 0.3137, 0.4079, 0.2607, 0.6321, 0.5385], [0.6389, 0.3857, 0.9060, 0.4683, 0.3464, 0.5044, 0.6220, 0.5043], [0.6267, 0.3907, 0.9125, 0.5158, 0.4654, 0.5760, 0.6488, 0.5189]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6332, 0.4165, 0.9100, 0.3350, 0.4187, 0.3683, 0.7438, 0.5528], [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038], [0.6275, 0.4111, 0.8462, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5987, 0.5700], [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350], [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967], [0.6251, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6087, 0.5183]]], device='cuda:0') loss_train_step before backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train: 0.015669901302317157 step: 11 running loss: 0.0014245364820288325 Train Steps: 11/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583], [ nan, nan, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6102, 0.4020, 0.8638, 0.3717, 0.3625, 0.5017, 0.6038, 0.5500], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882], [0.6147, 0.4081, 0.8538, 0.3400, 0.3663, 0.3150, 0.5142, 0.4875], [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6201, 0.4102, 0.7288, 0.2417, 0.4150, 0.2383, 0.6100, 0.5500], [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6875, 0.4300, 0.9321, 0.4452, 0.3982, 0.4376, 0.6041, 0.5300], [0.1043, 0.0346, 0.9332, 0.3317, 0.5154, 0.2238, 0.6927, 0.5586], [0.6547, 0.4155, 0.9068, 0.4049, 0.3867, 0.5096, 0.6642, 0.5283], [0.6674, 0.4435, 0.9025, 0.4268, 0.3800, 0.3965, 0.5753, 0.4976], [0.6864, 0.4318, 0.9109, 0.3977, 0.3846, 0.3165, 0.5818, 0.5175], [0.7053, 0.4394, 0.9215, 0.5413, 0.3811, 0.4696, 0.6207, 0.5601], [0.6223, 0.3908, 0.7892, 0.2697, 0.4158, 0.2622, 0.6334, 0.5526], [0.6602, 0.4094, 0.8130, 0.3379, 0.4130, 0.3217, 0.6394, 0.6013]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6212, 0.4033, 0.8938, 0.4167, 0.3812, 0.4267, 0.5612, 0.5583], [0.0000, 0.0000, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6102, 0.4020, 0.8637, 0.3717, 0.3625, 0.5017, 0.6037, 0.5500], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882], [0.6147, 0.4081, 0.8537, 0.3400, 0.3663, 0.3150, 0.5142, 0.4875], [0.6222, 0.4171, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6201, 0.4102, 0.7287, 0.2417, 0.4150, 0.2383, 0.6100, 0.5500], [0.6196, 0.4094, 0.7563, 0.2817, 0.3938, 0.3183, 0.6012, 0.6183]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.017088345863157883 step: 12 running loss: 0.0014240288219298236 Train Steps: 12/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.6069, 0.3975, 0.8625, 0.5083, 0.4388, 0.5483, 0.5650, 0.4967], [0.6083, 0.3957, 0.8638, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103], [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633], [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5963, 0.6217], [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926], [0.6125, 0.3974, 0.7725, 0.2517, 0.3538, 0.3317, 0.5887, 0.5500]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6531, 0.4223, 0.9179, 0.4122, 0.3533, 0.3904, 0.6084, 0.5556], [0.6554, 0.4147, 0.8955, 0.5243, 0.4580, 0.5364, 0.5704, 0.5223], [0.6251, 0.3943, 0.9011, 0.4907, 0.4358, 0.4927, 0.5703, 0.4922], [0.6102, 0.3983, 0.9248, 0.4518, 0.3971, 0.4422, 0.5567, 0.5213], [0.6523, 0.4251, 0.9362, 0.4552, 0.4239, 0.5249, 0.6157, 0.5596], [0.6320, 0.3951, 0.7458, 0.2448, 0.4413, 0.2426, 0.6295, 0.6072], [0.5919, 0.3868, 0.8213, 0.2366, 0.3866, 0.2564, 0.6227, 0.5115], [0.6126, 0.3744, 0.8217, 0.2681, 0.3642, 0.3385, 0.6378, 0.5634]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.6069, 0.3975, 0.8625, 0.5083, 0.4387, 0.5483, 0.5650, 0.4967], [0.6083, 0.3957, 0.8637, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103], [0.6189, 0.4049, 0.8888, 0.4417, 0.4212, 0.5200, 0.5987, 0.5633], [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5962, 0.6217], [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926], [0.6125, 0.3974, 0.7725, 0.2517, 0.3537, 0.3317, 0.5888, 0.5500]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.01765251744654961 step: 13 running loss: 0.001357885957426893 Train Steps: 13/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650], [0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148], [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364], [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633], [0.6246, 0.4028, 0.8738, 0.4867, 0.4088, 0.5667, 0.6362, 0.5200], [0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285], [0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6367, 0.4247, 0.8898, 0.4514, 0.4261, 0.5047, 0.5741, 0.5611], [0.5919, 0.4006, 0.8134, 0.2667, 0.3914, 0.2788, 0.5439, 0.5663], [0.6725, 0.4472, 0.9134, 0.4913, 0.3964, 0.4697, 0.5581, 0.5180], [0.6398, 0.4302, 0.8322, 0.2005, 0.4468, 0.2317, 0.6397, 0.5476], [0.6671, 0.4439, 0.9001, 0.4696, 0.4171, 0.5686, 0.6304, 0.5652], [0.7045, 0.4583, 0.8897, 0.4612, 0.4242, 0.5617, 0.5993, 0.5274], [0.6130, 0.4137, 0.9090, 0.4562, 0.4153, 0.5455, 0.6352, 0.5445], [0.4926, 0.3300, 0.8518, 0.3653, 0.3680, 0.2756, 0.5241, 0.5692]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6199, 0.4059, 0.8700, 0.4900, 0.4162, 0.5000, 0.6162, 0.5467], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650], [0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148], [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364], [0.6293, 0.4024, 0.8750, 0.5000, 0.4013, 0.5733, 0.7121, 0.5633], [0.6246, 0.4028, 0.8737, 0.4867, 0.4087, 0.5667, 0.6363, 0.5200], [0.6254, 0.3993, 0.8988, 0.4767, 0.3988, 0.5517, 0.6955, 0.5285], [0.6200, 0.4118, 0.8288, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.018795535230310634 step: 14 running loss: 0.0013425382307364739 Train Steps: 14/90 Loss: 0.0013 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507], [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6275, 0.4081, 0.8063, 0.2017, 0.4825, 0.1583, 0.6156, 0.4869], [0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783], [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552], [0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7229, 0.4724, 0.8251, 0.5350, 0.3827, 0.4518, 0.5353, 0.5470], [0.6008, 0.4034, 0.6870, 0.1803, 0.3881, 0.2368, 0.5260, 0.5752], [0.6170, 0.4062, 0.8478, 0.3946, 0.3342, 0.5308, 0.5184, 0.5252], [0.6327, 0.4191, 0.8496, 0.2966, 0.4430, 0.3363, 0.6651, 0.5467], [0.6231, 0.4200, 0.8101, 0.1875, 0.4821, 0.1824, 0.5935, 0.4996], [0.5662, 0.3985, 0.7558, 0.2272, 0.4465, 0.2259, 0.5278, 0.5828], [0.0313, 0.0332, 0.8566, 0.1995, 0.4916, 0.2640, 0.6464, 0.5660], [0.6997, 0.4572, 0.8597, 0.2787, 0.4170, 0.2444, 0.5558, 0.5374]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [0.6072, 0.4029, 0.7038, 0.2150, 0.3913, 0.2267, 0.5516, 0.5507], [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6275, 0.4081, 0.8062, 0.2017, 0.4825, 0.1583, 0.6156, 0.4869], [0.6229, 0.4198, 0.7663, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783], [0.0000, 0.0000, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552], [0.6250, 0.4131, 0.8687, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.019952477450715378 step: 15 running loss: 0.0013301651633810252 Train Steps: 15/90 Loss: 0.0013 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8888, 0.3100, 0.5262, 0.2817, 0.7145, 0.6003], [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285], [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367], [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038], [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517], [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.1825, 0.1387, 0.8564, 0.2657, 0.4663, 0.2815, 0.6554, 0.5692], [0.5838, 0.4080, 0.8435, 0.4816, 0.3514, 0.4680, 0.5071, 0.5772], [0.6640, 0.4447, 0.8520, 0.3452, 0.4214, 0.2683, 0.5725, 0.5439], [0.6804, 0.4549, 0.8350, 0.2710, 0.3790, 0.2731, 0.6021, 0.5237], [0.6128, 0.4227, 0.8599, 0.4379, 0.3931, 0.5277, 0.6612, 0.5508], [0.5641, 0.3913, 0.8124, 0.4318, 0.4103, 0.5226, 0.4824, 0.5051], [0.6186, 0.4344, 0.7072, 0.1699, 0.4006, 0.2641, 0.5666, 0.5555], [0.6137, 0.4181, 0.7518, 0.1825, 0.4275, 0.2110, 0.5735, 0.5496]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8888, 0.3100, 0.5263, 0.2817, 0.7145, 0.6003], [0.6222, 0.4171, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285], [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367], [0.6339, 0.4102, 0.9087, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038], [0.6200, 0.4071, 0.7337, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517], [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323]]], device='cuda:0') loss_train_step before backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train: 0.021972051385091618 step: 16 running loss: 0.001373253211568226 Train Steps: 16/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6371, 0.4092, 0.8337, 0.5850, 0.3950, 0.5117, 0.6559, 0.5262], [0.6203, 0.4021, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6298, 0.4151, 0.8153, 0.3393, 0.3600, 0.4291, 0.5379, 0.5698], [0.6456, 0.4376, 0.8645, 0.4351, 0.3727, 0.4266, 0.6316, 0.5136], [0.5620, 0.3916, 0.8048, 0.4880, 0.3995, 0.4805, 0.5509, 0.6130], [0.6626, 0.4515, 0.8499, 0.3342, 0.3620, 0.3796, 0.5293, 0.5116], [0.5962, 0.3994, 0.8743, 0.4134, 0.4052, 0.5115, 0.7128, 0.5608], [0.6262, 0.4437, 0.8169, 0.4994, 0.3806, 0.5024, 0.6557, 0.5502], [0.6586, 0.4178, 0.8642, 0.4511, 0.3658, 0.3891, 0.5872, 0.5320], [0.6053, 0.3949, 0.8285, 0.4581, 0.4370, 0.5583, 0.6096, 0.5243]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6082, 0.4024, 0.8737, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6339, 0.4102, 0.9087, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6371, 0.4092, 0.8338, 0.5850, 0.3950, 0.5117, 0.6559, 0.5262], [0.6203, 0.4020, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.022940666152862832 step: 17 running loss: 0.001349450950168402 Train Steps: 17/90 Loss: 0.0013 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650], [0.6097, 0.4000, 0.7325, 0.2667, 0.3450, 0.3517, 0.5284, 0.5045], [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822], [0.6151, 0.4058, 0.7068, 0.2680, 0.3400, 0.4083, 0.5775, 0.5733], [ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383], [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567], [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6877, 0.4486, 0.8495, 0.5078, 0.4421, 0.4954, 0.5811, 0.5105], [0.5926, 0.3816, 0.7462, 0.2623, 0.3489, 0.3333, 0.5847, 0.5139], [0.7007, 0.4712, 0.8677, 0.4622, 0.3739, 0.3210, 0.5782, 0.5522], [0.6480, 0.4357, 0.7169, 0.2784, 0.3488, 0.3984, 0.6073, 0.5504], [0.1842, 0.1286, 0.7313, 0.2295, 0.4224, 0.1709, 0.5615, 0.5122], [0.0856, 0.0792, 0.7627, 0.2752, 0.4135, 0.2344, 0.5505, 0.5307], [0.7567, 0.5055, 0.7287, 0.2531, 0.4355, 0.2495, 0.6603, 0.5967], [0.6695, 0.4580, 0.8895, 0.4715, 0.4324, 0.5336, 0.6547, 0.5361]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6098, 0.3991, 0.8637, 0.4717, 0.4263, 0.4967, 0.5213, 0.5650], [0.6097, 0.4000, 0.7325, 0.2667, 0.3450, 0.3517, 0.5284, 0.5045], [0.6182, 0.4058, 0.8737, 0.4350, 0.3562, 0.3400, 0.5290, 0.5822], [0.6151, 0.4058, 0.7068, 0.2680, 0.3400, 0.4083, 0.5775, 0.5733], [0.0000, 0.0000, 0.7268, 0.2333, 0.4125, 0.1933, 0.5113, 0.5383], [0.0000, 0.0000, 0.7515, 0.2708, 0.3988, 0.2267, 0.5163, 0.5567], [0.6193, 0.4078, 0.7287, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6189, 0.4049, 0.8888, 0.4417, 0.4212, 0.5200, 0.5987, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train: 0.025358807033626363 step: 18 running loss: 0.0014088226129792423 Train Steps: 18/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575], [0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846], [0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263], [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6221, 0.4107, 0.7788, 0.3033, 0.3950, 0.2817, 0.6075, 0.5517], [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5624, 0.3717, 0.8425, 0.3025, 0.4063, 0.4040, 0.7155, 0.5545], [0.5104, 0.3487, 0.8494, 0.4525, 0.3910, 0.4377, 0.5365, 0.5834], [0.5058, 0.3339, 0.7332, 0.2560, 0.4338, 0.1844, 0.6126, 0.5182], [0.5532, 0.3823, 0.8483, 0.4023, 0.3636, 0.3770, 0.5509, 0.4996], [0.6263, 0.4146, 0.7884, 0.2855, 0.4145, 0.2659, 0.6401, 0.5257], [0.5362, 0.3845, 0.8102, 0.5580, 0.4090, 0.4167, 0.5855, 0.5826], [0.5352, 0.3609, 0.8476, 0.4723, 0.3517, 0.3063, 0.5778, 0.5627], [0.4725, 0.3160, 0.8095, 0.5373, 0.4074, 0.4968, 0.6818, 0.5003]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575], [0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846], [0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263], [0.6082, 0.4024, 0.8737, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6221, 0.4107, 0.7788, 0.3033, 0.3950, 0.2817, 0.6075, 0.5517], [0.6212, 0.4159, 0.8675, 0.5783, 0.4087, 0.4317, 0.5612, 0.5917], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6299, 0.4008, 0.8450, 0.5350, 0.4212, 0.5000, 0.6350, 0.5100]]], device='cuda:0') loss_train_step before backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train: 0.027124431653646752 step: 19 running loss: 0.001427601665981408 Train Steps: 19/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967], [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864], [0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133], [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411], [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5778, 0.3690, 0.8085, 0.2754, 0.4514, 0.2019, 0.6591, 0.5103], [0.4937, 0.3214, 0.8699, 0.5001, 0.4092, 0.5391, 0.6217, 0.6006], [0.5210, 0.3367, 0.7454, 0.2287, 0.3928, 0.2534, 0.5894, 0.5159], [0.5524, 0.3582, 0.8768, 0.5135, 0.3693, 0.3642, 0.6268, 0.5321], [0.5612, 0.3727, 0.8623, 0.5504, 0.3960, 0.4826, 0.7284, 0.5798], [0.5088, 0.3190, 0.8487, 0.5750, 0.4073, 0.5123, 0.6379, 0.5277], [0.5967, 0.3906, 0.8153, 0.3364, 0.4398, 0.2178, 0.6045, 0.5534], [0.5869, 0.3812, 0.8239, 0.3602, 0.3586, 0.4270, 0.5885, 0.5591]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6199, 0.4102, 0.8950, 0.4417, 0.4013, 0.5367, 0.6112, 0.5967], [0.6165, 0.4106, 0.7575, 0.1733, 0.3837, 0.2650, 0.5680, 0.5116], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6314, 0.4107, 0.8750, 0.5100, 0.3787, 0.4900, 0.7121, 0.5864], [0.6308, 0.3990, 0.8687, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133], [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411], [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.028719637513859197 step: 20 running loss: 0.0014359818756929598 Train Steps: 20/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810], [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427], [0.6125, 0.3974, 0.7725, 0.2517, 0.3538, 0.3317, 0.5887, 0.5500], [ nan, nan, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933], [0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767], [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672], [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6311, 0.3910, 0.8631, 0.4934, 0.3649, 0.3741, 0.5482, 0.5658], [ 0.6389, 0.4029, 0.8508, 0.3148, 0.4316, 0.2565, 0.6621, 0.4937], [ 0.6481, 0.4064, 0.8025, 0.4326, 0.3589, 0.3769, 0.5521, 0.5612], [ 0.6320, 0.3742, 0.7733, 0.2867, 0.3728, 0.3405, 0.6450, 0.5739], [ 0.0239, -0.0033, 0.7480, 0.3124, 0.3933, 0.2245, 0.5602, 0.5838], [ 0.6112, 0.3836, 0.8664, 0.5335, 0.4437, 0.5030, 0.5833, 0.5835], [ 0.6534, 0.3981, 0.8733, 0.5679, 0.3811, 0.4323, 0.6578, 0.5283], [ 0.6353, 0.4044, 0.8751, 0.4401, 0.4185, 0.5854, 0.6248, 0.5464]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6151, 0.4125, 0.8737, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810], [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427], [0.6125, 0.3974, 0.7725, 0.2517, 0.3537, 0.3317, 0.5888, 0.5500], [0.0000, 0.0000, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933], [0.6148, 0.4076, 0.8666, 0.4820, 0.4137, 0.5067, 0.5250, 0.5767], [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672], [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.029402486368780956 step: 21 running loss: 0.0014001183985133789 Train Steps: 21/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350], [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235], [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533], [0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5887, 0.5017], [ nan, nan, 0.8363, 0.3317, 0.3563, 0.3367, 0.5329, 0.5142], [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367], [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7024, 0.4579, 0.8919, 0.4660, 0.3621, 0.3956, 0.5566, 0.5544], [0.6047, 0.3683, 0.8764, 0.5155, 0.4330, 0.5081, 0.5417, 0.5083], [0.5885, 0.3489, 0.9078, 0.5210, 0.4192, 0.5181, 0.6419, 0.5389], [0.2829, 0.1731, 0.7203, 0.2559, 0.4322, 0.1594, 0.5465, 0.5538], [0.6801, 0.4293, 0.8756, 0.5013, 0.3782, 0.4941, 0.5780, 0.5519], [0.3500, 0.2140, 0.8230, 0.3471, 0.3318, 0.3327, 0.5382, 0.5294], [0.6547, 0.4065, 0.8151, 0.2838, 0.4189, 0.2514, 0.6232, 0.5550], [0.5903, 0.3585, 0.7902, 0.3010, 0.3526, 0.3983, 0.6259, 0.5612]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350], [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235], [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533], [0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5888, 0.5017], [0.0000, 0.0000, 0.8363, 0.3317, 0.3562, 0.3367, 0.5329, 0.5142], [0.6204, 0.4110, 0.7912, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367], [0.6197, 0.4051, 0.7812, 0.2650, 0.3512, 0.4050, 0.6112, 0.5500]]], device='cuda:0') loss_train_step before backward: tensor(0.0058, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0058, device='cuda:0', grad_fn=) loss_train: 0.035240815341239795 step: 22 running loss: 0.001601855242783627 Train Steps: 22/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267], [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427], [0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050], [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426], [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533], [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100], [0.6284, 0.4093, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167], [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5260, 0.3243, 0.8976, 0.3190, 0.4337, 0.2847, 0.6278, 0.5195], [0.5957, 0.3750, 0.8220, 0.4138, 0.3370, 0.3987, 0.4945, 0.5374], [0.5451, 0.3495, 0.9044, 0.4324, 0.3525, 0.5072, 0.5708, 0.5915], [0.6046, 0.3757, 0.7889, 0.2466, 0.4441, 0.2022, 0.5529, 0.5444], [0.3285, 0.2059, 0.7271, 0.2460, 0.4224, 0.1879, 0.5087, 0.5453], [0.5804, 0.3538, 0.8625, 0.5487, 0.4006, 0.5067, 0.6161, 0.5008], [0.5873, 0.3687, 0.9221, 0.4838, 0.3646, 0.3840, 0.5822, 0.4999], [0.5823, 0.3752, 0.7353, 0.2549, 0.4104, 0.2001, 0.5275, 0.5490]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6266, 0.4067, 0.8587, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267], [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427], [0.6200, 0.4112, 0.8863, 0.4100, 0.3638, 0.4917, 0.6087, 0.6050], [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426], [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533], [0.6299, 0.4008, 0.8450, 0.5350, 0.4212, 0.5000, 0.6350, 0.5100], [0.6284, 0.4092, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167], [0.6154, 0.4112, 0.7038, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600]]], device='cuda:0') loss_train_step before backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train: 0.03808972876868211 step: 23 running loss: 0.0016560751638557438 Train Steps: 23/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124], [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496], [ nan, nan, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733], [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413], [0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319], [0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494], [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850], [0.6151, 0.4058, 0.7068, 0.2680, 0.3400, 0.4083, 0.5775, 0.5733]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5628, 0.3788, 0.8062, 0.2812, 0.3902, 0.2384, 0.5447, 0.4986], [0.5909, 0.3884, 0.8611, 0.2857, 0.3847, 0.2486, 0.5559, 0.5137], [0.0569, 0.0567, 0.7019, 0.2182, 0.4347, 0.1737, 0.5344, 0.5591], [0.6344, 0.4047, 0.9095, 0.4098, 0.4022, 0.2848, 0.6425, 0.5178], [0.5850, 0.3834, 0.7315, 0.3019, 0.3443, 0.3362, 0.4449, 0.4935], [0.6786, 0.4350, 0.8980, 0.3853, 0.4181, 0.3716, 0.6707, 0.5098], [0.5914, 0.3821, 0.9013, 0.4644, 0.4535, 0.5465, 0.5600, 0.5473], [0.6159, 0.4107, 0.7133, 0.2632, 0.3203, 0.3963, 0.5143, 0.5451]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124], [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496], [0.0000, 0.0000, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733], [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413], [0.6109, 0.4041, 0.6975, 0.3167, 0.3512, 0.3383, 0.5153, 0.5319], [0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494], [0.6184, 0.4079, 0.8838, 0.4617, 0.4837, 0.5650, 0.6175, 0.5850], [0.6151, 0.4058, 0.7068, 0.2680, 0.3400, 0.4083, 0.5775, 0.5733]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.039125452836742625 step: 24 running loss: 0.0016302272015309427 Train Steps: 24/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110], [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6286, 0.4055, 0.9000, 0.4717, 0.3763, 0.4683, 0.7018, 0.5494], [0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5659, 0.3794, 0.9019, 0.4457, 0.3871, 0.5255, 0.5347, 0.5126], [0.6188, 0.4026, 0.7115, 0.2296, 0.4164, 0.2545, 0.5583, 0.6091], [0.6343, 0.3958, 0.8512, 0.3105, 0.3461, 0.4123, 0.5670, 0.5198], [0.6368, 0.4226, 0.8449, 0.2465, 0.4428, 0.2691, 0.6653, 0.5504], [0.5647, 0.3627, 0.8970, 0.4503, 0.3598, 0.4549, 0.6685, 0.5267], [0.6035, 0.3871, 0.7022, 0.1994, 0.3821, 0.2455, 0.5232, 0.5336], [0.5298, 0.3530, 0.8426, 0.3095, 0.3614, 0.3037, 0.5191, 0.5351], [0.5681, 0.3602, 0.8807, 0.4949, 0.3830, 0.4460, 0.4697, 0.5351]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6186, 0.4129, 0.8900, 0.4567, 0.3938, 0.5417, 0.5734, 0.5110], [0.6193, 0.4078, 0.7287, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6286, 0.4055, 0.9000, 0.4717, 0.3762, 0.4683, 0.7018, 0.5494], [0.6187, 0.4103, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.03991389213479124 step: 25 running loss: 0.0015965556853916496 Train Steps: 25/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6230, 0.4113, 0.7213, 0.1983, 0.4325, 0.2367, 0.6262, 0.5400], [ nan, nan, 0.9088, 0.3783, 0.4562, 0.2617, 0.6741, 0.5575], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800], [0.6289, 0.4032, 0.8419, 0.5446, 0.4075, 0.5017, 0.6312, 0.5117]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5831, 0.3913, 0.7295, 0.1769, 0.4231, 0.2429, 0.5962, 0.5176], [0.3479, 0.2480, 0.9087, 0.3481, 0.4302, 0.2771, 0.6268, 0.5512], [0.5968, 0.4228, 0.8091, 0.2909, 0.3452, 0.4028, 0.5369, 0.5207], [0.6655, 0.4478, 0.7199, 0.2050, 0.4058, 0.2662, 0.5594, 0.6024], [0.6753, 0.4789, 0.8746, 0.4532, 0.3920, 0.5765, 0.6450, 0.5538], [0.6214, 0.4176, 0.8670, 0.5141, 0.3903, 0.4475, 0.5306, 0.5421], [0.5894, 0.4003, 0.7401, 0.1952, 0.4385, 0.2045, 0.5250, 0.5758], [0.6596, 0.4278, 0.8682, 0.4764, 0.3848, 0.4938, 0.6082, 0.5017]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6230, 0.4113, 0.7212, 0.1983, 0.4325, 0.2367, 0.6263, 0.5400], [0.0000, 0.0000, 0.9087, 0.3783, 0.4563, 0.2617, 0.6741, 0.5575], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6193, 0.4078, 0.7287, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6218, 0.4185, 0.7337, 0.2650, 0.4625, 0.1950, 0.5688, 0.5800], [0.6289, 0.4031, 0.8419, 0.5446, 0.4075, 0.5017, 0.6313, 0.5117]]], device='cuda:0') loss_train_step before backward: tensor(0.0038, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0038, device='cuda:0', grad_fn=) loss_train: 0.04367428258410655 step: 26 running loss: 0.0016797800993887135 Train Steps: 26/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064], [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967], [0.6177, 0.4086, 0.8738, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700], [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167], [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6063, 0.6083], [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411], [0.6083, 0.3957, 0.8638, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980], [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6200, 0.4533, 0.8285, 0.4511, 0.4742, 0.4791, 0.5561, 0.5512], [0.5848, 0.3958, 0.8567, 0.3728, 0.3393, 0.5062, 0.5970, 0.5217], [0.6453, 0.4443, 0.8480, 0.3489, 0.3609, 0.5387, 0.6867, 0.5467], [0.5991, 0.4077, 0.8601, 0.4618, 0.3996, 0.4926, 0.6376, 0.5499], [0.6198, 0.4574, 0.8238, 0.3098, 0.3500, 0.4181, 0.6127, 0.6146], [0.5995, 0.4142, 0.7769, 0.2683, 0.4217, 0.1999, 0.5966, 0.5539], [0.6304, 0.4276, 0.8377, 0.4263, 0.4319, 0.5067, 0.5758, 0.5189], [0.6700, 0.4491, 0.8450, 0.3254, 0.4215, 0.2076, 0.6212, 0.5116]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6128, 0.4022, 0.8737, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064], [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967], [0.6177, 0.4085, 0.8737, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700], [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167], [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6062, 0.6083], [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411], [0.6083, 0.3957, 0.8637, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980], [0.6273, 0.4110, 0.8900, 0.3817, 0.4187, 0.2167, 0.5858, 0.4835]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.04473406620672904 step: 27 running loss: 0.0016568172669158903 Train Steps: 27/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274], [0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6284, 0.4093, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167], [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804], [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6346, 0.4165, 0.9138, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6745, 0.4781, 0.7742, 0.3606, 0.4799, 0.2929, 0.5665, 0.6555], [0.5573, 0.3864, 0.8266, 0.2973, 0.4794, 0.2549, 0.6550, 0.5509], [0.6332, 0.4213, 0.7079, 0.2312, 0.4146, 0.2592, 0.5788, 0.5784], [0.5261, 0.3617, 0.8744, 0.4450, 0.3764, 0.4234, 0.6246, 0.5245], [0.6037, 0.4098, 0.8713, 0.3536, 0.3686, 0.4002, 0.6372, 0.5027], [0.7105, 0.4971, 0.6866, 0.2331, 0.4318, 0.2278, 0.5594, 0.5842], [0.5790, 0.3899, 0.8336, 0.3894, 0.3639, 0.5031, 0.5720, 0.5713], [0.5605, 0.3826, 0.8994, 0.3863, 0.4046, 0.4768, 0.7165, 0.5493]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6299, 0.4303, 0.7962, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274], [0.6193, 0.4050, 0.7312, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6284, 0.4092, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167], [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804], [0.6154, 0.4112, 0.7038, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6346, 0.4165, 0.9137, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.045994794316357 step: 28 running loss: 0.0016426712255841786 Train Steps: 28/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103], [0.6129, 0.4063, 0.8738, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874], [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960], [0.6160, 0.4093, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808], [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320], [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750], [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633], [ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6088, 0.4196, 0.8702, 0.4265, 0.4201, 0.4764, 0.5855, 0.5290], [0.7093, 0.4789, 0.8781, 0.5137, 0.4596, 0.4663, 0.5868, 0.5960], [0.7140, 0.4928, 0.8827, 0.4662, 0.4619, 0.4862, 0.6021, 0.5137], [0.7183, 0.4751, 0.8284, 0.4473, 0.3919, 0.4522, 0.5970, 0.5669], [0.6877, 0.4502, 0.8986, 0.4377, 0.3943, 0.4305, 0.7234, 0.5148], [0.7816, 0.5401, 0.6711, 0.2598, 0.4490, 0.1916, 0.5840, 0.5908], [0.7330, 0.4897, 0.8921, 0.4890, 0.4254, 0.5871, 0.7601, 0.5764], [0.1528, 0.1124, 0.7438, 0.2600, 0.3938, 0.2913, 0.5318, 0.5738]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103], [0.6130, 0.4063, 0.8737, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874], [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960], [0.6160, 0.4092, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808], [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320], [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5512, 0.5750], [0.6293, 0.4024, 0.8750, 0.5000, 0.4013, 0.5733, 0.7121, 0.5633], [0.0000, 0.0000, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0030, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0030, device='cuda:0', grad_fn=) loss_train: 0.048961690947180614 step: 29 running loss: 0.001688334170592435 Train Steps: 29/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6122, 0.3993, 0.8738, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927], [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092], [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869], [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250], [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162], [ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617], [0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6794, 0.4479, 0.8627, 0.3964, 0.3654, 0.3532, 0.6276, 0.5969], [0.6532, 0.4360, 0.8675, 0.4889, 0.4654, 0.5083, 0.5451, 0.5391], [0.6788, 0.4359, 0.8759, 0.5463, 0.4182, 0.5275, 0.6290, 0.5163], [0.6798, 0.4425, 0.8446, 0.3496, 0.4765, 0.2144, 0.6339, 0.5350], [0.6577, 0.4384, 0.8928, 0.4528, 0.4095, 0.4527, 0.6980, 0.5613], [0.6829, 0.4486, 0.8871, 0.4562, 0.3815, 0.4578, 0.7193, 0.5286], [0.3316, 0.2288, 0.7307, 0.2535, 0.4495, 0.2235, 0.5810, 0.5954], [0.6351, 0.4448, 0.8648, 0.4567, 0.4051, 0.4139, 0.5898, 0.5463]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6122, 0.3993, 0.8737, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927], [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092], [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869], [0.6263, 0.4065, 0.9038, 0.4317, 0.3587, 0.4550, 0.6325, 0.5250], [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162], [0.0000, 0.0000, 0.7513, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617], [0.6108, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117]]], device='cuda:0') loss_train_step before backward: tensor(0.0036, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0036, device='cuda:0', grad_fn=) loss_train: 0.05260280644870363 step: 30 running loss: 0.0017534268816234543 Train Steps: 30/90 Loss: 0.0018 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683], [0.6203, 0.4021, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6275, 0.4157, 0.8337, 0.5800, 0.3763, 0.4200, 0.5547, 0.6125], [0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700], [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450], [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5934, 0.3907, 0.8409, 0.3052, 0.4328, 0.2824, 0.6257, 0.5610], [0.5742, 0.3419, 0.8959, 0.5044, 0.3988, 0.3896, 0.5758, 0.5160], [0.5488, 0.3516, 0.8657, 0.3616, 0.3794, 0.3953, 0.6189, 0.5567], [0.6236, 0.3822, 0.8468, 0.5635, 0.4088, 0.4823, 0.6499, 0.5146], [0.7318, 0.4671, 0.8511, 0.5613, 0.4100, 0.4089, 0.5688, 0.5884], [0.6320, 0.3917, 0.8081, 0.3804, 0.4160, 0.2909, 0.5750, 0.5517], [0.6117, 0.3748, 0.8802, 0.4653, 0.4100, 0.4083, 0.6179, 0.5411], [0.6195, 0.3790, 0.8797, 0.5160, 0.4637, 0.5493, 0.6130, 0.5171]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6200, 0.4098, 0.8238, 0.2917, 0.4013, 0.2967, 0.6000, 0.5683], [0.6203, 0.4020, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [0.6219, 0.4097, 0.8737, 0.3400, 0.3562, 0.4117, 0.5975, 0.5683], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6275, 0.4157, 0.8338, 0.5800, 0.3762, 0.4200, 0.5547, 0.6125], [0.6212, 0.4171, 0.7875, 0.3633, 0.3812, 0.2933, 0.5675, 0.5700], [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450], [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.05339707116945647 step: 31 running loss: 0.0017224861667566602 Train Steps: 31/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6069, 0.3975, 0.8625, 0.5083, 0.4388, 0.5483, 0.5650, 0.4967], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550], [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116], [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461], [0.6201, 0.4151, 0.8588, 0.5467, 0.3700, 0.3950, 0.5637, 0.5933]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4726, 0.2672, 0.9171, 0.4492, 0.4305, 0.3561, 0.7157, 0.5673], [0.5934, 0.3507, 0.8860, 0.5450, 0.4450, 0.5328, 0.5489, 0.5182], [0.6378, 0.3892, 0.8093, 0.2568, 0.4691, 0.1763, 0.6213, 0.4853], [0.5002, 0.3126, 0.7706, 0.3071, 0.4582, 0.2126, 0.5596, 0.5627], [0.6113, 0.3908, 0.7203, 0.3242, 0.4085, 0.2067, 0.5273, 0.5523], [0.5958, 0.3624, 0.7685, 0.2393, 0.3788, 0.2428, 0.5749, 0.4980], [0.5660, 0.3472, 0.8770, 0.5506, 0.4446, 0.4987, 0.5519, 0.5381], [0.5380, 0.3342, 0.8785, 0.5946, 0.3659, 0.3793, 0.5535, 0.5536]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6069, 0.3975, 0.8625, 0.5083, 0.4387, 0.5483, 0.5650, 0.4967], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6229, 0.4198, 0.7663, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5362, 0.5550], [0.6165, 0.4106, 0.7575, 0.1733, 0.3837, 0.2650, 0.5680, 0.5116], [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461], [0.6202, 0.4151, 0.8587, 0.5467, 0.3700, 0.3950, 0.5638, 0.5933]]], device='cuda:0') loss_train_step before backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train: 0.05541980228736065 step: 32 running loss: 0.0017318688214800204 Train Steps: 32/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6087, 0.3951, 0.8387, 0.5833, 0.4188, 0.4933, 0.5146, 0.4830], [0.6132, 0.4037, 0.6963, 0.2217, 0.4100, 0.1950, 0.5395, 0.5175], [0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350], [0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [0.6198, 0.4115, 0.7762, 0.2717, 0.3713, 0.3200, 0.5837, 0.5683], [0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133], [0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320], [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6566, 0.3715, 0.8816, 0.6038, 0.4144, 0.4639, 0.4896, 0.5030], [0.5464, 0.3426, 0.7286, 0.2806, 0.3849, 0.1901, 0.4855, 0.4883], [0.6351, 0.3949, 0.7865, 0.3325, 0.4041, 0.2448, 0.5552, 0.5239], [0.6059, 0.3743, 0.7327, 0.3129, 0.4484, 0.1898, 0.5123, 0.5763], [0.5316, 0.3136, 0.8101, 0.3168, 0.3599, 0.3169, 0.5467, 0.5450], [0.7205, 0.4120, 0.9059, 0.5588, 0.3834, 0.5009, 0.5938, 0.4897], [0.6610, 0.3933, 0.9364, 0.4873, 0.3515, 0.5126, 0.6947, 0.5138], [0.0570, 0.0145, 0.8727, 0.3291, 0.5222, 0.1890, 0.6622, 0.5340]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6087, 0.3951, 0.8388, 0.5833, 0.4187, 0.4933, 0.5146, 0.4830], [0.6132, 0.4037, 0.6963, 0.2217, 0.4100, 0.1950, 0.5395, 0.5175], [0.6229, 0.4066, 0.7613, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350], [0.6270, 0.4266, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [0.6198, 0.4115, 0.7763, 0.2717, 0.3713, 0.3200, 0.5838, 0.5683], [0.6308, 0.3990, 0.8687, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133], [0.6250, 0.3993, 0.9137, 0.4333, 0.3762, 0.5217, 0.6995, 0.5320], [0.0000, 0.0000, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.05678630419424735 step: 33 running loss: 0.0017207970967953743 Train Steps: 33/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667], [0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913], [0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485], [0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471], [0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5069, 0.3218, 0.8555, 0.3668, 0.3512, 0.3650, 0.5693, 0.5543], [0.4332, 0.2523, 0.8236, 0.2801, 0.4832, 0.1434, 0.5698, 0.5128], [0.6045, 0.3508, 0.8627, 0.6034, 0.4019, 0.4417, 0.5702, 0.5253], [0.5790, 0.3402, 0.8591, 0.3597, 0.3533, 0.3820, 0.6203, 0.5123], [0.5639, 0.3524, 0.7444, 0.3332, 0.3590, 0.2914, 0.5039, 0.5421], [0.6302, 0.3901, 0.8920, 0.4865, 0.3971, 0.4280, 0.4932, 0.5448], [0.4977, 0.3001, 0.7139, 0.2390, 0.3956, 0.1843, 0.5345, 0.5176], [0.6276, 0.3901, 0.8870, 0.5276, 0.4145, 0.4734, 0.5482, 0.5294]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667], [0.6286, 0.4078, 0.8062, 0.2267, 0.4787, 0.1533, 0.5953, 0.4913], [0.6222, 0.3937, 0.8350, 0.5617, 0.4137, 0.4600, 0.5800, 0.5233], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6123, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485], [0.6127, 0.4084, 0.8700, 0.4467, 0.3988, 0.4317, 0.5013, 0.5471], [0.6054, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5412, 0.5717]]], device='cuda:0') loss_train_step before backward: tensor(0.0023, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0023, device='cuda:0', grad_fn=) loss_train: 0.05907099656178616 step: 34 running loss: 0.00173738225181724 Train Steps: 34/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123], [0.6144, 0.4032, 0.8563, 0.3283, 0.3525, 0.4200, 0.5775, 0.5583], [0.6224, 0.4179, 0.8700, 0.5683, 0.4037, 0.4683, 0.5650, 0.5600], [0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071], [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413], [ nan, nan, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220], [0.6275, 0.4048, 0.8488, 0.2883, 0.4463, 0.2033, 0.6321, 0.5155], [0.6176, 0.3911, 0.8738, 0.4217, 0.3488, 0.4033, 0.6025, 0.4817]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5963, 0.3928, 0.8422, 0.3795, 0.3410, 0.4309, 0.5365, 0.5404], [0.6860, 0.4379, 0.8387, 0.3597, 0.3536, 0.4246, 0.5370, 0.5621], [0.6555, 0.4334, 0.8594, 0.6063, 0.3841, 0.4639, 0.5436, 0.5663], [0.6021, 0.3857, 0.7330, 0.2327, 0.3687, 0.2826, 0.5496, 0.5033], [0.5491, 0.3401, 0.9177, 0.3947, 0.4646, 0.2704, 0.7094, 0.5530], [0.0126, 0.0070, 0.7575, 0.2523, 0.3890, 0.2771, 0.4934, 0.5518], [0.7066, 0.4475, 0.8518, 0.2999, 0.4414, 0.2243, 0.6433, 0.5179], [0.5654, 0.3630, 0.8516, 0.4159, 0.3413, 0.4084, 0.5547, 0.5023]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6213, 0.4131, 0.8438, 0.3550, 0.3512, 0.4400, 0.5716, 0.5123], [0.6144, 0.4032, 0.8562, 0.3283, 0.3525, 0.4200, 0.5775, 0.5583], [0.6224, 0.4179, 0.8700, 0.5683, 0.4038, 0.4683, 0.5650, 0.5600], [0.6163, 0.4114, 0.7650, 0.2017, 0.3762, 0.2867, 0.5631, 0.5071], [0.6271, 0.4040, 0.9137, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413], [0.0000, 0.0000, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220], [0.6275, 0.4048, 0.8487, 0.2883, 0.4462, 0.2033, 0.6321, 0.5155], [0.6176, 0.3911, 0.8737, 0.4217, 0.3487, 0.4033, 0.6025, 0.4817]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.05983991172979586 step: 35 running loss: 0.001709711763708453 Train Steps: 35/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804], [0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267], [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6234, 0.4179, 0.7825, 0.3450, 0.3813, 0.2867, 0.5675, 0.5617], [0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767], [0.6097, 0.4024, 0.8488, 0.3717, 0.3875, 0.5517, 0.5836, 0.5591], [0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5901, 0.4088, 0.8563, 0.3605, 0.3315, 0.3447, 0.5860, 0.4939], [0.5359, 0.3621, 0.8485, 0.2784, 0.4185, 0.2434, 0.6223, 0.5242], [0.5851, 0.3998, 0.8402, 0.5130, 0.3952, 0.5066, 0.5769, 0.5002], [0.5745, 0.3787, 0.8197, 0.3058, 0.3349, 0.3723, 0.6011, 0.5366], [0.5456, 0.3811, 0.7809, 0.3230, 0.3732, 0.2617, 0.5631, 0.5644], [0.6271, 0.4279, 0.8482, 0.4558, 0.3857, 0.4892, 0.5296, 0.5692], [0.5132, 0.3385, 0.8294, 0.3506, 0.3816, 0.5313, 0.5937, 0.5284], [0.5584, 0.3972, 0.8556, 0.4872, 0.4582, 0.4736, 0.5132, 0.5291]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804], [0.6266, 0.4067, 0.8587, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267], [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6234, 0.4179, 0.7825, 0.3450, 0.3812, 0.2867, 0.5675, 0.5617], [0.6148, 0.4076, 0.8666, 0.4820, 0.4137, 0.5067, 0.5250, 0.5767], [0.6097, 0.4024, 0.8487, 0.3717, 0.3875, 0.5517, 0.5836, 0.5591], [0.6128, 0.4022, 0.8737, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.060806872352259234 step: 36 running loss: 0.0016890797875627566 Train Steps: 36/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6081, 0.3950, 0.8538, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954], [0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378], [0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300], [0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983], [0.6097, 0.4000, 0.7325, 0.2667, 0.3450, 0.3517, 0.5284, 0.5045], [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397], [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412], [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6335, 0.4346, 0.8782, 0.4073, 0.3632, 0.4977, 0.5222, 0.5061], [0.6177, 0.4202, 0.8657, 0.5371, 0.3564, 0.4533, 0.5620, 0.5361], [0.6114, 0.4360, 0.7637, 0.1912, 0.3863, 0.3127, 0.6328, 0.5387], [0.6132, 0.4126, 0.8703, 0.4958, 0.4261, 0.4992, 0.5843, 0.4942], [0.5834, 0.4001, 0.7386, 0.2229, 0.3346, 0.3644, 0.5495, 0.5346], [0.6371, 0.4433, 0.6882, 0.1988, 0.3955, 0.1833, 0.5116, 0.5594], [0.6863, 0.4815, 0.8768, 0.5128, 0.4381, 0.4661, 0.5487, 0.5464], [0.2251, 0.1772, 0.7638, 0.1963, 0.3851, 0.3008, 0.6064, 0.5866]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6081, 0.3950, 0.8537, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954], [0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378], [0.6207, 0.4081, 0.7663, 0.2067, 0.3963, 0.3200, 0.6313, 0.5300], [0.6219, 0.3934, 0.8687, 0.5267, 0.4313, 0.4967, 0.5987, 0.4983], [0.6097, 0.4000, 0.7325, 0.2667, 0.3450, 0.3517, 0.5284, 0.5045], [0.6159, 0.4085, 0.6900, 0.2283, 0.4087, 0.1950, 0.5123, 0.5397], [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412], [0.0000, 0.0000, 0.7525, 0.2291, 0.3837, 0.3017, 0.6050, 0.5667]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.06275377355632372 step: 37 running loss: 0.0016960479339546953 Train Steps: 37/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6182, 0.3930, 0.8841, 0.3892, 0.3556, 0.4967, 0.6222, 0.5279], [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5923, 0.4131, 0.8545, 0.3751, 0.3961, 0.5505, 0.5932, 0.5513], [0.6174, 0.4432, 0.7769, 0.2844, 0.3408, 0.4038, 0.5863, 0.5461], [0.5197, 0.3747, 0.8393, 0.4754, 0.3550, 0.4011, 0.5515, 0.5350], [0.5182, 0.3501, 0.8382, 0.3487, 0.3390, 0.4937, 0.6212, 0.5191], [0.6206, 0.4425, 0.8216, 0.4711, 0.4638, 0.5038, 0.5184, 0.5643], [0.6210, 0.4301, 0.7798, 0.1495, 0.4564, 0.1728, 0.6289, 0.4910], [0.6128, 0.4325, 0.8455, 0.2361, 0.4737, 0.2206, 0.6600, 0.5483], [0.6007, 0.4236, 0.8583, 0.4114, 0.3887, 0.5358, 0.5525, 0.5445]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6122, 0.4006, 0.8850, 0.4217, 0.4087, 0.5517, 0.6062, 0.5517], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6182, 0.3930, 0.8841, 0.3892, 0.3556, 0.4967, 0.6222, 0.5279], [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5838, 0.5500]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.06382370277424343 step: 38 running loss: 0.001679571125637985 Train Steps: 38/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926], [0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567], [0.6230, 0.4113, 0.7213, 0.1983, 0.4325, 0.2367, 0.6262, 0.5400], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5740, 0.4027, 0.7804, 0.2006, 0.3989, 0.3092, 0.5924, 0.5124], [0.5530, 0.3796, 0.7529, 0.2496, 0.3690, 0.3354, 0.5114, 0.5020], [0.6239, 0.4282, 0.7437, 0.1662, 0.4529, 0.2339, 0.5796, 0.5355], [0.6111, 0.4053, 0.8515, 0.5777, 0.4085, 0.5404, 0.5554, 0.4905], [0.6401, 0.4081, 0.8852, 0.4734, 0.4462, 0.5911, 0.6910, 0.5573], [0.5619, 0.4020, 0.8669, 0.4254, 0.3718, 0.4445, 0.5490, 0.5866], [0.6114, 0.4139, 0.7461, 0.1707, 0.4561, 0.2525, 0.6348, 0.5516], [0.5993, 0.4027, 0.7931, 0.1914, 0.4720, 0.2240, 0.6078, 0.5352]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926], [0.6079, 0.3964, 0.7420, 0.2958, 0.3562, 0.2917, 0.5351, 0.4980], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567], [0.6230, 0.4113, 0.7212, 0.1983, 0.4325, 0.2367, 0.6263, 0.5400], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.06450154227786697 step: 39 running loss: 0.0016538856994324864 Train Steps: 39/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5837, 0.5583], [0.6250, 0.4110, 0.7238, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500], [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804], [0.6109, 0.4015, 0.7668, 0.3639, 0.3513, 0.3667, 0.5200, 0.5641], [0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5910, 0.4063, 0.8675, 0.3079, 0.3950, 0.3282, 0.6319, 0.5391], [0.6301, 0.4437, 0.8669, 0.4167, 0.4878, 0.5472, 0.5882, 0.5389], [0.5174, 0.3574, 0.7003, 0.1974, 0.4357, 0.2005, 0.5694, 0.5409], [0.5996, 0.4060, 0.8421, 0.4511, 0.3845, 0.5482, 0.5982, 0.5075], [0.5995, 0.4152, 0.8363, 0.4647, 0.3705, 0.4227, 0.5858, 0.5332], [0.6194, 0.4364, 0.8583, 0.3602, 0.3719, 0.3915, 0.6232, 0.4786], [0.6075, 0.4301, 0.7486, 0.3395, 0.3588, 0.4001, 0.5355, 0.5402], [0.6480, 0.4403, 0.8199, 0.2405, 0.4479, 0.2270, 0.6470, 0.4727]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5838, 0.5583], [0.6250, 0.4110, 0.7237, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6262, 0.4163, 0.8850, 0.5183, 0.3762, 0.4150, 0.6025, 0.5500], [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804], [0.6109, 0.4015, 0.7668, 0.3639, 0.3512, 0.3667, 0.5200, 0.5641], [0.6277, 0.4056, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.06523508534883149 step: 40 running loss: 0.0016308771337207872 Train Steps: 40/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8888, 0.3100, 0.5262, 0.2817, 0.7145, 0.6003], [0.6147, 0.4081, 0.8538, 0.3400, 0.3663, 0.3150, 0.5142, 0.4875], [0.6126, 0.4039, 0.8237, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138], [0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378], [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578], [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110], [0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.0891, 0.0631, 0.8773, 0.3103, 0.5095, 0.2728, 0.7551, 0.5479], [0.6648, 0.4404, 0.8403, 0.3713, 0.3801, 0.3216, 0.5263, 0.5005], [0.6272, 0.4338, 0.8008, 0.4137, 0.3686, 0.3707, 0.5899, 0.5850], [0.7378, 0.4701, 0.8775, 0.2688, 0.5286, 0.2359, 0.7385, 0.5006], [0.5938, 0.4020, 0.6731, 0.2113, 0.4046, 0.1986, 0.5346, 0.5199], [0.6902, 0.4420, 0.8249, 0.3392, 0.3674, 0.3924, 0.6394, 0.4805], [0.7369, 0.4858, 0.8802, 0.4684, 0.4083, 0.5596, 0.5782, 0.5072], [0.6505, 0.4323, 0.7702, 0.3830, 0.3922, 0.3041, 0.5700, 0.5416]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8888, 0.3100, 0.5263, 0.2817, 0.7145, 0.6003], [0.6147, 0.4081, 0.8537, 0.3400, 0.3663, 0.3150, 0.5142, 0.4875], [0.6126, 0.4038, 0.8238, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138], [0.6293, 0.4097, 0.8800, 0.2517, 0.5263, 0.2600, 0.7430, 0.5378], [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578], [0.6255, 0.4017, 0.8687, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6186, 0.4129, 0.8900, 0.4567, 0.3938, 0.5417, 0.5734, 0.5110], [0.6212, 0.4171, 0.7875, 0.3633, 0.3812, 0.2933, 0.5675, 0.5700]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.06645582363125868 step: 41 running loss: 0.0016208737471038702 Train Steps: 41/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019], [ nan, nan, 0.6512, 0.1717, 0.4100, 0.1983, 0.5253, 0.5240], [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240], [0.6139, 0.4019, 0.7137, 0.2150, 0.4375, 0.1533, 0.5293, 0.5006], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117], [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6891, 0.4599, 0.8777, 0.4752, 0.3979, 0.5489, 0.5949, 0.5166], [0.1478, 0.0824, 0.7012, 0.2130, 0.4257, 0.1784, 0.5816, 0.5343], [0.4423, 0.2698, 0.7500, 0.2366, 0.4303, 0.1242, 0.5547, 0.5396], [0.6466, 0.4050, 0.7299, 0.2141, 0.4476, 0.1427, 0.5879, 0.5038], [0.6940, 0.4596, 0.8240, 0.3588, 0.3734, 0.3761, 0.6182, 0.5333], [0.6037, 0.3890, 0.7641, 0.2457, 0.4580, 0.2087, 0.6180, 0.5329], [0.6408, 0.4135, 0.8916, 0.3551, 0.4042, 0.2893, 0.6089, 0.5080], [0.6872, 0.4480, 0.9009, 0.5420, 0.4146, 0.3651, 0.6688, 0.5193]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6161, 0.4099, 0.8737, 0.4383, 0.3787, 0.5483, 0.5605, 0.5019], [0.0000, 0.0000, 0.6513, 0.1717, 0.4100, 0.1983, 0.5253, 0.5240], [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240], [0.6139, 0.4019, 0.7138, 0.2150, 0.4375, 0.1533, 0.5293, 0.5006], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117], [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309]]], device='cuda:0') loss_train_step before backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train: 0.06857384138857014 step: 42 running loss: 0.00163271050925167 Train Steps: 42/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8300, 0.3150, 0.3588, 0.3383, 0.5208, 0.5194], [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583], [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6200, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367], [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528], [0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517], [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.1296, 0.0703, 0.8199, 0.3338, 0.3646, 0.2900, 0.5425, 0.4966], [0.7671, 0.4832, 0.7103, 0.2382, 0.3894, 0.2149, 0.5864, 0.5264], [0.7036, 0.4180, 0.8707, 0.5230, 0.4307, 0.4502, 0.5363, 0.5218], [0.6885, 0.4491, 0.8544, 0.5484, 0.4120, 0.5351, 0.6645, 0.5570], [0.6785, 0.4147, 0.8790, 0.4961, 0.3750, 0.4876, 0.5926, 0.5207], [0.6171, 0.3889, 0.9387, 0.3866, 0.4237, 0.3242, 0.7216, 0.5220], [0.6880, 0.4228, 0.9020, 0.4466, 0.4167, 0.5017, 0.5754, 0.5190], [0.7329, 0.4590, 0.8893, 0.5391, 0.3890, 0.4513, 0.6981, 0.5399]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8300, 0.3150, 0.3587, 0.3383, 0.5208, 0.5194], [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6087, 0.5583], [0.6126, 0.3954, 0.8537, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6199, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367], [0.6332, 0.4165, 0.9100, 0.3350, 0.4187, 0.3683, 0.7438, 0.5528], [0.6122, 0.4006, 0.8850, 0.4217, 0.4087, 0.5517, 0.6062, 0.5517], [0.6314, 0.4107, 0.8750, 0.5100, 0.3787, 0.4900, 0.7121, 0.5864]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.07046938224812038 step: 43 running loss: 0.0016388228429795439 Train Steps: 43/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6339, 0.4118, 0.7988, 0.5800, 0.3912, 0.4583, 0.7343, 0.5760], [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350], [0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233], [0.6353, 0.4128, 0.8488, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6533, 0.3930, 0.8091, 0.5610, 0.3700, 0.4263, 0.6663, 0.5426], [0.6025, 0.3455, 0.8612, 0.5094, 0.4215, 0.4670, 0.5519, 0.5381], [0.6022, 0.3682, 0.8667, 0.3334, 0.3617, 0.2666, 0.5935, 0.5086], [0.5780, 0.3655, 0.7793, 0.3759, 0.3514, 0.4040, 0.5146, 0.5343], [0.6462, 0.4082, 0.9039, 0.5446, 0.3658, 0.4411, 0.5685, 0.5704], [0.6143, 0.3738, 0.9155, 0.4762, 0.3818, 0.5059, 0.6533, 0.5145], [0.5468, 0.3345, 0.8641, 0.2554, 0.5273, 0.1434, 0.6819, 0.5426], [0.5857, 0.3659, 0.8191, 0.3233, 0.3502, 0.3641, 0.5801, 0.5197]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6339, 0.4118, 0.7987, 0.5800, 0.3913, 0.4583, 0.7343, 0.5760], [0.6126, 0.3954, 0.8537, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350], [0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6222, 0.4171, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233], [0.6353, 0.4128, 0.8487, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.07114345880108885 step: 44 running loss: 0.0016168967909338376 Train Steps: 44/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494], [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301], [0.6208, 0.4082, 0.8538, 0.3067, 0.3588, 0.3717, 0.6112, 0.5517], [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950], [ nan, nan, 0.8525, 0.2217, 0.5413, 0.2367, 0.7367, 0.5482], [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5837, 0.5167], [0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6648, 0.4051, 0.8547, 0.5253, 0.3593, 0.4408, 0.6620, 0.5442], [0.5646, 0.3528, 0.8634, 0.4902, 0.3912, 0.5007, 0.6026, 0.5536], [0.6554, 0.4185, 0.8312, 0.3112, 0.3282, 0.3703, 0.5874, 0.5649], [0.6001, 0.3667, 0.8595, 0.5331, 0.3922, 0.5166, 0.5689, 0.5086], [0.2252, 0.1352, 0.8452, 0.2125, 0.5094, 0.2397, 0.7010, 0.5618], [0.6177, 0.3958, 0.8650, 0.4102, 0.3560, 0.3034, 0.5887, 0.5341], [0.6059, 0.3636, 0.8642, 0.5232, 0.4138, 0.4863, 0.5702, 0.5012], [0.6755, 0.4330, 0.7479, 0.3293, 0.4549, 0.1847, 0.5319, 0.6135]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494], [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301], [0.6208, 0.4082, 0.8537, 0.3067, 0.3587, 0.3717, 0.6112, 0.5517], [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950], [0.0000, 0.0000, 0.8525, 0.2217, 0.5412, 0.2367, 0.7367, 0.5482], [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5838, 0.5167], [0.6219, 0.3934, 0.8687, 0.5267, 0.4313, 0.4967, 0.5987, 0.4983], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.07271968291024677 step: 45 running loss: 0.0016159929535610395 Train Steps: 45/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999], [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320], [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117], [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6353, 0.4128, 0.9138, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991], [0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6014, 0.3906, 0.8760, 0.5141, 0.3662, 0.3807, 0.5987, 0.5159], [0.5682, 0.3432, 0.8353, 0.4166, 0.3854, 0.5007, 0.5066, 0.5207], [0.5977, 0.3875, 0.8380, 0.5157, 0.3754, 0.5035, 0.6486, 0.5565], [0.6293, 0.3856, 0.8906, 0.4898, 0.3633, 0.4632, 0.6219, 0.5406], [0.6094, 0.3830, 0.8745, 0.4607, 0.4074, 0.5207, 0.6207, 0.5795], [0.5853, 0.3684, 0.8890, 0.4049, 0.4215, 0.3655, 0.7029, 0.5958], [0.4055, 0.2576, 0.8998, 0.3556, 0.4626, 0.3115, 0.7230, 0.6119], [0.6224, 0.4179, 0.8621, 0.4533, 0.3715, 0.4162, 0.5211, 0.5285]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6068, 0.3963, 0.8650, 0.4317, 0.4038, 0.5083, 0.5253, 0.4999], [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320], [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117], [0.6241, 0.4142, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6353, 0.4128, 0.9137, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991], [0.6108, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117]]], device='cuda:0') loss_train_step before backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train: 0.07447020555264316 step: 46 running loss: 0.0016189175120139819 Train Steps: 46/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6267, 0.4080, 0.8438, 0.2633, 0.4763, 0.1800, 0.6259, 0.5240], [0.6118, 0.4052, 0.8463, 0.3917, 0.3538, 0.3450, 0.5053, 0.5593], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5637, 0.5083], [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5837, 0.5167], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6041, 0.3938, 0.8301, 0.2768, 0.4639, 0.2062, 0.6434, 0.5514], [0.6029, 0.4004, 0.8234, 0.3782, 0.3560, 0.3713, 0.5147, 0.5802], [0.5090, 0.3451, 0.8887, 0.4574, 0.4401, 0.5328, 0.6144, 0.5899], [0.5273, 0.3276, 0.8808, 0.5208, 0.3927, 0.5246, 0.5674, 0.5295], [0.5351, 0.3431, 0.8607, 0.4114, 0.3778, 0.3320, 0.6319, 0.5436], [0.5631, 0.3568, 0.8728, 0.2954, 0.5005, 0.2429, 0.7542, 0.5450], [0.5440, 0.3390, 0.8046, 0.2910, 0.4050, 0.2686, 0.6255, 0.5356], [0.4561, 0.2983, 0.7701, 0.2767, 0.3897, 0.2866, 0.5682, 0.5719]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6267, 0.4080, 0.8438, 0.2633, 0.4762, 0.1800, 0.6259, 0.5240], [0.6118, 0.4052, 0.8462, 0.3917, 0.3537, 0.3450, 0.5053, 0.5593], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5638, 0.5083], [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5838, 0.5167], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6204, 0.4049, 0.7975, 0.2700, 0.3938, 0.2567, 0.5700, 0.5183]]], device='cuda:0') loss_train_step before backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train: 0.07631988901994191 step: 47 running loss: 0.0016238274259562108 Train Steps: 47/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6238, 0.5350], [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864], [0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954], [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950], [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391], [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5418, 0.3592, 0.8929, 0.3954, 0.3914, 0.5638, 0.6728, 0.5593], [0.5866, 0.3958, 0.8892, 0.5073, 0.3876, 0.5094, 0.7513, 0.5850], [0.6361, 0.4119, 0.8607, 0.3998, 0.4074, 0.2834, 0.5328, 0.5083], [0.5712, 0.3631, 0.8827, 0.5297, 0.4291, 0.5318, 0.6194, 0.4938], [0.5730, 0.3708, 0.8171, 0.2721, 0.4153, 0.2532, 0.6215, 0.5359], [0.5721, 0.3738, 0.7804, 0.2966, 0.3465, 0.3352, 0.5659, 0.5552], [0.5397, 0.3652, 0.8544, 0.3131, 0.3652, 0.3929, 0.6385, 0.5626], [0.5920, 0.4128, 0.8771, 0.5720, 0.4985, 0.4259, 0.6170, 0.6088]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6237, 0.5350], [0.6314, 0.4107, 0.8750, 0.5100, 0.3787, 0.4900, 0.7121, 0.5864], [0.6137, 0.4038, 0.8562, 0.4050, 0.3812, 0.2550, 0.5106, 0.4954], [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950], [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391], [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.0770722002198454 step: 48 running loss: 0.0016056708379134459 Train Steps: 48/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6102, 0.4020, 0.8638, 0.3717, 0.3625, 0.5017, 0.6038, 0.5500], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717], [0.6095, 0.3970, 0.8688, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940], [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567], [ nan, nan, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733], [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6116, 0.4067, 0.8828, 0.3820, 0.3839, 0.5019, 0.6557, 0.5338], [0.7061, 0.4629, 0.8807, 0.5306, 0.3652, 0.3922, 0.5974, 0.5166], [0.6063, 0.4012, 0.8859, 0.4928, 0.4343, 0.4867, 0.5892, 0.5541], [0.5990, 0.3772, 0.9000, 0.4734, 0.4776, 0.4876, 0.5473, 0.5143], [0.6460, 0.4354, 0.8652, 0.4529, 0.3676, 0.4091, 0.5995, 0.5648], [0.0335, 0.0334, 0.7170, 0.1800, 0.4606, 0.2172, 0.5830, 0.5723], [0.6761, 0.4458, 0.8849, 0.4941, 0.3624, 0.4209, 0.6187, 0.5309], [0.6916, 0.4472, 0.8236, 0.2844, 0.4038, 0.2530, 0.6177, 0.5026]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6102, 0.4020, 0.8637, 0.3717, 0.3625, 0.5017, 0.6037, 0.5500], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5412, 0.5717], [0.6095, 0.3970, 0.8687, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940], [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567], [0.0000, 0.0000, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733], [0.6262, 0.4163, 0.8850, 0.5183, 0.3762, 0.4150, 0.6025, 0.5500], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.07790248477249406 step: 49 running loss: 0.001589846628010083 Train Steps: 49/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374], [0.6117, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103], [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501], [0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983], [ nan, nan, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.5667, 0.3969, 0.8627, 0.5026, 0.3836, 0.4889, 0.5366, 0.5161], [ 0.7085, 0.4804, 0.8648, 0.3856, 0.3830, 0.3395, 0.6228, 0.5103], [ 0.6474, 0.4425, 0.6780, 0.2258, 0.3905, 0.2618, 0.5197, 0.4995], [ 0.7389, 0.5162, 0.8219, 0.2639, 0.4505, 0.2242, 0.6105, 0.5333], [ 0.7038, 0.4682, 0.8615, 0.5522, 0.3831, 0.4068, 0.6023, 0.5929], [-0.0587, -0.0297, 0.9008, 0.3119, 0.5381, 0.2447, 0.6911, 0.5574], [ 0.6270, 0.4166, 0.8636, 0.3521, 0.3723, 0.5205, 0.6568, 0.5517], [ 0.6500, 0.4548, 0.8787, 0.3977, 0.3747, 0.4232, 0.5579, 0.4989]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6038, 0.3946, 0.8413, 0.4883, 0.3562, 0.4550, 0.5266, 0.4693], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374], [0.6116, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103], [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501], [0.6260, 0.4214, 0.8537, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983], [0.0000, 0.0000, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6168, 0.4029, 0.8523, 0.3417, 0.3587, 0.5000, 0.6125, 0.5400], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.07921916476334445 step: 50 running loss: 0.001584383295266889 Train Steps: 50/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901], [0.6087, 0.3951, 0.8387, 0.5833, 0.4188, 0.4933, 0.5146, 0.4830], [0.6118, 0.4052, 0.8463, 0.3917, 0.3538, 0.3450, 0.5053, 0.5593], [0.6250, 0.4146, 0.8838, 0.3933, 0.3588, 0.4283, 0.6162, 0.5367], [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6218, 0.4098, 0.7238, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5915, 0.4134, 0.7551, 0.3165, 0.4903, 0.1958, 0.5103, 0.5853], [0.5764, 0.3818, 0.9127, 0.3342, 0.3573, 0.3764, 0.6141, 0.4978], [0.5345, 0.3572, 0.8517, 0.5715, 0.4077, 0.4864, 0.5188, 0.4861], [0.5781, 0.4059, 0.8441, 0.3646, 0.3541, 0.3400, 0.4695, 0.5328], [0.5979, 0.4039, 0.8465, 0.3703, 0.3558, 0.4157, 0.5717, 0.5015], [0.5912, 0.4007, 0.9070, 0.3955, 0.4255, 0.3726, 0.6829, 0.5573], [0.5950, 0.4042, 0.8472, 0.5555, 0.3779, 0.4802, 0.6432, 0.5175], [0.6436, 0.4484, 0.7460, 0.2079, 0.4379, 0.2331, 0.5944, 0.5265]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901], [0.6087, 0.3951, 0.8388, 0.5833, 0.4187, 0.4933, 0.5146, 0.4830], [0.6118, 0.4052, 0.8462, 0.3917, 0.3537, 0.3450, 0.5053, 0.5593], [0.6250, 0.4146, 0.8838, 0.3933, 0.3587, 0.4283, 0.6162, 0.5367], [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6218, 0.4098, 0.7237, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.07985461122007109 step: 51 running loss: 0.0015657766905896292 Train Steps: 51/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6268, 0.4061, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300], [0.6230, 0.4113, 0.7213, 0.1983, 0.4325, 0.2367, 0.6262, 0.5400], [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960], [0.6128, 0.4115, 0.7934, 0.3778, 0.3450, 0.4033, 0.5337, 0.5456], [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317], [0.6218, 0.4098, 0.7238, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350], [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6268, 0.4029, 0.8500, 0.2683, 0.3937, 0.3500, 0.6860, 0.5297]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5507, 0.3677, 0.8739, 0.2727, 0.4537, 0.1950, 0.5911, 0.5079], [0.6034, 0.4039, 0.7401, 0.2273, 0.4371, 0.2180, 0.5913, 0.5303], [0.5627, 0.3646, 0.9055, 0.5243, 0.4412, 0.4729, 0.5209, 0.4831], [0.6226, 0.3989, 0.8272, 0.4100, 0.3243, 0.4034, 0.5167, 0.5258], [0.5484, 0.3707, 0.7738, 0.2498, 0.4351, 0.2388, 0.5541, 0.5149], [0.6328, 0.4251, 0.7578, 0.2479, 0.4367, 0.2240, 0.5967, 0.5330], [0.5407, 0.3771, 0.7163, 0.2564, 0.4186, 0.1785, 0.4994, 0.5358], [0.5917, 0.3917, 0.8435, 0.3103, 0.3848, 0.3313, 0.6699, 0.5156]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6268, 0.4060, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300], [0.6230, 0.4113, 0.7212, 0.1983, 0.4325, 0.2367, 0.6263, 0.5400], [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960], [0.6128, 0.4115, 0.7934, 0.3778, 0.3450, 0.4033, 0.5337, 0.5456], [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317], [0.6218, 0.4098, 0.7237, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350], [0.6154, 0.4112, 0.7038, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6268, 0.4029, 0.8500, 0.2683, 0.3938, 0.3500, 0.6860, 0.5297]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.08078289640252478 step: 52 running loss: 0.001553517238510092 Train Steps: 52/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767], [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700], [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600], [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183], [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800], [0.6273, 0.4100, 0.7137, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.1549, 0.0988, 0.7728, 0.2984, 0.3965, 0.2483, 0.5293, 0.5679], [0.7679, 0.4968, 0.8719, 0.4534, 0.3593, 0.4609, 0.5512, 0.5227], [0.2748, 0.1950, 0.7612, 0.2568, 0.3874, 0.2569, 0.5394, 0.5318], [0.7197, 0.4644, 0.8651, 0.5402, 0.4529, 0.4879, 0.6104, 0.5005], [0.6871, 0.4434, 0.8564, 0.4625, 0.3680, 0.4583, 0.5583, 0.5248], [0.7213, 0.4817, 0.8850, 0.3534, 0.3950, 0.2741, 0.6116, 0.5200], [0.7321, 0.4945, 0.7385, 0.2308, 0.4484, 0.1679, 0.5341, 0.5670], [0.6125, 0.3907, 0.7119, 0.1921, 0.3792, 0.2419, 0.5768, 0.5231]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.7613, 0.3250, 0.4038, 0.2533, 0.5437, 0.5767], [0.6111, 0.3995, 0.8788, 0.4567, 0.3812, 0.4833, 0.5450, 0.5700], [0.0000, 0.0000, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600], [0.6202, 0.4053, 0.8637, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183], [0.6218, 0.4185, 0.7337, 0.2650, 0.4625, 0.1950, 0.5688, 0.5800], [0.6273, 0.4099, 0.7138, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0040, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0040, device='cuda:0', grad_fn=) loss_train: 0.08473318166215904 step: 53 running loss: 0.00159873927664451 Train Steps: 53/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6131, 0.4064, 0.8638, 0.5200, 0.4788, 0.4783, 0.5258, 0.5867], [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.6233, 0.4091, 0.8100, 0.2950, 0.3563, 0.3883, 0.6013, 0.5200], [0.6099, 0.4030, 0.8638, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575], [0.6250, 0.4110, 0.7238, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633], [0.6304, 0.4024, 0.8925, 0.4800, 0.3937, 0.4817, 0.7485, 0.5297], [ nan, nan, 0.9088, 0.3783, 0.4562, 0.2617, 0.6741, 0.5575]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6802, 0.4430, 0.8318, 0.5072, 0.4398, 0.4642, 0.5124, 0.5521], [0.6704, 0.4308, 0.8398, 0.3959, 0.3240, 0.3879, 0.5255, 0.5485], [0.6728, 0.4352, 0.7903, 0.2926, 0.3243, 0.3784, 0.5523, 0.5330], [0.6356, 0.4171, 0.8449, 0.5169, 0.4777, 0.4851, 0.4889, 0.5001], [0.7161, 0.4596, 0.8432, 0.2972, 0.3887, 0.3969, 0.6283, 0.5522], [0.6475, 0.4336, 0.6810, 0.2289, 0.3940, 0.1639, 0.4837, 0.5542], [0.7142, 0.4602, 0.8448, 0.4690, 0.3557, 0.4662, 0.6634, 0.5211], [0.1883, 0.1220, 0.8808, 0.3820, 0.4354, 0.2764, 0.6137, 0.5517]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6132, 0.4063, 0.8637, 0.5200, 0.4787, 0.4783, 0.5258, 0.5867], [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.6233, 0.4091, 0.8100, 0.2950, 0.3562, 0.3883, 0.6012, 0.5200], [0.6098, 0.4030, 0.8637, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575], [0.6250, 0.4110, 0.7237, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633], [0.6304, 0.4024, 0.8925, 0.4800, 0.3938, 0.4817, 0.7485, 0.5297], [0.0000, 0.0000, 0.9087, 0.3783, 0.4563, 0.2617, 0.6741, 0.5575]]], device='cuda:0') loss_train_step before backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train: 0.0869114036031533 step: 54 running loss: 0.0016094704370954315 Train Steps: 54/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750], [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391], [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4948, 0.3271, 0.7504, 0.2717, 0.3895, 0.2462, 0.5128, 0.5725], [0.6211, 0.3864, 0.8843, 0.3831, 0.4266, 0.3624, 0.7101, 0.5913], [0.5479, 0.3498, 0.8209, 0.2813, 0.4246, 0.2453, 0.5822, 0.5351], [0.5730, 0.3568, 0.8808, 0.3202, 0.3858, 0.3009, 0.5986, 0.5536], [0.6147, 0.3810, 0.8437, 0.3561, 0.3642, 0.3641, 0.5878, 0.5957], [0.5775, 0.3750, 0.6776, 0.2280, 0.4179, 0.2321, 0.5617, 0.5697], [0.5753, 0.3730, 0.8436, 0.5263, 0.4640, 0.4712, 0.5426, 0.5266], [0.5807, 0.3767, 0.7651, 0.3495, 0.3590, 0.4238, 0.5293, 0.5445]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6182, 0.4099, 0.7812, 0.3000, 0.3938, 0.2367, 0.5325, 0.5750], [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6199, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391], [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.08789730866556056 step: 55 running loss: 0.0015981328848283737 Train Steps: 55/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6115, 0.4005, 0.8838, 0.3867, 0.3763, 0.4700, 0.5800, 0.5550], [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999], [ nan, nan, 0.9088, 0.3783, 0.4562, 0.2617, 0.6741, 0.5575], [0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [0.6275, 0.4081, 0.8063, 0.2017, 0.4825, 0.1583, 0.6156, 0.4869], [0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650], [0.6188, 0.4099, 0.7400, 0.2433, 0.3962, 0.2750, 0.6162, 0.5467], [0.6127, 0.4118, 0.8650, 0.5083, 0.4088, 0.5367, 0.5300, 0.5456]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6512, 0.4229, 0.8618, 0.3790, 0.3667, 0.4765, 0.5779, 0.5789], [0.6838, 0.4149, 0.8431, 0.4274, 0.3979, 0.5157, 0.5459, 0.5217], [0.1031, 0.0614, 0.8903, 0.3869, 0.4564, 0.2801, 0.6815, 0.5916], [0.5853, 0.3887, 0.6898, 0.2776, 0.4559, 0.1826, 0.5637, 0.6347], [0.6206, 0.3911, 0.7994, 0.2031, 0.4864, 0.1657, 0.6258, 0.5273], [0.6177, 0.4136, 0.8885, 0.4072, 0.3660, 0.4113, 0.6229, 0.6092], [0.6280, 0.4210, 0.7120, 0.2187, 0.3850, 0.2793, 0.6336, 0.5902], [0.6331, 0.4122, 0.8470, 0.4942, 0.4314, 0.5270, 0.5640, 0.5644]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6115, 0.4005, 0.8838, 0.3867, 0.3762, 0.4700, 0.5800, 0.5550], [0.6068, 0.3963, 0.8650, 0.4317, 0.4038, 0.5083, 0.5253, 0.4999], [0.0000, 0.0000, 0.9087, 0.3783, 0.4563, 0.2617, 0.6741, 0.5575], [0.6270, 0.4266, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [0.6275, 0.4081, 0.8062, 0.2017, 0.4825, 0.1583, 0.6156, 0.4869], [0.6203, 0.4096, 0.8863, 0.4267, 0.3537, 0.4117, 0.6025, 0.5650], [0.6188, 0.4099, 0.7400, 0.2433, 0.3963, 0.2750, 0.6162, 0.5467], [0.6127, 0.4118, 0.8650, 0.5083, 0.4087, 0.5367, 0.5300, 0.5456]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.08864424467901699 step: 56 running loss: 0.001582932940696732 Train Steps: 56/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899], [ nan, nan, 0.8938, 0.2850, 0.4662, 0.3117, 0.7406, 0.5528], [0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317], [0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600], [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6746, 0.4382, 0.8947, 0.4301, 0.4024, 0.4315, 0.6823, 0.5842], [0.0795, 0.0425, 0.8813, 0.2713, 0.4756, 0.3016, 0.7231, 0.5805], [0.6175, 0.3949, 0.8288, 0.5494, 0.4417, 0.4728, 0.5768, 0.5437], [0.5646, 0.3568, 0.8741, 0.5262, 0.3934, 0.3711, 0.5676, 0.5532], [0.6259, 0.4015, 0.7931, 0.2748, 0.3971, 0.2923, 0.5981, 0.5459], [0.6373, 0.4096, 0.8234, 0.4940, 0.4075, 0.5483, 0.6636, 0.5948], [0.6988, 0.4403, 0.7122, 0.2193, 0.3589, 0.3156, 0.5807, 0.5981], [0.6013, 0.3702, 0.8606, 0.5430, 0.4338, 0.4950, 0.5684, 0.5725]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6346, 0.4144, 0.9087, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899], [0.0000, 0.0000, 0.8938, 0.2850, 0.4663, 0.3117, 0.7406, 0.5528], [0.6229, 0.4066, 0.8512, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317], [0.6201, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6132, 0.4066, 0.7259, 0.2402, 0.3587, 0.3300, 0.6000, 0.5600], [0.6192, 0.4128, 0.8512, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.08935375869623385 step: 57 running loss: 0.0015676098016883131 Train Steps: 57/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133], [0.6263, 0.4233, 0.7924, 0.4626, 0.3788, 0.2883, 0.5573, 0.6047], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319], [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617], [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5957, 0.3914, 0.8953, 0.5222, 0.4134, 0.4844, 0.6682, 0.5365], [0.5112, 0.3648, 0.8072, 0.4491, 0.4238, 0.3225, 0.6067, 0.6391], [0.4843, 0.3409, 0.7270, 0.2297, 0.4249, 0.2574, 0.5738, 0.5921], [0.5497, 0.3573, 0.8853, 0.4831, 0.4449, 0.5554, 0.6904, 0.5349], [0.5592, 0.3758, 0.7483, 0.2906, 0.3913, 0.3537, 0.5571, 0.5424], [0.5409, 0.3602, 0.9018, 0.5374, 0.4245, 0.4396, 0.6098, 0.5927], [0.5360, 0.3595, 0.8909, 0.2613, 0.4279, 0.2904, 0.6356, 0.5659], [0.4802, 0.3276, 0.8787, 0.4517, 0.3916, 0.3253, 0.5922, 0.6089]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133], [0.6263, 0.4232, 0.7924, 0.4626, 0.3787, 0.2883, 0.5573, 0.6047], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6109, 0.4041, 0.6975, 0.3167, 0.3512, 0.3383, 0.5153, 0.5319], [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617], [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767]]], device='cuda:0') loss_train_step before backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train: 0.09153115292428993 step: 58 running loss: 0.001578123326280861 Train Steps: 58/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567], [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [ nan, nan, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702], [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567], [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114], [0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6248, 0.4224, 0.8997, 0.4786, 0.4800, 0.4956, 0.6026, 0.5643], [0.7256, 0.4915, 0.8637, 0.5552, 0.3857, 0.5023, 0.7066, 0.5893], [0.6146, 0.4201, 0.8895, 0.4822, 0.4621, 0.5014, 0.5588, 0.5500], [0.7021, 0.4626, 0.8826, 0.5517, 0.3648, 0.4587, 0.6539, 0.4981], [0.0649, 0.0642, 0.9014, 0.3397, 0.4924, 0.2533, 0.7295, 0.5833], [0.0735, 0.0730, 0.7442, 0.2641, 0.3869, 0.2435, 0.5210, 0.5732], [0.6739, 0.4641, 0.8749, 0.5194, 0.4548, 0.5321, 0.6159, 0.5258], [0.6096, 0.4239, 0.8694, 0.4396, 0.3547, 0.3754, 0.5269, 0.5667]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6112, 0.4029, 0.8637, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567], [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.0000, 0.0000, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702], [0.0000, 0.0000, 0.7515, 0.2708, 0.3988, 0.2267, 0.5163, 0.5567], [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114], [0.6151, 0.4125, 0.8737, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.09251674797269516 step: 59 running loss: 0.0015680804741134773 Train Steps: 59/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038], [0.6126, 0.4067, 0.8638, 0.5383, 0.4188, 0.4850, 0.5016, 0.5392], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283], [0.6200, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683], [0.6199, 0.4093, 0.7913, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5696, 0.3680, 0.8549, 0.4859, 0.4327, 0.5179, 0.5572, 0.4864], [0.5494, 0.3816, 0.8490, 0.5808, 0.4114, 0.5042, 0.5461, 0.5231], [0.5412, 0.3755, 0.8835, 0.4247, 0.3414, 0.3799, 0.6158, 0.5338], [0.5183, 0.3418, 0.8871, 0.4990, 0.3579, 0.4325, 0.6339, 0.5204], [0.4538, 0.3013, 0.8366, 0.2473, 0.5292, 0.2839, 0.7284, 0.5587], [0.5399, 0.3747, 0.9038, 0.4655, 0.3930, 0.3018, 0.6449, 0.5132], [0.5060, 0.3411, 0.8525, 0.5727, 0.3688, 0.3681, 0.5648, 0.5554], [0.5151, 0.3688, 0.7626, 0.2425, 0.4081, 0.2603, 0.6358, 0.5804]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038], [0.6126, 0.4067, 0.8637, 0.5383, 0.4187, 0.4850, 0.5016, 0.5392], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283], [0.6201, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6300, 0.4133, 0.8537, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5412, 0.5683], [0.6198, 0.4093, 0.7912, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700]]], device='cuda:0') loss_train_step before backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train: 0.09467396742547862 step: 60 running loss: 0.0015778994570913103 Train Steps: 60/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166], [0.6202, 0.4079, 0.8025, 0.2500, 0.3763, 0.3217, 0.6125, 0.5533], [0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343], [0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110], [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5558, 0.3568, 0.8592, 0.3219, 0.4439, 0.2026, 0.6285, 0.4938], [0.5836, 0.3509, 0.9035, 0.5453, 0.4080, 0.5315, 0.6395, 0.5006], [0.6229, 0.4163, 0.8235, 0.3249, 0.3960, 0.3336, 0.6160, 0.5354], [0.5874, 0.3863, 0.8409, 0.5990, 0.4015, 0.4852, 0.6877, 0.5168], [0.4982, 0.3207, 0.9290, 0.5280, 0.4096, 0.5368, 0.5549, 0.4946], [0.5610, 0.3600, 0.8111, 0.3305, 0.3712, 0.4111, 0.6245, 0.5420], [0.5940, 0.3892, 0.8602, 0.3784, 0.3739, 0.3779, 0.6145, 0.5378], [0.5522, 0.3593, 0.7109, 0.2939, 0.4385, 0.2026, 0.5390, 0.5488]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6277, 0.4056, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166], [0.6202, 0.4079, 0.8025, 0.2500, 0.3762, 0.3217, 0.6125, 0.5533], [0.6311, 0.3998, 0.7975, 0.5767, 0.3837, 0.4850, 0.7327, 0.5343], [0.6186, 0.4129, 0.8900, 0.4567, 0.3938, 0.5417, 0.5734, 0.5110], [0.6197, 0.4051, 0.7812, 0.2650, 0.3512, 0.4050, 0.6112, 0.5500], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6149, 0.4054, 0.6712, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.09618941575172357 step: 61 running loss: 0.0015768756680610422 Train Steps: 61/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240], [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598], [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800], [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500], [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5837, 0.5317], [ nan, nan, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933], [0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6293, 0.3999, 0.7578, 0.2511, 0.4518, 0.1490, 0.5578, 0.5300], [0.6185, 0.4012, 0.8560, 0.5950, 0.4820, 0.4643, 0.5926, 0.5478], [0.5366, 0.3654, 0.7644, 0.3100, 0.3725, 0.2572, 0.5464, 0.5521], [0.5998, 0.3699, 0.8737, 0.3380, 0.3826, 0.3640, 0.6650, 0.4941], [0.5691, 0.3774, 0.9011, 0.5399, 0.3759, 0.4019, 0.6141, 0.5238], [0.6344, 0.4208, 0.8663, 0.3617, 0.3727, 0.3084, 0.6055, 0.5220], [0.0288, 0.0073, 0.7313, 0.2803, 0.4008, 0.2289, 0.5316, 0.5480], [0.6353, 0.4239, 0.8864, 0.3825, 0.3525, 0.4215, 0.5744, 0.4790]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240], [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598], [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5462, 0.5800], [0.6255, 0.4017, 0.8687, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6262, 0.4163, 0.8850, 0.5183, 0.3762, 0.4150, 0.6025, 0.5500], [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5838, 0.5317], [0.0000, 0.0000, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933], [0.6213, 0.4131, 0.8438, 0.3550, 0.3512, 0.4400, 0.5716, 0.5123]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.09672151939594187 step: 62 running loss: 0.0015600245063861593 Train Steps: 62/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999], [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050], [0.6277, 0.4083, 0.8350, 0.2717, 0.4562, 0.1800, 0.5918, 0.4878], [0.6117, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103], [0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300], [0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183], [0.6125, 0.4076, 0.8488, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5845, 0.3731, 0.7125, 0.2705, 0.4301, 0.1965, 0.5510, 0.5530], [0.5448, 0.3305, 0.8915, 0.4691, 0.3918, 0.5389, 0.5213, 0.4772], [0.5425, 0.3449, 0.9177, 0.5062, 0.3573, 0.4249, 0.5814, 0.4759], [0.6598, 0.4127, 0.8657, 0.3145, 0.4771, 0.2235, 0.6076, 0.4885], [0.5708, 0.3631, 0.7040, 0.2629, 0.3926, 0.2395, 0.5489, 0.5093], [0.5866, 0.3729, 0.7770, 0.2500, 0.3852, 0.3128, 0.6369, 0.5207], [0.6762, 0.4158, 0.7889, 0.2567, 0.4456, 0.1642, 0.6077, 0.5199], [0.5105, 0.3045, 0.8920, 0.4341, 0.3555, 0.3648, 0.4901, 0.5239]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6149, 0.4054, 0.6712, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6068, 0.3963, 0.8650, 0.4317, 0.4038, 0.5083, 0.5253, 0.4999], [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050], [0.6277, 0.4083, 0.8350, 0.2717, 0.4563, 0.1800, 0.5918, 0.4878], [0.6116, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103], [0.6207, 0.4081, 0.7663, 0.2067, 0.3963, 0.3200, 0.6313, 0.5300], [0.6213, 0.4001, 0.7713, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183], [0.6125, 0.4076, 0.8487, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.0980854112713132 step: 63 running loss: 0.0015569112900208446 Train Steps: 63/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6136, 0.4085, 0.6688, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783], [0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233], [0.6272, 0.4045, 0.8538, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6183, 0.4076, 0.8838, 0.4517, 0.3813, 0.4483, 0.5775, 0.5633], [0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165], [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332], [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6443, 0.4123, 0.6926, 0.2116, 0.3916, 0.2383, 0.5444, 0.5459], [0.5970, 0.3596, 0.8289, 0.5472, 0.4022, 0.4737, 0.5319, 0.5218], [0.6078, 0.3907, 0.8406, 0.5781, 0.3599, 0.4388, 0.5628, 0.4617], [0.5568, 0.3553, 0.8705, 0.3905, 0.3469, 0.3844, 0.4892, 0.4773], [0.7234, 0.4420, 0.8791, 0.4286, 0.3780, 0.4776, 0.5646, 0.5362], [0.6287, 0.4008, 0.8808, 0.3129, 0.3524, 0.3089, 0.5696, 0.4900], [0.7104, 0.4353, 0.9023, 0.4677, 0.3627, 0.4362, 0.6397, 0.5030], [0.6305, 0.4165, 0.8075, 0.1914, 0.4149, 0.2623, 0.6327, 0.5142]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6136, 0.4085, 0.6687, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783], [0.6222, 0.3937, 0.8350, 0.5617, 0.4137, 0.4600, 0.5800, 0.5233], [0.6271, 0.4045, 0.8537, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6082, 0.4024, 0.8737, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6183, 0.4076, 0.8838, 0.4517, 0.3812, 0.4483, 0.5775, 0.5633], [0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165], [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332], [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.09885636498802342 step: 64 running loss: 0.0015446307029378659 Train Steps: 64/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650], [ nan, nan, 0.9088, 0.3783, 0.4562, 0.2617, 0.6741, 0.5575], [0.6224, 0.4179, 0.8700, 0.5683, 0.4037, 0.4683, 0.5650, 0.5600], [ nan, nan, 0.8525, 0.2217, 0.5413, 0.2367, 0.7367, 0.5482], [0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7994, 0.5367, 0.7996, 0.5162, 0.3479, 0.4840, 0.5989, 0.5018], [0.7304, 0.4841, 0.8247, 0.4373, 0.4467, 0.4928, 0.4858, 0.4879], [0.6837, 0.4597, 0.7335, 0.2450, 0.3503, 0.2686, 0.5322, 0.5364], [0.0720, 0.0430, 0.8790, 0.3516, 0.4122, 0.2363, 0.6346, 0.5353], [0.7726, 0.5192, 0.8251, 0.5377, 0.3656, 0.4561, 0.5352, 0.5271], [0.1233, 0.0741, 0.8074, 0.1840, 0.4907, 0.2171, 0.6621, 0.5080], [0.6856, 0.4354, 0.8384, 0.2577, 0.4017, 0.2215, 0.5753, 0.5097], [0.8259, 0.5425, 0.8471, 0.4142, 0.4311, 0.4802, 0.5386, 0.5119]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650], [0.0000, 0.0000, 0.9087, 0.3783, 0.4563, 0.2617, 0.6741, 0.5575], [0.6224, 0.4179, 0.8700, 0.5683, 0.4038, 0.4683, 0.5650, 0.5600], [0.0000, 0.0000, 0.8525, 0.2217, 0.5412, 0.2367, 0.7367, 0.5482], [0.6250, 0.4131, 0.8687, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567]]], device='cuda:0') loss_train_step before backward: tensor(0.0043, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0043, device='cuda:0', grad_fn=) loss_train: 0.10314330333494581 step: 65 running loss: 0.0015868200513068586 Train Steps: 65/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033], [0.6350, 0.4043, 0.8738, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950], [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892], [0.6276, 0.4120, 0.8738, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892], [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750], [0.6224, 0.4061, 0.8988, 0.4300, 0.3838, 0.4750, 0.6112, 0.5483], [0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773], [0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6405, 0.4135, 0.8524, 0.4461, 0.4034, 0.5525, 0.5545, 0.5209], [0.6956, 0.4378, 0.8569, 0.5324, 0.3790, 0.4938, 0.6071, 0.5266], [0.5868, 0.3892, 0.8500, 0.4279, 0.3564, 0.3593, 0.5179, 0.5958], [0.6459, 0.4150, 0.8801, 0.3083, 0.4352, 0.2120, 0.6126, 0.5217], [0.6575, 0.4345, 0.7819, 0.2809, 0.3977, 0.2372, 0.5261, 0.5703], [0.6123, 0.4102, 0.8703, 0.3776, 0.3593, 0.4997, 0.5979, 0.5484], [0.6556, 0.4435, 0.7015, 0.2308, 0.4147, 0.2139, 0.5435, 0.6041], [0.5834, 0.3792, 0.8637, 0.3678, 0.4143, 0.5444, 0.5634, 0.5441]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033], [0.6350, 0.4043, 0.8737, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950], [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892], [0.6276, 0.4120, 0.8737, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892], [0.6182, 0.4099, 0.7812, 0.3000, 0.3938, 0.2367, 0.5325, 0.5750], [0.6224, 0.4061, 0.8988, 0.4300, 0.3837, 0.4750, 0.6112, 0.5483], [0.6147, 0.4026, 0.6600, 0.2467, 0.4087, 0.2150, 0.5489, 0.5773], [0.6122, 0.4006, 0.8850, 0.4217, 0.4087, 0.5517, 0.6062, 0.5517]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.1037804797233548 step: 66 running loss: 0.0015724315109599213 Train Steps: 66/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869], [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285], [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700], [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540], [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [ nan, nan, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6785, 0.4481, 0.8139, 0.2020, 0.4749, 0.1850, 0.6138, 0.5308], [0.6331, 0.4025, 0.8234, 0.2862, 0.3667, 0.3415, 0.6191, 0.5370], [0.6745, 0.4569, 0.8635, 0.4377, 0.4150, 0.5758, 0.6779, 0.5648], [0.7044, 0.4675, 0.8349, 0.4179, 0.3905, 0.4767, 0.5015, 0.5859], [0.6103, 0.4076, 0.8960, 0.4278, 0.4193, 0.3025, 0.6647, 0.5860], [0.7442, 0.4936, 0.8570, 0.4359, 0.4789, 0.4910, 0.5747, 0.5813], [0.7321, 0.5025, 0.8436, 0.4162, 0.3645, 0.3410, 0.6136, 0.5306], [0.0770, 0.0437, 0.7523, 0.2256, 0.3834, 0.2551, 0.5500, 0.5512]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6276, 0.4095, 0.8238, 0.2250, 0.4663, 0.1783, 0.6171, 0.4869], [0.6255, 0.4017, 0.8687, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6254, 0.3993, 0.8988, 0.4767, 0.3988, 0.5517, 0.6955, 0.5285], [0.6111, 0.3995, 0.8788, 0.4567, 0.3812, 0.4833, 0.5450, 0.5700], [0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540], [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.0000, 0.0000, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220]]], device='cuda:0') loss_train_step before backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train: 0.10556276931311004 step: 67 running loss: 0.0015755637210911947 Train Steps: 67/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999], [0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131], [0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5837, 0.5317], [0.6216, 0.4167, 0.8588, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667], [0.6273, 0.4105, 0.8988, 0.4517, 0.3912, 0.2550, 0.5894, 0.4811], [0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5519, 0.3711, 0.8373, 0.3983, 0.3957, 0.4941, 0.5430, 0.5392], [0.5763, 0.3926, 0.8314, 0.3637, 0.3484, 0.4117, 0.5856, 0.5564], [0.5165, 0.3468, 0.8993, 0.3403, 0.4354, 0.3505, 0.7177, 0.5684], [0.6442, 0.4390, 0.8608, 0.4409, 0.3806, 0.4275, 0.5905, 0.5781], [0.6688, 0.4501, 0.8287, 0.3181, 0.3599, 0.2964, 0.6153, 0.5782], [0.6583, 0.4522, 0.8290, 0.5357, 0.4189, 0.5248, 0.6321, 0.5842], [0.6497, 0.4241, 0.8771, 0.4061, 0.4136, 0.2432, 0.6155, 0.5340], [0.6057, 0.4289, 0.8389, 0.4843, 0.4893, 0.4952, 0.5470, 0.5456]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6068, 0.3963, 0.8650, 0.4317, 0.4038, 0.5083, 0.5253, 0.4999], [0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131], [0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5838, 0.5317], [0.6216, 0.4167, 0.8587, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667], [0.6273, 0.4105, 0.8988, 0.4517, 0.3913, 0.2550, 0.5894, 0.4811], [0.6128, 0.4022, 0.8737, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.10667031232151203 step: 68 running loss: 0.0015686810635516474 Train Steps: 68/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110], [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650], [0.6332, 0.4118, 0.9238, 0.4267, 0.4012, 0.4733, 0.7525, 0.5436], [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822], [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411], [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461], [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631], [0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6513, 0.4373, 0.8968, 0.4452, 0.4122, 0.5273, 0.6020, 0.5217], [0.6055, 0.3957, 0.8620, 0.4681, 0.4406, 0.4793, 0.5524, 0.5462], [0.5849, 0.3926, 0.9103, 0.4227, 0.4073, 0.4678, 0.7246, 0.5526], [0.6474, 0.4221, 0.8637, 0.4193, 0.3624, 0.3218, 0.5677, 0.5930], [0.5914, 0.3841, 0.8379, 0.3149, 0.4315, 0.1825, 0.6207, 0.5486], [0.5509, 0.3660, 0.8650, 0.5164, 0.4588, 0.5093, 0.5628, 0.5499], [0.5631, 0.3833, 0.8470, 0.3880, 0.3613, 0.4727, 0.5943, 0.5614], [0.6208, 0.4106, 0.8653, 0.5627, 0.3887, 0.4721, 0.6646, 0.5145]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6186, 0.4129, 0.8900, 0.4567, 0.3938, 0.5417, 0.5734, 0.5110], [0.6098, 0.3991, 0.8637, 0.4717, 0.4263, 0.4967, 0.5213, 0.5650], [0.6332, 0.4118, 0.9237, 0.4267, 0.4013, 0.4733, 0.7525, 0.5436], [0.6182, 0.4058, 0.8737, 0.4350, 0.3562, 0.3400, 0.5290, 0.5822], [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411], [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461], [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631], [0.6310, 0.4017, 0.8562, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.10714027981157415 step: 69 running loss: 0.001552757678428611 Train Steps: 69/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6197, 0.4118, 0.8688, 0.5517, 0.4037, 0.5233, 0.5875, 0.5600], [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400], [0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6262, 0.5367], [0.6261, 0.4066, 0.8325, 0.2150, 0.4763, 0.2667, 0.7002, 0.5633], [ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567], [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6797, 0.4425, 0.8824, 0.5833, 0.4246, 0.5367, 0.5965, 0.5615], [0.5917, 0.3821, 0.8896, 0.3713, 0.3764, 0.4707, 0.6431, 0.5447], [0.6304, 0.4132, 0.9070, 0.4863, 0.4543, 0.5783, 0.6085, 0.5383], [0.6103, 0.3874, 0.8507, 0.2457, 0.4849, 0.2317, 0.7062, 0.5364], [0.1297, 0.0855, 0.7537, 0.2486, 0.4253, 0.2378, 0.5901, 0.5524], [0.6390, 0.4320, 0.9092, 0.5563, 0.3869, 0.4732, 0.5894, 0.5641], [0.6987, 0.4476, 0.8577, 0.6106, 0.3919, 0.5049, 0.6735, 0.5354], [0.6582, 0.4399, 0.8386, 0.3634, 0.4016, 0.2942, 0.5853, 0.5336]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6197, 0.4118, 0.8687, 0.5517, 0.4038, 0.5233, 0.5875, 0.5600], [0.6168, 0.4029, 0.8523, 0.3417, 0.3587, 0.5000, 0.6125, 0.5400], [0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6263, 0.5367], [0.6261, 0.4066, 0.8325, 0.2150, 0.4762, 0.2667, 0.7002, 0.5633], [0.0000, 0.0000, 0.7425, 0.2117, 0.3938, 0.2433, 0.5437, 0.5567], [0.6222, 0.4171, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.10814244553330354 step: 70 running loss: 0.0015448920790471935 Train Steps: 70/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999], [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650], [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134], [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869], [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402], [0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5551, 0.3648, 0.8834, 0.4773, 0.3928, 0.5299, 0.5418, 0.5109], [0.5776, 0.3809, 0.8719, 0.5290, 0.4246, 0.5201, 0.5414, 0.5491], [0.5532, 0.3692, 0.8817, 0.5963, 0.4139, 0.5325, 0.6002, 0.5089], [0.5847, 0.3793, 0.8621, 0.2714, 0.5321, 0.2797, 0.7289, 0.5481], [0.5375, 0.3378, 0.8474, 0.2894, 0.3923, 0.2840, 0.6445, 0.4855], [0.5882, 0.3816, 0.8420, 0.2702, 0.4611, 0.2218, 0.6174, 0.4954], [0.5419, 0.3502, 0.8819, 0.5863, 0.3925, 0.4763, 0.6229, 0.5348], [0.5298, 0.3446, 0.7433, 0.2817, 0.4003, 0.2351, 0.5706, 0.5667]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6068, 0.3963, 0.8650, 0.4317, 0.4038, 0.5083, 0.5253, 0.4999], [0.6098, 0.3991, 0.8637, 0.4717, 0.4263, 0.4967, 0.5213, 0.5650], [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134], [0.6300, 0.4133, 0.8537, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.6257, 0.4034, 0.8288, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6276, 0.4095, 0.8238, 0.2250, 0.4663, 0.1783, 0.6171, 0.4869], [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402], [0.6193, 0.4050, 0.7312, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.10943879742990248 step: 71 running loss: 0.001541391513097218 Train Steps: 71/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567], [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426], [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.1746, 0.1223, 0.7293, 0.2208, 0.4302, 0.2071, 0.5374, 0.5330], [0.6888, 0.4421, 0.9003, 0.5702, 0.3797, 0.4544, 0.6007, 0.5144], [0.6518, 0.4239, 0.9269, 0.5022, 0.4492, 0.5962, 0.6580, 0.5112], [0.6560, 0.4232, 0.8784, 0.5940, 0.3866, 0.4556, 0.6057, 0.5096], [0.0818, 0.0514, 0.7669, 0.2419, 0.4034, 0.2672, 0.5743, 0.5333], [0.6324, 0.4228, 0.7939, 0.2588, 0.4446, 0.2049, 0.5882, 0.5233], [0.7643, 0.4976, 0.8812, 0.6012, 0.3931, 0.5022, 0.5749, 0.5795], [0.5979, 0.3965, 0.9247, 0.4614, 0.4477, 0.5889, 0.6334, 0.5222]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [0.0000, 0.0000, 0.7425, 0.2117, 0.3938, 0.2433, 0.5437, 0.5567], [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426], [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.11130989887169562 step: 72 running loss: 0.0015459708176624393 Train Steps: 72/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117], [0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585], [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123], [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850], [0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799], [0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5244, 0.3230, 0.7322, 0.2232, 0.3741, 0.2648, 0.5930, 0.5196], [0.5423, 0.3285, 0.7587, 0.2411, 0.4187, 0.2186, 0.5792, 0.4869], [0.5157, 0.3403, 0.8917, 0.4692, 0.4561, 0.5478, 0.5834, 0.5112], [0.5739, 0.3679, 0.8928, 0.5662, 0.3678, 0.4318, 0.6299, 0.5162], [0.5439, 0.3567, 0.7502, 0.2889, 0.3702, 0.2684, 0.5522, 0.4780], [0.5751, 0.3509, 0.9098, 0.4848, 0.4810, 0.5481, 0.6122, 0.5371], [0.5244, 0.3461, 0.8728, 0.4333, 0.3907, 0.3227, 0.5516, 0.5536], [0.5542, 0.3530, 0.8331, 0.3390, 0.3716, 0.4300, 0.6096, 0.5287]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6114, 0.4018, 0.7212, 0.1967, 0.3762, 0.2700, 0.5875, 0.5533], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5788, 0.5117], [0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585], [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123], [0.6184, 0.4079, 0.8838, 0.4617, 0.4837, 0.5650, 0.6175, 0.5850], [0.6250, 0.4236, 0.8637, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799], [0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.11289354818291031 step: 73 running loss: 0.0015464869614097302 Train Steps: 73/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6099, 0.4030, 0.8638, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388], [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879], [0.6198, 0.4101, 0.8838, 0.5283, 0.3763, 0.5267, 0.5913, 0.5567], [0.6325, 0.4165, 0.9000, 0.4617, 0.3813, 0.4900, 0.7485, 0.5447], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263], [0.6275, 0.4048, 0.8488, 0.2883, 0.4463, 0.2033, 0.6321, 0.5155]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5338, 0.3522, 0.8780, 0.5271, 0.4792, 0.5190, 0.4882, 0.5306], [0.5412, 0.3423, 0.8516, 0.2336, 0.5212, 0.2795, 0.6972, 0.5319], [0.5630, 0.3671, 0.8744, 0.4487, 0.3698, 0.5784, 0.5697, 0.4878], [0.5473, 0.3473, 0.8602, 0.5516, 0.3654, 0.5584, 0.5371, 0.5472], [0.5457, 0.3635, 0.9200, 0.4802, 0.3689, 0.5329, 0.6613, 0.5378], [0.5589, 0.3510, 0.7413, 0.3505, 0.4733, 0.2244, 0.5237, 0.6045], [0.5656, 0.3641, 0.7410, 0.2551, 0.4121, 0.2023, 0.5469, 0.5126], [0.5246, 0.3155, 0.8481, 0.2755, 0.4280, 0.2449, 0.6312, 0.5021]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6098, 0.4030, 0.8637, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388], [0.6300, 0.4133, 0.8537, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879], [0.6198, 0.4101, 0.8838, 0.5283, 0.3762, 0.5267, 0.5913, 0.5567], [0.6325, 0.4165, 0.9000, 0.4617, 0.3812, 0.4900, 0.7485, 0.5447], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263], [0.6275, 0.4048, 0.8487, 0.2883, 0.4462, 0.2033, 0.6321, 0.5155]]], device='cuda:0') loss_train_step before backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train: 0.11465108217089437 step: 74 running loss: 0.0015493389482553292 Train Steps: 74/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019], [0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433], [0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6238, 0.5350], [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6200, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5727, 0.3657, 0.8181, 0.2616, 0.4693, 0.2311, 0.5641, 0.5257], [0.6030, 0.3700, 0.8953, 0.4802, 0.3780, 0.4110, 0.6093, 0.5120], [0.5460, 0.3730, 0.8862, 0.4525, 0.4609, 0.5342, 0.5684, 0.5456], [0.6147, 0.4029, 0.8721, 0.3928, 0.3868, 0.5654, 0.5933, 0.5465], [0.5614, 0.3649, 0.8514, 0.5074, 0.3873, 0.5299, 0.6509, 0.5383], [0.6059, 0.3993, 0.8745, 0.4442, 0.3633, 0.3604, 0.5813, 0.5028], [0.5465, 0.3564, 0.8698, 0.4750, 0.3883, 0.4464, 0.5409, 0.5462], [0.5784, 0.3736, 0.7716, 0.2147, 0.4488, 0.2392, 0.6080, 0.5363]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6275, 0.4111, 0.8462, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019], [0.6250, 0.4054, 0.8770, 0.4723, 0.4663, 0.5367, 0.6162, 0.5433], [0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6237, 0.5350], [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6201, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.1157409083971288 step: 75 running loss: 0.0015432121119617174 Train Steps: 75/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700], [ nan, nan, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220], [0.6166, 0.4008, 0.8563, 0.5667, 0.4388, 0.4933, 0.5575, 0.5567], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.6198, 0.4130, 0.8762, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810], [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.7047, 0.4521, 0.8542, 0.4457, 0.3852, 0.4534, 0.5695, 0.5806], [-0.0310, -0.0268, 0.7352, 0.2170, 0.3854, 0.2452, 0.5475, 0.5456], [ 0.6074, 0.3916, 0.8318, 0.5185, 0.4381, 0.4620, 0.5896, 0.5820], [ 0.6495, 0.4064, 0.7906, 0.2347, 0.4436, 0.2241, 0.6480, 0.5400], [ 0.6545, 0.4307, 0.8227, 0.5311, 0.4055, 0.4264, 0.5485, 0.5173], [ 0.6268, 0.4137, 0.8465, 0.3939, 0.3636, 0.4709, 0.5830, 0.5404], [ 0.6327, 0.4152, 0.8263, 0.2706, 0.4183, 0.2436, 0.6445, 0.5048], [ 0.6758, 0.4513, 0.8386, 0.3769, 0.4227, 0.5609, 0.6045, 0.5502]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6111, 0.3995, 0.8788, 0.4567, 0.3812, 0.4833, 0.5450, 0.5700], [0.0000, 0.0000, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220], [0.6166, 0.4008, 0.8562, 0.5667, 0.4387, 0.4933, 0.5575, 0.5567], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.6198, 0.4130, 0.8763, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810], [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.11660776424105279 step: 76 running loss: 0.0015343126873822736 Train Steps: 76/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767], [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540], [0.6361, 0.4076, 0.8862, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297], [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118], [0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773], [0.6214, 0.4112, 0.7838, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100], [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5444, 0.3557, 0.8830, 0.4480, 0.4301, 0.5141, 0.5388, 0.5729], [0.6076, 0.3915, 0.9206, 0.4102, 0.4174, 0.3018, 0.6659, 0.5342], [0.6002, 0.3981, 0.8804, 0.5211, 0.3667, 0.4781, 0.6635, 0.5433], [0.6297, 0.4335, 0.7276, 0.2955, 0.4258, 0.2230, 0.5619, 0.6147], [0.5596, 0.3659, 0.6780, 0.1973, 0.4080, 0.2274, 0.5393, 0.5786], [0.5413, 0.3425, 0.7714, 0.1700, 0.3657, 0.2916, 0.5894, 0.5097], [0.6796, 0.4358, 0.8973, 0.3880, 0.3720, 0.3797, 0.6493, 0.5082], [0.6840, 0.4333, 0.8701, 0.5319, 0.3848, 0.3931, 0.6216, 0.4820]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6148, 0.4076, 0.8666, 0.4820, 0.4137, 0.5067, 0.5250, 0.5767], [0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540], [0.6361, 0.4076, 0.8863, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297], [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118], [0.6147, 0.4026, 0.6600, 0.2467, 0.4087, 0.2150, 0.5489, 0.5773], [0.6214, 0.4112, 0.7837, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100], [0.6272, 0.4071, 0.8737, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.11740422362345271 step: 77 running loss: 0.0015247301769279573 Train Steps: 77/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748], [0.6200, 0.3993, 0.8519, 0.4923, 0.3962, 0.4717, 0.6013, 0.5433], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167], [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852], [ nan, nan, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999], [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097], [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5963, 0.6217]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6762, 0.4645, 0.8378, 0.5068, 0.3814, 0.3885, 0.6824, 0.5250], [0.6535, 0.4385, 0.8755, 0.4775, 0.3767, 0.4595, 0.6373, 0.5371], [0.6861, 0.4507, 0.8526, 0.5125, 0.3286, 0.3462, 0.5766, 0.5065], [0.6160, 0.4223, 0.7629, 0.2434, 0.3948, 0.2627, 0.6210, 0.6043], [0.6554, 0.4457, 0.8855, 0.4622, 0.4515, 0.4659, 0.5474, 0.5615], [0.1650, 0.1163, 0.7443, 0.1991, 0.4005, 0.1677, 0.5614, 0.4940], [0.6359, 0.4271, 0.8692, 0.4393, 0.4413, 0.5079, 0.5420, 0.4931], [0.7311, 0.4831, 0.6914, 0.1935, 0.4226, 0.2248, 0.6182, 0.5892]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748], [0.6200, 0.3993, 0.8519, 0.4923, 0.3963, 0.4717, 0.6012, 0.5433], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167], [0.6140, 0.4070, 0.8700, 0.5000, 0.4613, 0.4900, 0.5260, 0.5852], [0.0000, 0.0000, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999], [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097], [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5962, 0.6217]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.11927059848676436 step: 78 running loss: 0.0015291102370097993 Train Steps: 78/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927], [0.6087, 0.3951, 0.8387, 0.5833, 0.4188, 0.4933, 0.5146, 0.4830], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717], [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367], [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461], [0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960], [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413], [0.6127, 0.4115, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6987, 0.4584, 0.8383, 0.5464, 0.3512, 0.4217, 0.6277, 0.4938], [0.5863, 0.3896, 0.8132, 0.5523, 0.3801, 0.4898, 0.5403, 0.5217], [0.5919, 0.4097, 0.8451, 0.4679, 0.4123, 0.4491, 0.5641, 0.5819], [0.6159, 0.4131, 0.8378, 0.2951, 0.3678, 0.2500, 0.6546, 0.5304], [0.5996, 0.4130, 0.8316, 0.5114, 0.4123, 0.5076, 0.5484, 0.5542], [0.5672, 0.3895, 0.8641, 0.2796, 0.4345, 0.1819, 0.6205, 0.5191], [0.6335, 0.4170, 0.9235, 0.3661, 0.4421, 0.2307, 0.7317, 0.5447], [0.5672, 0.3924, 0.7009, 0.2516, 0.3476, 0.2907, 0.5554, 0.5630]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6321, 0.4048, 0.8737, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927], [0.6087, 0.3951, 0.8388, 0.5833, 0.4187, 0.4933, 0.5146, 0.4830], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5412, 0.5717], [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367], [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461], [0.6259, 0.4156, 0.8813, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960], [0.6271, 0.4040, 0.9137, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413], [0.6127, 0.4114, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.11994522050372325 step: 79 running loss: 0.0015182939304268766 Train Steps: 79/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6167, 0.4048, 0.6831, 0.3639, 0.3763, 0.3017, 0.5700, 0.5883], [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383], [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100], [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650], [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.1057, 0.0799, 0.6776, 0.2266, 0.4075, 0.1826, 0.5185, 0.5593], [0.5822, 0.4002, 0.7446, 0.3650, 0.3553, 0.3069, 0.5613, 0.5871], [0.7067, 0.4896, 0.7654, 0.2641, 0.4527, 0.1474, 0.6178, 0.5330], [0.6936, 0.4618, 0.8660, 0.5623, 0.4117, 0.5137, 0.6381, 0.5277], [0.6334, 0.4676, 0.8516, 0.4369, 0.3563, 0.3323, 0.5630, 0.5715], [0.7058, 0.4797, 0.8861, 0.5675, 0.3770, 0.4973, 0.6390, 0.5419], [0.6569, 0.4481, 0.7611, 0.2014, 0.4641, 0.1767, 0.6098, 0.5115], [0.6362, 0.4427, 0.8369, 0.2680, 0.4433, 0.2233, 0.7059, 0.5670]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6167, 0.4048, 0.6831, 0.3639, 0.3762, 0.3017, 0.5700, 0.5883], [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383], [0.6299, 0.4008, 0.8450, 0.5350, 0.4212, 0.5000, 0.6350, 0.5100], [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650], [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.121143347438192 step: 80 running loss: 0.0015142918429774 Train Steps: 80/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700], [0.6198, 0.4130, 0.8762, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103], [0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272], [0.6193, 0.4108, 0.7438, 0.2700, 0.3650, 0.3683, 0.6238, 0.5717], [0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550], [0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5598, 0.3661, 0.7617, 0.2352, 0.4312, 0.1759, 0.5951, 0.5085], [0.6472, 0.4201, 0.8735, 0.5111, 0.3916, 0.4521, 0.5697, 0.5639], [0.6271, 0.4169, 0.8698, 0.4614, 0.3704, 0.4657, 0.5780, 0.5201], [0.5978, 0.3887, 0.8416, 0.4152, 0.4066, 0.2550, 0.5161, 0.5368], [0.6568, 0.4365, 0.7498, 0.3367, 0.3714, 0.3466, 0.6398, 0.5593], [0.6814, 0.4443, 0.8768, 0.5257, 0.3883, 0.4427, 0.6180, 0.5738], [0.5511, 0.3807, 0.8901, 0.5058, 0.4081, 0.5096, 0.5988, 0.5190], [0.7312, 0.4653, 0.8871, 0.5383, 0.4363, 0.5173, 0.6679, 0.5432]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6202, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6111, 0.3995, 0.8788, 0.4567, 0.3812, 0.4833, 0.5450, 0.5700], [0.6198, 0.4130, 0.8763, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103], [0.6186, 0.4080, 0.8625, 0.3483, 0.3787, 0.2650, 0.5320, 0.5272], [0.6193, 0.4108, 0.7437, 0.2700, 0.3650, 0.3683, 0.6237, 0.5717], [0.6197, 0.4091, 0.8800, 0.4783, 0.3537, 0.4767, 0.5950, 0.5550], [0.6186, 0.4129, 0.8900, 0.4567, 0.3938, 0.5417, 0.5734, 0.5110], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.12236118371947668 step: 81 running loss: 0.001510631897771317 Train Steps: 81/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6145, 0.4007, 0.8775, 0.4533, 0.4562, 0.5533, 0.6088, 0.5533], [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100], [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267], [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250], [0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467], [0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297], [0.6198, 0.4101, 0.8838, 0.5283, 0.3763, 0.5267, 0.5913, 0.5567]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6552, 0.4201, 0.8652, 0.4655, 0.4552, 0.5254, 0.5912, 0.5429], [0.6301, 0.3977, 0.8483, 0.5500, 0.4006, 0.4668, 0.6007, 0.4875], [0.6143, 0.4100, 0.8121, 0.5916, 0.3956, 0.4262, 0.5624, 0.5386], [0.6091, 0.4006, 0.8325, 0.2715, 0.4438, 0.2067, 0.6716, 0.5364], [0.6086, 0.3938, 0.8833, 0.4870, 0.4127, 0.5155, 0.5829, 0.5079], [0.5988, 0.3928, 0.8847, 0.5002, 0.4260, 0.4873, 0.5860, 0.5524], [0.6430, 0.4311, 0.9103, 0.4571, 0.4090, 0.2978, 0.6228, 0.5287], [0.6248, 0.3931, 0.8567, 0.5376, 0.3810, 0.5050, 0.5612, 0.5507]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6145, 0.4007, 0.8775, 0.4533, 0.4563, 0.5533, 0.6087, 0.5533], [0.6275, 0.4024, 0.8500, 0.5383, 0.3913, 0.4883, 0.6288, 0.5100], [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267], [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250], [0.6199, 0.4059, 0.8700, 0.4900, 0.4162, 0.5000, 0.6162, 0.5467], [0.6364, 0.4165, 0.9087, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297], [0.6198, 0.4101, 0.8838, 0.5283, 0.3762, 0.5267, 0.5913, 0.5567]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.12267718787188642 step: 82 running loss: 0.0014960632667303222 Train Steps: 82/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136], [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954], [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5914, 0.3665, 0.7146, 0.2745, 0.4171, 0.2598, 0.5285, 0.5583], [0.6297, 0.3911, 0.7009, 0.2508, 0.3974, 0.2863, 0.5636, 0.5424], [0.5483, 0.3389, 0.8805, 0.5764, 0.4454, 0.5368, 0.5776, 0.5156], [0.6216, 0.3915, 0.8177, 0.3364, 0.3854, 0.4146, 0.5627, 0.5295], [0.6665, 0.4182, 0.8659, 0.3129, 0.5133, 0.2287, 0.6636, 0.5608], [0.5599, 0.3312, 0.8511, 0.6297, 0.4080, 0.5062, 0.5952, 0.4846], [0.6774, 0.4302, 0.8439, 0.4182, 0.3562, 0.4139, 0.5344, 0.5245], [0.6530, 0.4147, 0.8346, 0.2688, 0.4997, 0.1857, 0.6275, 0.5368]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6043, 0.4022, 0.6888, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136], [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6076, 0.3952, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954], [0.6266, 0.4065, 0.8313, 0.2467, 0.4787, 0.1733, 0.6313, 0.5133]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.12356227377313189 step: 83 running loss: 0.0014887020936521914 Train Steps: 83/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6104, 0.4029, 0.8738, 0.4900, 0.4088, 0.4533, 0.5070, 0.5510], [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917], [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850], [0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517], [0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843], [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6121, 0.3956, 0.8771, 0.5035, 0.4204, 0.4715, 0.4994, 0.5032], [0.6205, 0.3745, 0.9002, 0.5052, 0.3913, 0.5386, 0.6705, 0.4973], [0.6060, 0.3819, 0.7070, 0.3297, 0.3670, 0.3024, 0.5258, 0.5516], [0.6212, 0.4016, 0.8674, 0.5632, 0.4227, 0.4285, 0.5475, 0.5770], [0.6462, 0.4000, 0.9094, 0.4689, 0.4834, 0.5711, 0.6178, 0.5542], [0.5842, 0.3633, 0.8862, 0.4659, 0.4760, 0.5078, 0.5525, 0.5156], [0.6476, 0.3985, 0.9158, 0.4879, 0.3884, 0.5060, 0.6152, 0.4727], [0.6824, 0.4276, 0.7175, 0.2365, 0.3930, 0.2785, 0.5990, 0.5228]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6104, 0.4029, 0.8737, 0.4900, 0.4087, 0.4533, 0.5070, 0.5510], [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6212, 0.4159, 0.8675, 0.5783, 0.4087, 0.4317, 0.5612, 0.5917], [0.6184, 0.4079, 0.8838, 0.4617, 0.4837, 0.5650, 0.6175, 0.5850], [0.6175, 0.3957, 0.8700, 0.4817, 0.4663, 0.5133, 0.5800, 0.5517], [0.6264, 0.4049, 0.8988, 0.4633, 0.3812, 0.4983, 0.6326, 0.4843], [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6087, 0.5583]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.12402151894639246 step: 84 running loss: 0.0014764466541237198 Train Steps: 84/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433], [0.6109, 0.4015, 0.7668, 0.3639, 0.3513, 0.3667, 0.5200, 0.5641], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650], [0.6277, 0.4083, 0.8350, 0.2717, 0.4562, 0.1800, 0.5918, 0.4878], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750], [0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5706, 0.3611, 0.8415, 0.3558, 0.3894, 0.3482, 0.5369, 0.5443], [0.6374, 0.4090, 0.7719, 0.3737, 0.3650, 0.3895, 0.5323, 0.5581], [0.5539, 0.3452, 0.8682, 0.3352, 0.3994, 0.3267, 0.5938, 0.5154], [0.6228, 0.4052, 0.7520, 0.2746, 0.4052, 0.2969, 0.5474, 0.5626], [0.5389, 0.3360, 0.8147, 0.2744, 0.4840, 0.2404, 0.5875, 0.4859], [0.5996, 0.3857, 0.8043, 0.3106, 0.3818, 0.4226, 0.5740, 0.5148], [0.6152, 0.3860, 0.7280, 0.2786, 0.4149, 0.3206, 0.6044, 0.5611], [0.6196, 0.3619, 0.8757, 0.3720, 0.4327, 0.2936, 0.6399, 0.4696]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5412, 0.5433], [0.6109, 0.4015, 0.7668, 0.3639, 0.3512, 0.3667, 0.5200, 0.5641], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650], [0.6277, 0.4083, 0.8350, 0.2717, 0.4563, 0.1800, 0.5918, 0.4878], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750], [0.6277, 0.4036, 0.8687, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.1248362893529702 step: 85 running loss: 0.0014686622276820023 Train Steps: 85/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6233, 0.4091, 0.8100, 0.2950, 0.3563, 0.3883, 0.6013, 0.5200], [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207], [0.6126, 0.4067, 0.8638, 0.5383, 0.4188, 0.4850, 0.5016, 0.5392], [0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6439, 0.4169, 0.7913, 0.2965, 0.3638, 0.4242, 0.6197, 0.5277], [0.6116, 0.3866, 0.8863, 0.4314, 0.4029, 0.4827, 0.5048, 0.5232], [0.6431, 0.4087, 0.8547, 0.3027, 0.3808, 0.3170, 0.5720, 0.5248], [0.6157, 0.3800, 0.8009, 0.2299, 0.4458, 0.2494, 0.6712, 0.5019], [0.6382, 0.4040, 0.8611, 0.5433, 0.4226, 0.5305, 0.5289, 0.5456], [0.6433, 0.3967, 0.7910, 0.2740, 0.4048, 0.3044, 0.6042, 0.5222], [0.6362, 0.4055, 0.8299, 0.3050, 0.3622, 0.4071, 0.6183, 0.5390], [0.6083, 0.3849, 0.6878, 0.2513, 0.4030, 0.2424, 0.5473, 0.5537]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6233, 0.4091, 0.8100, 0.2950, 0.3562, 0.3883, 0.6012, 0.5200], [0.6185, 0.4067, 0.8838, 0.4450, 0.4038, 0.4733, 0.5213, 0.5142], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207], [0.6126, 0.4067, 0.8637, 0.5383, 0.4187, 0.4850, 0.5016, 0.5392], [0.6201, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5362, 0.5550]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.12514446672867052 step: 86 running loss: 0.0014551682177752386 Train Steps: 86/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329], [0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148], [0.6087, 0.3976, 0.8337, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5966, 0.3957, 0.8675, 0.5086, 0.3638, 0.5120, 0.6810, 0.5625], [0.7951, 0.5214, 0.8327, 0.3075, 0.3375, 0.3722, 0.5612, 0.5145], [0.6586, 0.4217, 0.8096, 0.3004, 0.3758, 0.3178, 0.5457, 0.5269], [0.6461, 0.4297, 0.8695, 0.5199, 0.3957, 0.4628, 0.5302, 0.5536], [0.6294, 0.4333, 0.8799, 0.4842, 0.3766, 0.5092, 0.5607, 0.4978], [0.6841, 0.4662, 0.8497, 0.3571, 0.3594, 0.3482, 0.5710, 0.5243], [0.6373, 0.4358, 0.8713, 0.2233, 0.4394, 0.2694, 0.6967, 0.5514], [0.2011, 0.1628, 0.6991, 0.1745, 0.4166, 0.2143, 0.5079, 0.5705]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329], [0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148], [0.6087, 0.3976, 0.8338, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.0000, 0.0000, 0.6413, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700]]], device='cuda:0') loss_train_step before backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train: 0.12772876137751155 step: 87 running loss: 0.0014681466825001326 Train Steps: 87/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864], [0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901], [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6357, 0.4118, 0.8400, 0.2500, 0.5413, 0.1633, 0.6725, 0.5586], [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5803, 0.3842, 0.8580, 0.5116, 0.3559, 0.4980, 0.6933, 0.5874], [0.6300, 0.4038, 0.8794, 0.4717, 0.3554, 0.5003, 0.7029, 0.5902], [0.6064, 0.4055, 0.7442, 0.2507, 0.3430, 0.2914, 0.5109, 0.5141], [0.5785, 0.3940, 0.8552, 0.3603, 0.3619, 0.5005, 0.5622, 0.5352], [0.5799, 0.3884, 0.7843, 0.2502, 0.3571, 0.2743, 0.5150, 0.5242], [0.5797, 0.3719, 0.8946, 0.4518, 0.3488, 0.4259, 0.5736, 0.5385], [0.5465, 0.3934, 0.8544, 0.2093, 0.5169, 0.1659, 0.6492, 0.5658], [0.6566, 0.4615, 0.8477, 0.3786, 0.3487, 0.3242, 0.5568, 0.5618]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6314, 0.4107, 0.8750, 0.5100, 0.3787, 0.4900, 0.7121, 0.5864], [0.6079, 0.3964, 0.7420, 0.2958, 0.3562, 0.2917, 0.5351, 0.4980], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901], [0.6236, 0.3965, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6357, 0.4118, 0.8400, 0.2500, 0.5412, 0.1633, 0.6725, 0.5586], [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.12843954053823836 step: 88 running loss: 0.0014595402333890722 Train Steps: 88/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117], [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823], [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019], [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767], [0.6131, 0.4037, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6554, 0.4362, 0.8934, 0.4392, 0.3842, 0.4940, 0.5825, 0.5158], [0.6204, 0.4117, 0.7705, 0.1987, 0.3993, 0.1871, 0.6053, 0.5255], [0.5925, 0.3959, 0.7826, 0.3638, 0.3346, 0.3605, 0.5689, 0.5980], [0.5463, 0.3805, 0.7074, 0.2054, 0.4121, 0.1620, 0.5610, 0.5656], [0.6076, 0.4064, 0.8212, 0.2460, 0.3942, 0.2600, 0.6509, 0.5222], [0.6327, 0.4301, 0.8894, 0.4261, 0.3661, 0.5427, 0.6239, 0.5256], [0.5979, 0.3990, 0.9151, 0.5030, 0.3897, 0.4423, 0.5752, 0.5977], [0.6053, 0.3951, 0.7183, 0.2489, 0.3607, 0.2643, 0.5485, 0.5817]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6068, 0.3963, 0.8650, 0.4317, 0.4038, 0.5083, 0.5253, 0.4999], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5788, 0.5117], [0.6170, 0.4102, 0.7468, 0.3695, 0.3462, 0.3767, 0.5238, 0.5823], [0.6154, 0.4112, 0.7038, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6161, 0.4099, 0.8737, 0.4383, 0.3787, 0.5483, 0.5605, 0.5019], [0.6163, 0.4001, 0.8788, 0.5033, 0.4013, 0.4633, 0.5337, 0.5767], [0.6131, 0.4036, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.12903116384404711 step: 89 running loss: 0.0014497883577982821 Train Steps: 89/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767], [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367], [0.6216, 0.4099, 0.7225, 0.2033, 0.4188, 0.2217, 0.5975, 0.5283], [0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617], [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6138, 0.5883], [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833], [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350], [0.6198, 0.4115, 0.7762, 0.2717, 0.3713, 0.3200, 0.5837, 0.5683]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6364, 0.4298, 0.8848, 0.4947, 0.3976, 0.4907, 0.5543, 0.5624], [0.5528, 0.3813, 0.7996, 0.2573, 0.3886, 0.2307, 0.6113, 0.5471], [0.6042, 0.4352, 0.7250, 0.2135, 0.3953, 0.1917, 0.5677, 0.5342], [0.6216, 0.3989, 0.8161, 0.3435, 0.3605, 0.2686, 0.5757, 0.5322], [0.5356, 0.3706, 0.9046, 0.4557, 0.4116, 0.5370, 0.6324, 0.5777], [0.4950, 0.3409, 0.7071, 0.2959, 0.3453, 0.2621, 0.5593, 0.5702], [0.6456, 0.4085, 0.8733, 0.4935, 0.3967, 0.4418, 0.5593, 0.5491], [0.6524, 0.4392, 0.7946, 0.2740, 0.3508, 0.2938, 0.5992, 0.5466]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6148, 0.4076, 0.8666, 0.4820, 0.4137, 0.5067, 0.5250, 0.5767], [0.6204, 0.4110, 0.7912, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367], [0.6216, 0.4099, 0.7225, 0.2033, 0.4187, 0.2217, 0.5975, 0.5283], [0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617], [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6137, 0.5883], [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833], [0.6126, 0.3954, 0.8537, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350], [0.6198, 0.4115, 0.7763, 0.2717, 0.3713, 0.3200, 0.5838, 0.5683]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.12994244889705442 step: 90 running loss: 0.001443804987745049 Valid Steps: 10/10 Loss: nan 3.3351 -------------------------------------------------- Epoch: 5 Train Loss: 0.0014 Valid Loss: nan -------------------------------------------------- size of train loader is: 90 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413], [ nan, nan, 0.8850, 0.3000, 0.5363, 0.2250, 0.7343, 0.5771], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6117, 0.4019, 0.8538, 0.4067, 0.3513, 0.3583, 0.5663, 0.5133], [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200], [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417], [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748], [0.6248, 0.4032, 0.7738, 0.1900, 0.4813, 0.1400, 0.5941, 0.4904]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6332, 0.4250, 0.8920, 0.3640, 0.4325, 0.2550, 0.6897, 0.5502], [-0.0614, -0.0035, 0.8266, 0.2532, 0.4840, 0.2003, 0.6934, 0.5828], [ 0.5656, 0.3930, 0.8296, 0.4740, 0.3736, 0.4827, 0.5273, 0.5308], [ 0.6408, 0.4271, 0.8211, 0.3995, 0.3206, 0.3348, 0.5158, 0.5318], [ 0.6423, 0.4379, 0.8163, 0.2955, 0.4129, 0.2113, 0.5810, 0.5201], [ 0.6365, 0.4531, 0.7563, 0.2654, 0.3638, 0.2782, 0.5941, 0.5576], [ 0.5648, 0.4057, 0.7927, 0.5257, 0.3614, 0.3913, 0.6561, 0.5536], [ 0.5543, 0.3796, 0.7082, 0.1927, 0.4567, 0.1175, 0.5567, 0.5030]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6271, 0.4040, 0.9137, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413], [0.0000, 0.0000, 0.8850, 0.3000, 0.5362, 0.2250, 0.7343, 0.5771], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6116, 0.4019, 0.8537, 0.4067, 0.3512, 0.3583, 0.5663, 0.5133], [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200], [0.6243, 0.4128, 0.7763, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417], [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748], [0.6248, 0.4032, 0.7738, 0.1900, 0.4812, 0.1400, 0.5941, 0.4904]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.0010950572323054075 step: 1 running loss: 0.0010950572323054075 Train Steps: 1/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240], [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413], [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567], [0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297], [0.6197, 0.4118, 0.8688, 0.5517, 0.4037, 0.5233, 0.5875, 0.5600], [0.6250, 0.4146, 0.8838, 0.3933, 0.3588, 0.4283, 0.6162, 0.5367], [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4767, 0.3139, 0.7132, 0.2130, 0.4388, 0.1563, 0.5310, 0.5519], [0.5141, 0.3293, 0.8923, 0.3159, 0.5079, 0.2378, 0.7557, 0.5437], [0.0024, 0.0341, 0.7160, 0.2549, 0.4101, 0.2055, 0.5066, 0.5869], [0.6908, 0.4591, 0.9057, 0.4473, 0.4126, 0.2974, 0.6531, 0.5544], [0.7181, 0.4650, 0.8476, 0.5692, 0.3929, 0.5315, 0.6112, 0.5648], [0.6992, 0.4510, 0.8456, 0.4030, 0.3658, 0.4038, 0.6252, 0.5446], [0.5817, 0.4095, 0.7255, 0.2901, 0.3826, 0.2522, 0.5198, 0.5925], [0.5966, 0.4142, 0.7107, 0.2639, 0.3920, 0.2604, 0.5655, 0.5287]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240], [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413], [0.0000, 0.0000, 0.7515, 0.2708, 0.3988, 0.2267, 0.5163, 0.5567], [0.6364, 0.4165, 0.9087, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297], [0.6197, 0.4118, 0.8687, 0.5517, 0.4038, 0.5233, 0.5875, 0.5600], [0.6250, 0.4146, 0.8838, 0.3933, 0.3587, 0.4283, 0.6162, 0.5367], [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5462, 0.5800], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.002500983187928796 step: 2 running loss: 0.001250491593964398 Train Steps: 2/90 Loss: 0.0013 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983], [0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267], [0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6307, 0.4029, 0.8650, 0.5200, 0.3763, 0.4017, 0.7311, 0.5366], [0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165], [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567], [0.6171, 0.4127, 0.8900, 0.4800, 0.4325, 0.5783, 0.5769, 0.5090], [0.6218, 0.4137, 0.7263, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5942, 0.3765, 0.8441, 0.5555, 0.3895, 0.3756, 0.5572, 0.5978], [0.6839, 0.4461, 0.8516, 0.2989, 0.4596, 0.2601, 0.6697, 0.5343], [0.5452, 0.3702, 0.6386, 0.2365, 0.4353, 0.1893, 0.5224, 0.5777], [0.6697, 0.4203, 0.8474, 0.5115, 0.4019, 0.3696, 0.7155, 0.5159], [0.6741, 0.4410, 0.8573, 0.3529, 0.3951, 0.2907, 0.6152, 0.5146], [0.0102, 0.0326, 0.7223, 0.2657, 0.4187, 0.2061, 0.5159, 0.5756], [0.6716, 0.4447, 0.9013, 0.4962, 0.4554, 0.5630, 0.6017, 0.5144], [0.6185, 0.3953, 0.7324, 0.2696, 0.4223, 0.2574, 0.6101, 0.5704]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6260, 0.4214, 0.8537, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983], [0.6266, 0.4067, 0.8587, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267], [0.6149, 0.4054, 0.6712, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6307, 0.4029, 0.8650, 0.5200, 0.3762, 0.4017, 0.7311, 0.5366], [0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165], [0.0000, 0.0000, 0.7515, 0.2708, 0.3988, 0.2267, 0.5163, 0.5567], [0.6171, 0.4127, 0.8900, 0.4800, 0.4325, 0.5783, 0.5769, 0.5090], [0.6218, 0.4137, 0.7262, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.0031757151591591537 step: 3 running loss: 0.001058571719719718 Train Steps: 3/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033], [0.6040, 0.4002, 0.7338, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778], [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407], [0.6144, 0.4032, 0.8563, 0.3283, 0.3525, 0.4200, 0.5775, 0.5583], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343], [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413], [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5998, 0.3881, 0.8489, 0.5266, 0.4269, 0.5364, 0.5653, 0.5266], [0.6322, 0.4066, 0.7096, 0.2361, 0.4164, 0.1925, 0.5609, 0.5039], [0.6330, 0.4206, 0.8351, 0.4194, 0.3907, 0.3015, 0.5033, 0.5504], [0.6071, 0.3998, 0.8333, 0.3566, 0.3592, 0.4133, 0.5639, 0.5432], [0.6248, 0.3960, 0.8594, 0.5358, 0.4098, 0.5365, 0.6945, 0.5580], [0.5699, 0.3728, 0.7798, 0.5499, 0.4036, 0.4531, 0.6909, 0.5258], [0.5120, 0.3214, 0.8967, 0.3325, 0.5200, 0.2344, 0.7362, 0.5293], [0.6220, 0.4074, 0.8722, 0.4873, 0.4371, 0.5165, 0.6189, 0.5546]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033], [0.6040, 0.4002, 0.7337, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778], [0.6133, 0.4094, 0.8495, 0.4028, 0.3587, 0.3200, 0.5003, 0.5407], [0.6144, 0.4032, 0.8562, 0.3283, 0.3525, 0.4200, 0.5775, 0.5583], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6311, 0.3998, 0.7975, 0.5767, 0.3837, 0.4850, 0.7327, 0.5343], [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413], [0.6241, 0.4142, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.003960852511227131 step: 4 running loss: 0.0009902131278067827 Train Steps: 4/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6125, 0.4076, 0.8488, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505], [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5787, 0.5600], [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367], [0.6265, 0.4088, 0.8025, 0.1850, 0.4163, 0.2500, 0.6290, 0.4947], [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6783, 0.4287, 0.8345, 0.4148, 0.3889, 0.3670, 0.5106, 0.5487], [0.6346, 0.3948, 0.8567, 0.5327, 0.4835, 0.5207, 0.6150, 0.5423], [0.6474, 0.3986, 0.8573, 0.3465, 0.4161, 0.2739, 0.6702, 0.5225], [0.6479, 0.4147, 0.7972, 0.2488, 0.4486, 0.2405, 0.6617, 0.5058], [0.0196, 0.0039, 0.7029, 0.2429, 0.4289, 0.2278, 0.5220, 0.5362], [0.5678, 0.3465, 0.8812, 0.5202, 0.3915, 0.5057, 0.6697, 0.4872], [0.6323, 0.3935, 0.8804, 0.3828, 0.3995, 0.3069, 0.6250, 0.5382], [0.6513, 0.4062, 0.7634, 0.2697, 0.4707, 0.2048, 0.6481, 0.5327]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6125, 0.4076, 0.8487, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505], [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5788, 0.5600], [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367], [0.6265, 0.4088, 0.8025, 0.1850, 0.4162, 0.2500, 0.6290, 0.4947], [0.0000, 0.0000, 0.6900, 0.1917, 0.3938, 0.2367, 0.5240, 0.5246], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.00465098989661783 step: 5 running loss: 0.0009301979793235659 Train Steps: 5/90 Loss: 0.0009 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540], [0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019], [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6552, 0.4114, 0.8774, 0.4658, 0.3782, 0.4515, 0.6856, 0.5181], [0.6604, 0.4108, 0.9086, 0.4488, 0.4350, 0.3231, 0.7029, 0.5382], [0.5866, 0.3753, 0.8584, 0.4249, 0.4075, 0.5679, 0.5804, 0.5033], [0.6180, 0.3723, 0.8611, 0.5224, 0.4083, 0.5331, 0.5848, 0.5022], [0.6441, 0.4107, 0.8707, 0.4167, 0.3678, 0.4041, 0.6107, 0.5390], [0.5974, 0.3602, 0.8996, 0.5247, 0.3950, 0.3934, 0.6271, 0.4822], [0.6110, 0.3776, 0.8502, 0.4215, 0.3777, 0.3466, 0.5408, 0.5714], [0.5673, 0.3445, 0.8787, 0.4207, 0.3819, 0.3933, 0.6621, 0.5070]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540], [0.6161, 0.4099, 0.8737, 0.4383, 0.3787, 0.5483, 0.5605, 0.5019], [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6182, 0.4058, 0.8737, 0.4350, 0.3562, 0.3400, 0.5290, 0.5822], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.00511719353380613 step: 6 running loss: 0.0008528655889676884 Train Steps: 6/90 Loss: 0.0009 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100], [ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117], [0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [0.6293, 0.3982, 0.8700, 0.5300, 0.3763, 0.4717, 0.7050, 0.5297], [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6628, 0.4258, 0.8829, 0.5384, 0.3582, 0.5162, 0.6053, 0.5552], [0.1885, 0.1040, 0.7453, 0.2245, 0.3961, 0.2227, 0.5406, 0.5308], [0.6685, 0.4067, 0.7905, 0.2101, 0.4101, 0.2576, 0.5969, 0.4991], [0.6771, 0.4192, 0.8103, 0.2036, 0.4245, 0.2512, 0.6132, 0.4906], [0.6017, 0.3746, 0.8931, 0.4406, 0.4620, 0.3175, 0.5585, 0.5956], [0.6587, 0.4070, 0.8789, 0.5625, 0.3910, 0.4548, 0.6079, 0.5264], [0.6750, 0.4058, 0.8986, 0.5292, 0.3593, 0.5000, 0.6943, 0.5169], [0.6772, 0.4275, 0.8966, 0.5189, 0.4330, 0.5449, 0.5616, 0.5367]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100], [0.0000, 0.0000, 0.7268, 0.2333, 0.4125, 0.1933, 0.5113, 0.5383], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5788, 0.5117], [0.6202, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [0.6293, 0.3982, 0.8700, 0.5300, 0.3762, 0.4717, 0.7050, 0.5297], [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.006649552000453696 step: 7 running loss: 0.0009499360000648137 Train Steps: 7/90 Loss: 0.0009 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517], [0.6126, 0.4039, 0.8237, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138], [0.6115, 0.4005, 0.8838, 0.3867, 0.3763, 0.4700, 0.5800, 0.5550], [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690], [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791], [0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5887, 0.3806, 0.9167, 0.4414, 0.3467, 0.4701, 0.6545, 0.4994], [0.6237, 0.4254, 0.9156, 0.4351, 0.4059, 0.5172, 0.5768, 0.5177], [0.6543, 0.4100, 0.7849, 0.2122, 0.3987, 0.2520, 0.6252, 0.5278], [0.6230, 0.3921, 0.8515, 0.4160, 0.3300, 0.3534, 0.6035, 0.5913], [0.6065, 0.3897, 0.8989, 0.3883, 0.3638, 0.4858, 0.5655, 0.5183], [0.6534, 0.4116, 0.8756, 0.5226, 0.3947, 0.5489, 0.7233, 0.5426], [0.6262, 0.3883, 0.9243, 0.4744, 0.3783, 0.5022, 0.6143, 0.4519], [0.6220, 0.3931, 0.8964, 0.5730, 0.4454, 0.4293, 0.5510, 0.5697]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6200, 0.4071, 0.7337, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517], [0.6126, 0.4038, 0.8238, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138], [0.6115, 0.4005, 0.8838, 0.3867, 0.3762, 0.4700, 0.5800, 0.5550], [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690], [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791], [0.6248, 0.4185, 0.8500, 0.5767, 0.4462, 0.4550, 0.5612, 0.5917]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.007122047507436946 step: 8 running loss: 0.0008902559384296183 Train Steps: 8/90 Loss: 0.0009 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835], [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6362, 0.5200], [ nan, nan, 0.6512, 0.1717, 0.4100, 0.1983, 0.5253, 0.5240], [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833], [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767], [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6781, 0.4366, 0.8292, 0.2767, 0.3947, 0.3024, 0.6299, 0.5228], [0.6502, 0.4163, 0.9356, 0.3927, 0.4165, 0.2404, 0.6043, 0.5037], [0.6243, 0.4212, 0.8959, 0.3969, 0.3492, 0.5258, 0.6535, 0.5265], [0.0754, 0.0272, 0.7135, 0.2025, 0.3964, 0.2239, 0.5353, 0.5389], [0.7175, 0.4695, 0.9051, 0.5078, 0.3554, 0.4924, 0.5934, 0.5303], [0.6314, 0.4253, 0.9020, 0.5113, 0.4033, 0.4586, 0.5315, 0.6031], [0.6845, 0.4592, 0.8323, 0.3439, 0.3670, 0.4061, 0.5821, 0.6231], [0.6510, 0.4404, 0.9164, 0.4900, 0.4029, 0.5442, 0.6044, 0.5414]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6273, 0.4110, 0.8900, 0.3817, 0.4187, 0.2167, 0.5858, 0.4835], [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6363, 0.5200], [0.0000, 0.0000, 0.6513, 0.1717, 0.4100, 0.1983, 0.5253, 0.5240], [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6037, 0.4833], [0.6163, 0.4001, 0.8788, 0.5033, 0.4013, 0.4633, 0.5337, 0.5767], [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6037, 0.6167], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.008067115588346496 step: 9 running loss: 0.000896346176482944 Train Steps: 9/90 Loss: 0.0009 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6353, 0.4128, 0.9138, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991], [ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617], [0.6275, 0.4157, 0.8337, 0.5800, 0.3763, 0.4200, 0.5547, 0.6125], [0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617], [0.6265, 0.4088, 0.8025, 0.1850, 0.4163, 0.2500, 0.6290, 0.4947], [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6273, 0.4100, 0.7137, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6229, 0.4118, 0.8863, 0.2859, 0.4184, 0.2557, 0.5658, 0.5340], [0.6148, 0.4197, 0.9315, 0.3354, 0.4334, 0.3215, 0.6852, 0.5903], [0.0754, 0.0627, 0.7600, 0.2218, 0.3954, 0.2259, 0.5242, 0.5600], [0.5883, 0.4249, 0.8646, 0.5609, 0.3616, 0.4497, 0.5613, 0.6069], [0.5997, 0.4042, 0.8556, 0.5515, 0.4356, 0.4546, 0.4996, 0.5693], [0.6244, 0.4190, 0.8221, 0.1943, 0.4030, 0.2628, 0.5984, 0.5144], [0.6360, 0.4487, 0.9374, 0.4746, 0.3753, 0.5383, 0.7273, 0.5499], [0.5903, 0.3867, 0.7336, 0.2222, 0.3727, 0.2676, 0.5683, 0.5639]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6250, 0.4131, 0.8687, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6353, 0.4128, 0.9137, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991], [0.0000, 0.0000, 0.7513, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617], [0.6275, 0.4157, 0.8338, 0.5800, 0.3762, 0.4200, 0.5547, 0.6125], [0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617], [0.6265, 0.4088, 0.8025, 0.1850, 0.4162, 0.2500, 0.6290, 0.4947], [0.6339, 0.4102, 0.9087, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6273, 0.4099, 0.7138, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.008671639865497127 step: 10 running loss: 0.0008671639865497127 Train Steps: 10/90 Loss: 0.0009 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528], [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103], [0.6289, 0.4081, 0.8720, 0.3487, 0.3900, 0.3183, 0.6703, 0.5376], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285], [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822], [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5651, 0.3872, 0.9091, 0.3589, 0.4060, 0.3687, 0.7129, 0.5629], [0.5442, 0.3813, 0.8742, 0.4722, 0.4190, 0.4401, 0.5225, 0.5954], [0.5596, 0.3881, 0.8712, 0.4179, 0.3868, 0.4584, 0.5164, 0.5327], [0.5353, 0.3653, 0.8768, 0.3212, 0.3934, 0.3146, 0.6572, 0.5564], [0.5771, 0.3766, 0.8791, 0.4666, 0.4206, 0.5151, 0.6954, 0.5690], [0.5786, 0.3886, 0.7983, 0.4074, 0.3795, 0.4221, 0.5060, 0.5502], [0.5158, 0.3602, 0.8586, 0.4000, 0.3567, 0.3225, 0.5129, 0.5861], [0.6043, 0.4215, 0.8886, 0.4290, 0.4082, 0.4468, 0.5101, 0.5186]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6332, 0.4165, 0.9100, 0.3350, 0.4187, 0.3683, 0.7438, 0.5528], [0.6163, 0.4001, 0.8788, 0.5033, 0.4013, 0.4633, 0.5337, 0.5767], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103], [0.6289, 0.4081, 0.8720, 0.3487, 0.3900, 0.3183, 0.6703, 0.5376], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285], [0.6182, 0.4058, 0.8737, 0.4350, 0.3562, 0.3400, 0.5290, 0.5822], [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.00954322874895297 step: 11 running loss: 0.0008675662499048154 Train Steps: 11/90 Loss: 0.0009 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6048, 0.3928, 0.8538, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719], [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417], [0.6361, 0.4076, 0.8862, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297], [0.6203, 0.4021, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [0.6273, 0.4100, 0.7137, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5483, 0.3622, 0.8770, 0.4779, 0.3990, 0.5093, 0.6017, 0.5437], [0.5784, 0.3904, 0.8539, 0.5166, 0.3958, 0.4806, 0.5558, 0.5333], [0.5637, 0.3742, 0.7941, 0.2559, 0.3872, 0.3076, 0.6280, 0.5515], [0.5998, 0.4185, 0.8796, 0.5261, 0.3549, 0.4438, 0.6606, 0.5561], [0.6237, 0.4060, 0.8862, 0.4990, 0.3595, 0.3872, 0.5895, 0.5282], [0.5888, 0.3756, 0.7120, 0.2081, 0.3834, 0.2500, 0.6050, 0.5684], [0.5443, 0.3876, 0.8513, 0.4988, 0.4388, 0.5221, 0.5204, 0.5598], [0.5425, 0.3668, 0.8625, 0.4351, 0.4530, 0.4921, 0.5606, 0.5558]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6048, 0.3928, 0.8537, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719], [0.6243, 0.4128, 0.7763, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417], [0.6361, 0.4076, 0.8863, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297], [0.6203, 0.4020, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [0.6273, 0.4099, 0.7138, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6175, 0.3957, 0.8700, 0.4817, 0.4663, 0.5133, 0.5800, 0.5517]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.010210297914454713 step: 12 running loss: 0.0008508581595378928 Train Steps: 12/90 Loss: 0.0009 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5613, 0.5900], [0.6173, 0.4114, 0.7325, 0.2500, 0.4213, 0.1917, 0.5338, 0.5700], [ nan, nan, 0.7225, 0.2167, 0.3987, 0.2283, 0.5427, 0.5181], [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583], [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500], [0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100], [ nan, nan, 0.8525, 0.2217, 0.5413, 0.2367, 0.7367, 0.5482], [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6665, 0.4454, 0.8174, 0.4735, 0.3770, 0.3649, 0.5131, 0.5624], [0.5387, 0.3739, 0.7161, 0.2188, 0.4464, 0.2063, 0.5391, 0.5532], [0.2135, 0.1407, 0.6820, 0.1918, 0.4179, 0.2471, 0.5318, 0.5153], [0.6443, 0.4274, 0.6857, 0.2046, 0.4088, 0.2694, 0.6036, 0.5470], [0.6675, 0.4486, 0.8739, 0.4873, 0.3920, 0.4055, 0.5779, 0.5486], [0.6897, 0.4687, 0.8466, 0.5476, 0.3902, 0.4891, 0.5805, 0.5826], [0.2292, 0.1626, 0.8446, 0.2121, 0.5398, 0.2417, 0.7055, 0.5412], [0.6525, 0.4344, 0.8758, 0.5061, 0.3929, 0.4984, 0.7030, 0.5711]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5612, 0.5900], [0.6173, 0.4114, 0.7325, 0.2500, 0.4212, 0.1917, 0.5337, 0.5700], [0.0000, 0.0000, 0.7225, 0.2167, 0.3988, 0.2283, 0.5427, 0.5181], [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6087, 0.5583], [0.6262, 0.4163, 0.8850, 0.5183, 0.3762, 0.4150, 0.6025, 0.5500], [0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100], [0.0000, 0.0000, 0.8525, 0.2217, 0.5412, 0.2367, 0.7367, 0.5482], [0.6314, 0.4107, 0.8750, 0.5100, 0.3787, 0.4900, 0.7121, 0.5864]]], device='cuda:0') loss_train_step before backward: tensor(0.0029, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0029, device='cuda:0', grad_fn=) loss_train: 0.013074482005322352 step: 13 running loss: 0.0010057293850247962 Train Steps: 13/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6087, 0.3951, 0.8387, 0.5833, 0.4188, 0.4933, 0.5146, 0.4830], [0.6034, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084], [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967], [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6260, 0.4120, 0.8013, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6209, 0.3946, 0.7829, 0.2354, 0.4578, 0.2138, 0.5985, 0.5199], [0.5912, 0.3801, 0.8002, 0.5580, 0.4329, 0.5129, 0.5151, 0.5371], [0.5538, 0.3627, 0.6871, 0.2188, 0.3659, 0.3580, 0.5647, 0.5449], [0.5544, 0.3579, 0.8420, 0.5472, 0.3969, 0.4023, 0.6108, 0.4973], [0.5533, 0.3713, 0.8420, 0.4789, 0.3996, 0.3498, 0.6161, 0.5338], [0.5912, 0.3869, 0.8465, 0.3534, 0.4348, 0.3184, 0.6907, 0.5468], [0.6274, 0.3913, 0.8715, 0.4603, 0.3700, 0.3854, 0.6073, 0.5570], [0.5417, 0.3396, 0.7500, 0.1926, 0.4977, 0.1744, 0.6412, 0.5392]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6277, 0.4056, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6087, 0.3951, 0.8388, 0.5833, 0.4187, 0.4933, 0.5146, 0.4830], [0.6033, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084], [0.6272, 0.4071, 0.8737, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967], [0.6275, 0.4003, 0.9100, 0.3783, 0.4387, 0.3133, 0.7058, 0.5343], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6259, 0.4120, 0.8012, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.014277539594331756 step: 14 running loss: 0.0010198242567379826 Train Steps: 14/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578], [0.6083, 0.3957, 0.8638, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980], [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6201, 0.4017, 0.8871, 0.4621, 0.3517, 0.4675, 0.5999, 0.5106], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426], [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5333, 0.3506, 0.6552, 0.2076, 0.4242, 0.1751, 0.5727, 0.5451], [0.4893, 0.3109, 0.8448, 0.5058, 0.4601, 0.5073, 0.5673, 0.4904], [0.5904, 0.3878, 0.7141, 0.1966, 0.4354, 0.2181, 0.6287, 0.5480], [0.5772, 0.3840, 0.8524, 0.4117, 0.3834, 0.3702, 0.5531, 0.5023], [0.6300, 0.4052, 0.8221, 0.5879, 0.3911, 0.4689, 0.6662, 0.5207], [0.6077, 0.3779, 0.8499, 0.4657, 0.3675, 0.4600, 0.6185, 0.5338], [0.6065, 0.3866, 0.8750, 0.4366, 0.3635, 0.3120, 0.6016, 0.5102], [0.5697, 0.3748, 0.8739, 0.4813, 0.3705, 0.4117, 0.6781, 0.5255]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578], [0.6083, 0.3957, 0.8637, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980], [0.6200, 0.4071, 0.7337, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6201, 0.4017, 0.8871, 0.4621, 0.3517, 0.4675, 0.5999, 0.5106], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426], [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.015163687261519954 step: 15 running loss: 0.0010109124841013303 Train Steps: 15/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6157, 0.4102, 0.8513, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890], [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5410, 0.3491, 0.8061, 0.3897, 0.3737, 0.3296, 0.5710, 0.5361], [0.5908, 0.3836, 0.8486, 0.5070, 0.3837, 0.4084, 0.5487, 0.5290], [0.5794, 0.3698, 0.8578, 0.4783, 0.4464, 0.5413, 0.6610, 0.5036], [0.5673, 0.3616, 0.8306, 0.5433, 0.3512, 0.3234, 0.6101, 0.5062], [0.5835, 0.3757, 0.8322, 0.3436, 0.3784, 0.2795, 0.6196, 0.4811], [0.5601, 0.3804, 0.8507, 0.4410, 0.4294, 0.4775, 0.6118, 0.5142], [0.6267, 0.4019, 0.8304, 0.4744, 0.3647, 0.2856, 0.5849, 0.5461], [0.6285, 0.4035, 0.8221, 0.4240, 0.3711, 0.4418, 0.5872, 0.4826]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6157, 0.4102, 0.8512, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890], [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6084, 0.4005, 0.8400, 0.4317, 0.3762, 0.4750, 0.5476, 0.5058]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.016262349992757663 step: 16 running loss: 0.001016396874547354 Train Steps: 16/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050], [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683], [0.6200, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6273, 0.4105, 0.8988, 0.4517, 0.3912, 0.2550, 0.5894, 0.4811], [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700], [0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748], [0.6138, 0.4101, 0.8800, 0.5083, 0.4637, 0.5950, 0.5587, 0.5077], [0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6855, 0.4411, 0.8811, 0.4829, 0.3468, 0.4056, 0.5976, 0.4854], [0.6003, 0.3924, 0.8620, 0.5050, 0.4516, 0.4777, 0.5454, 0.5403], [0.7025, 0.4502, 0.8847, 0.4959, 0.3405, 0.4132, 0.5959, 0.4986], [0.6217, 0.3892, 0.8945, 0.4343, 0.3822, 0.2371, 0.5838, 0.4497], [0.0274, 0.0246, 0.6846, 0.2371, 0.3901, 0.1621, 0.5692, 0.5287], [0.6936, 0.4408, 0.8705, 0.5191, 0.3849, 0.3495, 0.7009, 0.5208], [0.6504, 0.4256, 0.8840, 0.5347, 0.4355, 0.5537, 0.5773, 0.5028], [0.6316, 0.4177, 0.7242, 0.3255, 0.3338, 0.3280, 0.5260, 0.4997]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050], [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683], [0.6201, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6273, 0.4105, 0.8988, 0.4517, 0.3913, 0.2550, 0.5894, 0.4811], [0.0000, 0.0000, 0.6413, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700], [0.6336, 0.4191, 0.8938, 0.5167, 0.3938, 0.3517, 0.7343, 0.5748], [0.6138, 0.4101, 0.8800, 0.5083, 0.4638, 0.5950, 0.5587, 0.5077], [0.6109, 0.4041, 0.6975, 0.3167, 0.3512, 0.3383, 0.5153, 0.5319]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.0170363754441496 step: 17 running loss: 0.0010021397320088 Train Steps: 17/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6120, 0.4014, 0.6863, 0.2817, 0.3700, 0.2783, 0.5513, 0.5667], [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869], [ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533], [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6260, 0.4011, 0.7258, 0.3199, 0.3612, 0.2915, 0.5442, 0.5361], [0.7346, 0.4767, 0.9288, 0.5064, 0.4090, 0.3807, 0.7193, 0.5532], [0.6979, 0.4502, 0.8781, 0.4688, 0.3707, 0.4679, 0.5312, 0.5525], [0.6071, 0.3951, 0.8454, 0.2538, 0.4587, 0.1876, 0.6201, 0.4572], [0.0940, 0.0501, 0.8083, 0.3363, 0.3641, 0.2925, 0.5379, 0.5242], [0.6665, 0.4243, 0.8967, 0.5811, 0.4485, 0.5496, 0.5898, 0.5064], [0.6506, 0.4104, 0.7844, 0.2195, 0.4589, 0.1643, 0.5733, 0.4783], [0.6472, 0.4191, 0.7120, 0.3332, 0.3684, 0.2936, 0.5573, 0.5656]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6120, 0.4013, 0.6862, 0.2817, 0.3700, 0.2783, 0.5512, 0.5667], [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6276, 0.4095, 0.8238, 0.2250, 0.4663, 0.1783, 0.6171, 0.4869], [0.0000, 0.0000, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533], [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.018217981763882563 step: 18 running loss: 0.0010121100979934756 Train Steps: 18/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6173, 0.4114, 0.7325, 0.2500, 0.4213, 0.1917, 0.5338, 0.5700], [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5787, 0.5600], [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823], [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726], [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6201, 0.4064, 0.8688, 0.5050, 0.4225, 0.5100, 0.6138, 0.5500]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5512, 0.3725, 0.7604, 0.2443, 0.4207, 0.1884, 0.5394, 0.5562], [0.6294, 0.3869, 0.9030, 0.5222, 0.4701, 0.5182, 0.5802, 0.5584], [0.5952, 0.3779, 0.7930, 0.3845, 0.3451, 0.3834, 0.5343, 0.5833], [0.6804, 0.4306, 0.8026, 0.3019, 0.3634, 0.2921, 0.5605, 0.4898], [0.6347, 0.4243, 0.7262, 0.2597, 0.4223, 0.1819, 0.5501, 0.5555], [0.5753, 0.3615, 0.9085, 0.5114, 0.4387, 0.5237, 0.5442, 0.5066], [0.5027, 0.3365, 0.8310, 0.3479, 0.3453, 0.2930, 0.5062, 0.5514], [0.6423, 0.4073, 0.9061, 0.5245, 0.4004, 0.5121, 0.5782, 0.5466]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6173, 0.4114, 0.7325, 0.2500, 0.4212, 0.1917, 0.5337, 0.5700], [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5788, 0.5600], [0.6170, 0.4102, 0.7468, 0.3695, 0.3462, 0.3767, 0.5238, 0.5823], [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726], [0.6154, 0.4112, 0.7038, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6201, 0.4064, 0.8687, 0.5050, 0.4225, 0.5100, 0.6137, 0.5500]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.019125032442389056 step: 19 running loss: 0.0010065806548625819 Train Steps: 19/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412], [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822], [0.6200, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913], [0.6179, 0.4040, 0.7412, 0.1850, 0.3825, 0.2783, 0.5837, 0.5600], [0.6208, 0.4082, 0.8538, 0.3067, 0.3588, 0.3717, 0.6112, 0.5517], [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6359, 0.4048, 0.9052, 0.3762, 0.3995, 0.2797, 0.5899, 0.5466], [0.6064, 0.3951, 0.8949, 0.4519, 0.3761, 0.3334, 0.5104, 0.6051], [0.6164, 0.3700, 0.8848, 0.5386, 0.4169, 0.5256, 0.5715, 0.5245], [0.6481, 0.4276, 0.7565, 0.2404, 0.3964, 0.2855, 0.5759, 0.5843], [0.6212, 0.4118, 0.8703, 0.3438, 0.3663, 0.3934, 0.5951, 0.5677], [0.6131, 0.3926, 0.8612, 0.3559, 0.3660, 0.3619, 0.5658, 0.5549], [0.5858, 0.3855, 0.7570, 0.2178, 0.4388, 0.2146, 0.5812, 0.5519], [0.5676, 0.3535, 0.8760, 0.4405, 0.4186, 0.5101, 0.5055, 0.5193]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412], [0.6182, 0.4058, 0.8737, 0.4350, 0.3562, 0.3400, 0.5290, 0.5822], [0.6199, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913], [0.6179, 0.4040, 0.7412, 0.1850, 0.3825, 0.2783, 0.5838, 0.5600], [0.6208, 0.4082, 0.8537, 0.3067, 0.3587, 0.3717, 0.6112, 0.5517], [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.6068, 0.3963, 0.8650, 0.4317, 0.4038, 0.5083, 0.5253, 0.4999]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.01953231156221591 step: 20 running loss: 0.0009766155781107955 Train Steps: 20/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6206, 0.4001, 0.8900, 0.3933, 0.3588, 0.3567, 0.5837, 0.5083], [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317], [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600], [0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064], [0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617], [ nan, nan, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7210, 0.4565, 0.9084, 0.3731, 0.3526, 0.3772, 0.5655, 0.5376], [0.8087, 0.5395, 0.8758, 0.5171, 0.3853, 0.3925, 0.5778, 0.5738], [0.1119, 0.0785, 0.7751, 0.2568, 0.4144, 0.2698, 0.5499, 0.5953], [0.6822, 0.4636, 0.8693, 0.4949, 0.4949, 0.5419, 0.5108, 0.5413], [0.7736, 0.5026, 0.8111, 0.3225, 0.3772, 0.3114, 0.5589, 0.5785], [0.0275, 0.0309, 0.7832, 0.2323, 0.4275, 0.1838, 0.5191, 0.5356], [0.7302, 0.4657, 0.9031, 0.4539, 0.3850, 0.5179, 0.6317, 0.5228], [0.6900, 0.4722, 0.8493, 0.4946, 0.4511, 0.5650, 0.5092, 0.5610]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6206, 0.4001, 0.8900, 0.3933, 0.3587, 0.3567, 0.5838, 0.5083], [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317], [0.0000, 0.0000, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600], [0.6128, 0.4022, 0.8737, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064], [0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617], [0.0000, 0.0000, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466]]], device='cuda:0') loss_train_step before backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train: 0.022330504696583375 step: 21 running loss: 0.0010633573665039702 Train Steps: 21/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377], [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719], [0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471], [0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [ nan, nan, 0.8888, 0.3100, 0.5262, 0.2817, 0.7145, 0.6003], [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800], [0.6126, 0.4039, 0.8237, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6185, 0.3869, 0.8579, 0.5263, 0.4154, 0.5089, 0.5843, 0.5349], [0.6103, 0.4033, 0.8438, 0.3815, 0.3781, 0.4424, 0.5075, 0.5624], [0.6637, 0.4565, 0.8744, 0.4178, 0.4033, 0.4657, 0.4512, 0.5411], [0.6865, 0.4526, 0.8055, 0.3225, 0.3858, 0.3246, 0.5445, 0.5547], [0.6587, 0.4332, 0.8624, 0.2587, 0.5050, 0.2517, 0.6967, 0.5345], [0.1376, 0.0907, 0.8921, 0.2710, 0.5097, 0.2876, 0.7036, 0.5853], [0.6178, 0.4209, 0.7452, 0.2694, 0.3747, 0.3018, 0.5023, 0.5806], [0.6423, 0.4236, 0.8180, 0.3788, 0.3601, 0.3860, 0.5589, 0.6233]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377], [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719], [0.6127, 0.4084, 0.8700, 0.4467, 0.3988, 0.4317, 0.5013, 0.5471], [0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.0000, 0.0000, 0.8888, 0.3100, 0.5263, 0.2817, 0.7145, 0.6003], [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5462, 0.5800], [0.6126, 0.4038, 0.8238, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.023386229382595047 step: 22 running loss: 0.001063010426481593 Train Steps: 22/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6048, 0.3928, 0.8538, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719], [0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413], [0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183], [0.6234, 0.4179, 0.7825, 0.3450, 0.3813, 0.2867, 0.5675, 0.5617]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5246, 0.3330, 0.8417, 0.5379, 0.3996, 0.5314, 0.5079, 0.5074], [0.5499, 0.3492, 0.8426, 0.2148, 0.5311, 0.2304, 0.6864, 0.5399], [0.5512, 0.3678, 0.6624, 0.2023, 0.4101, 0.2173, 0.4776, 0.5696], [0.5862, 0.3871, 0.8917, 0.2849, 0.4597, 0.2297, 0.5993, 0.5250], [0.5173, 0.3275, 0.7987, 0.2612, 0.3914, 0.2580, 0.5748, 0.5194], [0.5717, 0.3551, 0.8947, 0.4054, 0.3440, 0.4823, 0.6041, 0.5235], [0.5840, 0.3897, 0.7675, 0.2568, 0.3996, 0.3258, 0.5925, 0.6365], [0.5287, 0.3531, 0.7851, 0.3243, 0.3914, 0.2944, 0.5629, 0.5665]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6048, 0.3928, 0.8537, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719], [0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413], [0.6149, 0.4054, 0.6712, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6259, 0.4156, 0.8813, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.6196, 0.4094, 0.7563, 0.2817, 0.3938, 0.3183, 0.6012, 0.6183], [0.6234, 0.4179, 0.7825, 0.3450, 0.3812, 0.2867, 0.5675, 0.5617]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.024698010383872315 step: 23 running loss: 0.001073826538429231 Train Steps: 23/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917], [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600], [0.6203, 0.4078, 0.8800, 0.5083, 0.3900, 0.5000, 0.6100, 0.5583], [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609], [0.6357, 0.4118, 0.8400, 0.2500, 0.5413, 0.1633, 0.6725, 0.5586], [0.6157, 0.4102, 0.8513, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890], [0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5903, 0.3847, 0.8431, 0.5428, 0.4399, 0.4542, 0.5502, 0.5792], [0.5266, 0.3435, 0.7182, 0.2179, 0.3495, 0.3179, 0.5764, 0.5612], [0.5280, 0.3317, 0.8819, 0.4689, 0.3673, 0.4914, 0.5831, 0.5414], [0.6012, 0.3880, 0.8877, 0.2876, 0.4039, 0.2774, 0.6887, 0.5500], [0.5292, 0.3549, 0.8352, 0.2285, 0.5218, 0.1683, 0.6503, 0.5430], [0.5647, 0.3668, 0.8404, 0.3564, 0.3523, 0.3591, 0.5268, 0.5653], [0.5416, 0.3683, 0.8392, 0.5033, 0.4493, 0.5229, 0.5332, 0.5184], [0.4986, 0.3387, 0.7712, 0.2292, 0.4214, 0.1707, 0.5749, 0.5097]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6248, 0.4185, 0.8500, 0.5767, 0.4462, 0.4550, 0.5612, 0.5917], [0.6132, 0.4066, 0.7259, 0.2402, 0.3587, 0.3300, 0.6000, 0.5600], [0.6203, 0.4078, 0.8800, 0.5083, 0.3900, 0.5000, 0.6100, 0.5583], [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609], [0.6357, 0.4118, 0.8400, 0.2500, 0.5412, 0.1633, 0.6725, 0.5586], [0.6157, 0.4102, 0.8512, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890], [0.6075, 0.4000, 0.8512, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.02608701281133108 step: 24 running loss: 0.0010869588671387949 Train Steps: 24/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [ nan, nan, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6092, 0.4001, 0.8638, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064], [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575], [ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383], [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528], [ nan, nan, 0.8363, 0.3317, 0.3563, 0.3367, 0.5329, 0.5142]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7749, 0.5050, 0.8667, 0.5227, 0.4023, 0.5295, 0.7262, 0.5520], [0.1078, 0.0715, 0.8621, 0.3088, 0.5078, 0.1750, 0.6640, 0.5756], [0.7126, 0.4748, 0.8329, 0.4779, 0.4340, 0.5367, 0.5249, 0.5027], [0.7413, 0.5045, 0.8681, 0.4425, 0.4100, 0.4426, 0.5186, 0.5116], [0.7770, 0.5188, 0.8618, 0.3211, 0.4173, 0.4002, 0.6991, 0.5632], [0.0340, 0.0423, 0.6935, 0.2104, 0.4144, 0.1369, 0.5162, 0.5456], [0.7058, 0.4849, 0.8911, 0.3628, 0.4073, 0.3391, 0.7341, 0.5740], [0.1683, 0.1309, 0.7931, 0.3116, 0.3325, 0.3041, 0.5446, 0.5366]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.0000, 0.0000, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6092, 0.4001, 0.8637, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064], [0.6185, 0.4067, 0.8838, 0.4450, 0.4038, 0.4733, 0.5213, 0.5142], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575], [0.0000, 0.0000, 0.7268, 0.2333, 0.4125, 0.1933, 0.5113, 0.5383], [0.6332, 0.4165, 0.9100, 0.3350, 0.4187, 0.3683, 0.7438, 0.5528], [0.0000, 0.0000, 0.8363, 0.3317, 0.3562, 0.3367, 0.5329, 0.5142]]], device='cuda:0') loss_train_step before backward: tensor(0.0032, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0032, device='cuda:0', grad_fn=) loss_train: 0.02925560690346174 step: 25 running loss: 0.0011702242761384696 Train Steps: 25/90 Loss: 0.0012 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633], [0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131], [0.6201, 0.4064, 0.8688, 0.5050, 0.4225, 0.5100, 0.6138, 0.5500], [0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319], [0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5933, 0.3969, 0.8577, 0.3590, 0.4118, 0.2590, 0.5607, 0.5364], [0.5607, 0.3531, 0.8947, 0.4054, 0.3543, 0.4176, 0.6658, 0.5124], [0.5784, 0.3753, 0.8585, 0.5589, 0.4376, 0.4992, 0.6307, 0.5515], [0.5546, 0.3648, 0.7714, 0.2658, 0.4262, 0.2079, 0.6115, 0.5436], [0.4776, 0.3190, 0.8711, 0.3610, 0.3545, 0.3455, 0.6230, 0.5217], [0.5311, 0.3501, 0.8663, 0.4774, 0.4244, 0.4933, 0.6378, 0.5392], [0.5744, 0.3876, 0.7364, 0.2986, 0.3520, 0.3252, 0.5476, 0.5329], [0.6285, 0.4031, 0.8640, 0.5768, 0.3735, 0.3970, 0.6315, 0.5407]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6186, 0.4080, 0.8625, 0.3483, 0.3787, 0.2650, 0.5320, 0.5272], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.6192, 0.4128, 0.8512, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633], [0.6229, 0.4066, 0.7613, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131], [0.6201, 0.4064, 0.8687, 0.5050, 0.4225, 0.5100, 0.6137, 0.5500], [0.6109, 0.4041, 0.6975, 0.3167, 0.3512, 0.3383, 0.5153, 0.5319], [0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.03047404598328285 step: 26 running loss: 0.001172078691664725 Train Steps: 26/90 Loss: 0.0012 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167], [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621], [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285], [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.5571, 0.3781, 0.8592, 0.4977, 0.3940, 0.5043, 0.6304, 0.5167], [-0.0641, -0.0320, 0.8451, 0.2732, 0.5266, 0.1686, 0.6831, 0.5694], [ 0.5501, 0.3686, 0.8577, 0.4885, 0.4755, 0.5070, 0.5333, 0.5305], [ 0.6006, 0.4094, 0.8789, 0.5209, 0.3491, 0.3783, 0.6467, 0.4951], [ 0.5343, 0.3618, 0.8626, 0.4401, 0.4468, 0.5363, 0.6245, 0.5325], [ 0.5859, 0.4038, 0.8746, 0.5054, 0.3550, 0.4501, 0.6062, 0.5472], [ 0.5593, 0.3798, 0.8754, 0.3642, 0.4343, 0.2383, 0.6505, 0.5330], [ 0.6434, 0.4352, 0.8833, 0.4716, 0.3494, 0.3238, 0.6562, 0.5142]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167], [0.0000, 0.0000, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621], [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6222, 0.4171, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285], [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.03130909425090067 step: 27 running loss: 0.0011595960833666915 Train Steps: 27/90 Loss: 0.0012 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6203, 0.4078, 0.8800, 0.5083, 0.3900, 0.5000, 0.6100, 0.5583], [0.6153, 0.4117, 0.8688, 0.5167, 0.4895, 0.5647, 0.5524, 0.5136], [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5837, 0.5317], [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329], [0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [ nan, nan, 0.8525, 0.2217, 0.5413, 0.2367, 0.7367, 0.5482], [0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [ nan, nan, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6480, 0.4196, 0.8940, 0.5324, 0.3837, 0.5083, 0.6373, 0.5335], [ 0.6795, 0.4661, 0.8864, 0.5449, 0.4735, 0.5319, 0.6194, 0.5147], [ 0.5716, 0.3851, 0.8499, 0.3645, 0.3668, 0.3217, 0.6308, 0.5389], [ 0.6398, 0.4191, 0.8459, 0.3649, 0.3495, 0.3682, 0.6308, 0.5194], [ 0.6781, 0.4620, 0.7139, 0.3219, 0.4760, 0.1850, 0.5917, 0.5935], [-0.0955, -0.0567, 0.8607, 0.2603, 0.5467, 0.2307, 0.7484, 0.5299], [ 0.6616, 0.4385, 0.8714, 0.4713, 0.3620, 0.3759, 0.5641, 0.5374], [ 0.1023, 0.0534, 0.6864, 0.2369, 0.4190, 0.1628, 0.5558, 0.5566]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6203, 0.4078, 0.8800, 0.5083, 0.3900, 0.5000, 0.6100, 0.5583], [0.6154, 0.4117, 0.8687, 0.5167, 0.4895, 0.5647, 0.5524, 0.5136], [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5838, 0.5317], [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329], [0.6270, 0.4266, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [0.0000, 0.0000, 0.8525, 0.2217, 0.5412, 0.2367, 0.7367, 0.5482], [0.6151, 0.4125, 0.8737, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.0000, 0.0000, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.03243761681369506 step: 28 running loss: 0.0011584863147748234 Train Steps: 28/90 Loss: 0.0012 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6208, 0.4082, 0.8538, 0.3067, 0.3588, 0.3717, 0.6112, 0.5517], [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954], [0.6332, 0.4118, 0.9238, 0.4267, 0.4012, 0.4733, 0.7525, 0.5436], [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5333, 0.3519, 0.8224, 0.5819, 0.4291, 0.4459, 0.5833, 0.5994], [0.5432, 0.3570, 0.8515, 0.3269, 0.3724, 0.3711, 0.6106, 0.5296], [0.5232, 0.3436, 0.8307, 0.3767, 0.3578, 0.3805, 0.5443, 0.4934], [0.6259, 0.4124, 0.9127, 0.4304, 0.4193, 0.4777, 0.7340, 0.5407], [0.5638, 0.3596, 0.8785, 0.5704, 0.3951, 0.3730, 0.6073, 0.4628], [0.5675, 0.3640, 0.8701, 0.5401, 0.4035, 0.4203, 0.6903, 0.5244], [0.5548, 0.3612, 0.8543, 0.4221, 0.3882, 0.4384, 0.5354, 0.5408], [0.5765, 0.3649, 0.8960, 0.4664, 0.4072, 0.3573, 0.5822, 0.5698]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6208, 0.4082, 0.8537, 0.3067, 0.3587, 0.3717, 0.6112, 0.5517], [0.6076, 0.3952, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954], [0.6332, 0.4118, 0.9237, 0.4267, 0.4013, 0.4733, 0.7525, 0.5436], [0.6272, 0.4071, 0.8737, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.03350451713777147 step: 29 running loss: 0.0011553281771645334 Train Steps: 29/90 Loss: 0.0012 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167], [0.6317, 0.4038, 0.8287, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986], [0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183], [0.6261, 0.4029, 0.8720, 0.3364, 0.3665, 0.3753, 0.6531, 0.5183], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6097, 0.4000, 0.7325, 0.2667, 0.3450, 0.3517, 0.5284, 0.5045], [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5300, 0.3320, 0.8162, 0.3578, 0.3960, 0.3841, 0.5884, 0.6032], [0.5531, 0.3594, 0.8705, 0.6327, 0.3978, 0.4552, 0.6268, 0.5013], [0.6140, 0.3880, 0.8141, 0.3311, 0.4078, 0.2734, 0.5962, 0.5032], [0.5975, 0.3754, 0.8768, 0.3859, 0.3943, 0.3580, 0.6395, 0.5108], [0.5600, 0.3510, 0.8141, 0.3194, 0.3890, 0.3210, 0.6084, 0.5404], [0.5502, 0.3410, 0.7539, 0.3059, 0.3619, 0.3273, 0.5426, 0.5130], [0.5694, 0.3608, 0.8149, 0.3133, 0.4233, 0.2520, 0.5978, 0.5442], [0.5955, 0.3840, 0.7417, 0.2904, 0.3905, 0.2555, 0.5303, 0.5149]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6037, 0.6167], [0.6317, 0.4038, 0.8288, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986], [0.6201, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183], [0.6261, 0.4029, 0.8720, 0.3364, 0.3665, 0.3753, 0.6531, 0.5183], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6097, 0.4000, 0.7325, 0.2667, 0.3450, 0.3517, 0.5284, 0.5045], [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.03453243928379379 step: 30 running loss: 0.001151081309459793 Train Steps: 30/90 Loss: 0.0012 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626], [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6064, 0.3953, 0.8738, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416], [0.6272, 0.4045, 0.8538, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [ nan, nan, 0.8850, 0.2817, 0.5112, 0.2183, 0.7184, 0.5436], [0.6357, 0.4097, 0.9038, 0.3883, 0.4213, 0.2950, 0.6686, 0.5390], [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5837, 0.5583]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6427, 0.4202, 0.8314, 0.6064, 0.3901, 0.5030, 0.5913, 0.4913], [0.6279, 0.4157, 0.8776, 0.5284, 0.3637, 0.4915, 0.5434, 0.5713], [0.6431, 0.4021, 0.8803, 0.4585, 0.3645, 0.4955, 0.5290, 0.5244], [0.6660, 0.4271, 0.8366, 0.5968, 0.3659, 0.4531, 0.5882, 0.4971], [0.2643, 0.1526, 0.8646, 0.2992, 0.5143, 0.2550, 0.6820, 0.5459], [0.6371, 0.4096, 0.8860, 0.3984, 0.3991, 0.3012, 0.6437, 0.5549], [0.6351, 0.4026, 0.8868, 0.4923, 0.3529, 0.4365, 0.5662, 0.5320], [0.6200, 0.4089, 0.8712, 0.4522, 0.4741, 0.5675, 0.5527, 0.5659]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626], [0.6222, 0.4171, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6064, 0.3952, 0.8737, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416], [0.6271, 0.4045, 0.8537, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.0000, 0.0000, 0.8850, 0.2817, 0.5113, 0.2183, 0.7184, 0.5436], [0.6357, 0.4097, 0.9038, 0.3883, 0.4212, 0.2950, 0.6686, 0.5390], [0.6236, 0.3965, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5838, 0.5583]]], device='cuda:0') loss_train_step before backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0018, device='cuda:0', grad_fn=) loss_train: 0.03629230541992001 step: 31 running loss: 0.0011707195296748391 Train Steps: 31/90 Loss: 0.0012 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412], [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131], [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397], [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312], [0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823], [0.6110, 0.4047, 0.8700, 0.4483, 0.3713, 0.3967, 0.5088, 0.5517]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6312, 0.4087, 0.8552, 0.2854, 0.4418, 0.2874, 0.6899, 0.5743], [0.6343, 0.4257, 0.8424, 0.5669, 0.4541, 0.5228, 0.5158, 0.5439], [0.5952, 0.3849, 0.8683, 0.5273, 0.4134, 0.5349, 0.5800, 0.5326], [0.6040, 0.3875, 0.8703, 0.3866, 0.3464, 0.3959, 0.5834, 0.5296], [0.5615, 0.3480, 0.6886, 0.2595, 0.4079, 0.2172, 0.4974, 0.5599], [0.6166, 0.3884, 0.8692, 0.4099, 0.3551, 0.4637, 0.5817, 0.5406], [0.5760, 0.3547, 0.7143, 0.2348, 0.3887, 0.2061, 0.5262, 0.5102], [0.5566, 0.3679, 0.8671, 0.4734, 0.3623, 0.4301, 0.5038, 0.5463]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412], [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131], [0.6159, 0.4085, 0.6900, 0.2283, 0.4087, 0.1950, 0.5123, 0.5397], [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312], [0.6055, 0.4015, 0.7425, 0.2033, 0.4112, 0.1883, 0.5217, 0.4823], [0.6110, 0.4047, 0.8700, 0.4483, 0.3713, 0.3967, 0.5088, 0.5517]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.03677627723664045 step: 32 running loss: 0.0011492586636450142 Train Steps: 32/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617], [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183], [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717], [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835], [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999], [0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517], [ nan, nan, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819], [0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6957, 0.4481, 0.8929, 0.5009, 0.4728, 0.5124, 0.5399, 0.5542], [0.6731, 0.4449, 0.7781, 0.3020, 0.3991, 0.3333, 0.6170, 0.6263], [0.6312, 0.4144, 0.6878, 0.2701, 0.3600, 0.2677, 0.5420, 0.5699], [0.6798, 0.4307, 0.9188, 0.4046, 0.4165, 0.2358, 0.5908, 0.4885], [0.6334, 0.4017, 0.8684, 0.4303, 0.3911, 0.5226, 0.5477, 0.4909], [0.6938, 0.4446, 0.8939, 0.4294, 0.4075, 0.5651, 0.5999, 0.5187], [0.1302, 0.0652, 0.6944, 0.2737, 0.3742, 0.2653, 0.5407, 0.5516], [0.6296, 0.4025, 0.6964, 0.2256, 0.3679, 0.2693, 0.5542, 0.5215]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617], [0.6196, 0.4094, 0.7563, 0.2817, 0.3938, 0.3183, 0.6012, 0.6183], [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717], [0.6273, 0.4110, 0.8900, 0.3817, 0.4187, 0.2167, 0.5858, 0.4835], [0.6068, 0.3963, 0.8650, 0.4317, 0.4038, 0.5083, 0.5253, 0.4999], [0.6122, 0.4006, 0.8850, 0.4217, 0.4087, 0.5517, 0.6062, 0.5517], [0.0000, 0.0000, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819], [0.6043, 0.4022, 0.6888, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.037722738285083324 step: 33 running loss: 0.0011431132813661613 Train Steps: 33/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6273, 0.4100, 0.7137, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633], [0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583], [0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533], [0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517], [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6207, 0.4087, 0.7299, 0.2346, 0.3832, 0.2794, 0.5773, 0.5598], [0.6340, 0.4120, 0.9302, 0.4444, 0.3620, 0.4391, 0.5457, 0.5405], [0.6096, 0.4085, 0.8734, 0.5473, 0.5027, 0.5276, 0.4847, 0.5311], [0.6392, 0.4208, 0.9139, 0.4806, 0.3919, 0.4799, 0.5381, 0.5347], [0.6476, 0.4382, 0.7389, 0.2453, 0.3673, 0.3284, 0.5825, 0.5613], [0.5024, 0.3274, 0.6944, 0.2168, 0.3967, 0.2365, 0.5230, 0.5248], [0.5912, 0.4161, 0.7449, 0.2064, 0.4182, 0.2596, 0.5937, 0.5524], [0.6630, 0.4463, 0.8269, 0.5501, 0.3866, 0.5200, 0.6644, 0.5626]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6273, 0.4099, 0.7138, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633], [0.6212, 0.4033, 0.8938, 0.4167, 0.3812, 0.4267, 0.5612, 0.5583], [0.6084, 0.4008, 0.8587, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533], [0.6054, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6200, 0.4071, 0.7337, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517], [0.6346, 0.4086, 0.7937, 0.5500, 0.3963, 0.4867, 0.7343, 0.5702]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.03848643408855423 step: 34 running loss: 0.001131953943781007 Train Steps: 34/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397], [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533], [0.6222, 0.3957, 0.8838, 0.5017, 0.3937, 0.4600, 0.5900, 0.5017], [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500], [0.6224, 0.4097, 0.7438, 0.2267, 0.3850, 0.2850, 0.5988, 0.5250], [0.6200, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6030, 0.3969, 0.7988, 0.3917, 0.3450, 0.3667, 0.5266, 0.4700], [0.6275, 0.4048, 0.8488, 0.2883, 0.4463, 0.2033, 0.6321, 0.5155]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6183, 0.4021, 0.8597, 0.4573, 0.3796, 0.3890, 0.5009, 0.5528], [0.6546, 0.4281, 0.8510, 0.4200, 0.3696, 0.3607, 0.5795, 0.5611], [0.6465, 0.4086, 0.8639, 0.4932, 0.3910, 0.4767, 0.5734, 0.5107], [0.6153, 0.4060, 0.7664, 0.2453, 0.3581, 0.4111, 0.6184, 0.5636], [0.6052, 0.4063, 0.7235, 0.2073, 0.3895, 0.3052, 0.5874, 0.5488], [0.6432, 0.4107, 0.8695, 0.4533, 0.3793, 0.4444, 0.5860, 0.5381], [0.6142, 0.4123, 0.8144, 0.3713, 0.3490, 0.3675, 0.5293, 0.5152], [0.7097, 0.4605, 0.8296, 0.2603, 0.4559, 0.2119, 0.6177, 0.5250]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6127, 0.4084, 0.8737, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397], [0.6252, 0.4162, 0.8662, 0.4467, 0.3625, 0.3567, 0.6037, 0.5533], [0.6222, 0.3957, 0.8838, 0.5017, 0.3938, 0.4600, 0.5900, 0.5017], [0.6197, 0.4051, 0.7812, 0.2650, 0.3512, 0.4050, 0.6112, 0.5500], [0.6224, 0.4097, 0.7437, 0.2267, 0.3850, 0.2850, 0.5987, 0.5250], [0.6201, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6030, 0.3969, 0.7987, 0.3917, 0.3450, 0.3667, 0.5266, 0.4700], [0.6275, 0.4048, 0.8487, 0.2883, 0.4462, 0.2033, 0.6321, 0.5155]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.03889687912305817 step: 35 running loss: 0.0011113394035159477 Train Steps: 35/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183], [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083], [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822], [0.6147, 0.4081, 0.8538, 0.3400, 0.3663, 0.3150, 0.5142, 0.4875], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483], [0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6339, 0.4102, 0.8588, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6272, 0.4205, 0.8835, 0.3393, 0.3981, 0.2917, 0.6411, 0.5323], [0.5420, 0.3742, 0.8473, 0.4490, 0.3551, 0.4554, 0.6004, 0.5777], [0.6386, 0.4225, 0.8350, 0.4646, 0.4252, 0.4898, 0.5376, 0.5558], [0.6313, 0.4140, 0.8614, 0.3493, 0.3677, 0.3204, 0.5472, 0.5091], [0.5479, 0.3567, 0.8535, 0.3825, 0.3773, 0.4866, 0.6150, 0.5179], [0.5576, 0.3567, 0.6646, 0.1769, 0.3972, 0.2270, 0.5550, 0.5160], [0.6751, 0.4624, 0.7140, 0.2613, 0.4381, 0.2245, 0.5698, 0.5634], [0.6748, 0.4515, 0.8537, 0.2882, 0.4649, 0.2031, 0.6719, 0.5120]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183], [0.6186, 0.4060, 0.8750, 0.5050, 0.3537, 0.4367, 0.5813, 0.6083], [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822], [0.6147, 0.4081, 0.8537, 0.3400, 0.3663, 0.3150, 0.5142, 0.4875], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483], [0.6054, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6339, 0.4102, 0.8587, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.03962359408615157 step: 36 running loss: 0.001100655391281988 Train Steps: 36/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800], [0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397], [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167], [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799], [0.6202, 0.4079, 0.8025, 0.2500, 0.3763, 0.3217, 0.6125, 0.5533], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220], [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6731, 0.4660, 0.7442, 0.2541, 0.3794, 0.2471, 0.5337, 0.5815], [0.5771, 0.3816, 0.8796, 0.4264, 0.3821, 0.3529, 0.5279, 0.5435], [0.5705, 0.3869, 0.9043, 0.3565, 0.3812, 0.3285, 0.6388, 0.5319], [0.5005, 0.3500, 0.8975, 0.4178, 0.3760, 0.4214, 0.6897, 0.5422], [0.5129, 0.3521, 0.8153, 0.3398, 0.3702, 0.2708, 0.5315, 0.5734], [0.5919, 0.4103, 0.7996, 0.1988, 0.3947, 0.3031, 0.6229, 0.5450], [0.6546, 0.4317, 0.6917, 0.1848, 0.4577, 0.1434, 0.5603, 0.5350], [0.6887, 0.4524, 0.8398, 0.3022, 0.3768, 0.2874, 0.5113, 0.5499]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5462, 0.5800], [0.6127, 0.4084, 0.8737, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397], [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6263, 0.5167], [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6200, 0.4118, 0.8288, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799], [0.6202, 0.4079, 0.8025, 0.2500, 0.3762, 0.3217, 0.6125, 0.5533], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220], [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.04108484456082806 step: 37 running loss: 0.0011104012043467043 Train Steps: 37/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6132, 0.4037, 0.6963, 0.2217, 0.4100, 0.1950, 0.5395, 0.5175], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117], [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417], [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319], [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7004, 0.4781, 0.6918, 0.2217, 0.4056, 0.1450, 0.5147, 0.5004], [0.6227, 0.4156, 0.8690, 0.3062, 0.3745, 0.2790, 0.5618, 0.5130], [0.5683, 0.3657, 0.9082, 0.3851, 0.4262, 0.3278, 0.6973, 0.5923], [0.6048, 0.4049, 0.7707, 0.2541, 0.3962, 0.2739, 0.6194, 0.5442], [0.5766, 0.3658, 0.8741, 0.4772, 0.4067, 0.5390, 0.6700, 0.5635], [0.6280, 0.4125, 0.8869, 0.4231, 0.3597, 0.2912, 0.6195, 0.4975], [0.5745, 0.3832, 0.7137, 0.2926, 0.3510, 0.3062, 0.5140, 0.5181], [0.6141, 0.4022, 0.8545, 0.4257, 0.3587, 0.3034, 0.5714, 0.5359]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6132, 0.4037, 0.6963, 0.2217, 0.4100, 0.1950, 0.5395, 0.5175], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117], [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6243, 0.4128, 0.7763, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417], [0.6293, 0.4024, 0.8750, 0.5000, 0.4013, 0.5733, 0.7121, 0.5633], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6109, 0.4041, 0.6975, 0.3167, 0.3512, 0.3383, 0.5153, 0.5319], [0.6252, 0.4162, 0.8662, 0.4467, 0.3625, 0.3567, 0.6037, 0.5533]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.04196064005373046 step: 38 running loss: 0.001104227369835012 Train Steps: 38/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274], [0.6059, 0.4002, 0.7562, 0.2767, 0.3538, 0.3033, 0.5529, 0.5455], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6138, 0.5883], [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578], [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822], [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6063, 0.6083], [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6322, 0.4343, 0.8624, 0.3162, 0.4473, 0.1855, 0.6437, 0.4917], [0.6319, 0.4302, 0.7483, 0.2713, 0.3327, 0.2648, 0.5815, 0.5005], [0.6248, 0.3952, 0.8612, 0.4800, 0.4064, 0.4749, 0.7014, 0.5224], [0.6160, 0.4161, 0.8835, 0.4645, 0.4237, 0.5160, 0.5874, 0.5665], [0.6446, 0.4308, 0.6868, 0.2093, 0.3712, 0.1665, 0.5641, 0.5247], [0.6253, 0.4151, 0.8410, 0.4777, 0.4036, 0.4330, 0.5291, 0.5450], [0.5632, 0.3792, 0.8631, 0.3644, 0.3287, 0.3986, 0.6136, 0.5751], [0.6078, 0.4056, 0.8220, 0.5465, 0.4432, 0.4151, 0.5733, 0.5377]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274], [0.6059, 0.4002, 0.7563, 0.2767, 0.3537, 0.3033, 0.5529, 0.5455], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6137, 0.5883], [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578], [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822], [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6062, 0.6083], [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.042734951653983444 step: 39 running loss: 0.0010957679911277806 Train Steps: 39/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285], [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800], [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344], [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117], [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283], [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6414, 0.4290, 0.8107, 0.5311, 0.3755, 0.4731, 0.6912, 0.5280], [0.6003, 0.3886, 0.7825, 0.4177, 0.3560, 0.4139, 0.5226, 0.5284], [0.6446, 0.4353, 0.7344, 0.3098, 0.3624, 0.2710, 0.5068, 0.5727], [0.5883, 0.3823, 0.8884, 0.4842, 0.3630, 0.3670, 0.5167, 0.5230], [0.6610, 0.4082, 0.8864, 0.4826, 0.3667, 0.4435, 0.6328, 0.5101], [0.6120, 0.3974, 0.8367, 0.4157, 0.3598, 0.4524, 0.5596, 0.4991], [0.7246, 0.4780, 0.9052, 0.3613, 0.3900, 0.2460, 0.6239, 0.5306], [0.6220, 0.4103, 0.8976, 0.4861, 0.3990, 0.3335, 0.7187, 0.5409]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6289, 0.4019, 0.8112, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285], [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5462, 0.5800], [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344], [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117], [0.6084, 0.4005, 0.8400, 0.4317, 0.3762, 0.4750, 0.5476, 0.5058], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6187, 0.5283], [0.6307, 0.4029, 0.8988, 0.4817, 0.3938, 0.3500, 0.7311, 0.5378]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.0431966645992361 step: 40 running loss: 0.0010799166149809026 Train Steps: 40/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5837, 0.5317], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6063, 0.5617]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6134, 0.3949, 0.8595, 0.3987, 0.3494, 0.3724, 0.5910, 0.5731], [0.6684, 0.4354, 0.8558, 0.3522, 0.3534, 0.3221, 0.6058, 0.5432], [0.6412, 0.4086, 0.8021, 0.3002, 0.3626, 0.3407, 0.6261, 0.5365], [0.6907, 0.4500, 0.8277, 0.5763, 0.4382, 0.4893, 0.5934, 0.5552], [0.6440, 0.4196, 0.8511, 0.5767, 0.3726, 0.4335, 0.6587, 0.4990], [0.6427, 0.4119, 0.8444, 0.4761, 0.3769, 0.4700, 0.5662, 0.5726], [0.6348, 0.3972, 0.8611, 0.3470, 0.3555, 0.3311, 0.6533, 0.5098], [0.6770, 0.4490, 0.8951, 0.5009, 0.3560, 0.4613, 0.6049, 0.5530]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5838, 0.5317], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6229, 0.4066, 0.8512, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6321, 0.4048, 0.8737, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6255, 0.4017, 0.8687, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6062, 0.5617]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.04362316819606349 step: 41 running loss: 0.0010639797120991095 Train Steps: 41/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6164, 0.4119, 0.7913, 0.2650, 0.3538, 0.3500, 0.5614, 0.5038], [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633], [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869], [0.6198, 0.4115, 0.7762, 0.2717, 0.3713, 0.3200, 0.5837, 0.5683], [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290], [0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397], [0.6203, 0.4021, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [0.6222, 0.4169, 0.8638, 0.5650, 0.4313, 0.4783, 0.5637, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5952, 0.3894, 0.7800, 0.2822, 0.3461, 0.3827, 0.6112, 0.5220], [0.6692, 0.4349, 0.8421, 0.5854, 0.4277, 0.5497, 0.5884, 0.5794], [0.7066, 0.4452, 0.8616, 0.3625, 0.4512, 0.2253, 0.6105, 0.5051], [0.6223, 0.3899, 0.7771, 0.2893, 0.3718, 0.3561, 0.5877, 0.5753], [0.5444, 0.3587, 0.7548, 0.2279, 0.4270, 0.2348, 0.6474, 0.5424], [0.6348, 0.4049, 0.8640, 0.4997, 0.3731, 0.3949, 0.5174, 0.5608], [0.6215, 0.3988, 0.8642, 0.5334, 0.3727, 0.4115, 0.6142, 0.5432], [0.6647, 0.4366, 0.8561, 0.5870, 0.4228, 0.4839, 0.5994, 0.5959]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6164, 0.4119, 0.7912, 0.2650, 0.3537, 0.3500, 0.5614, 0.5038], [0.6192, 0.4128, 0.8512, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633], [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869], [0.6198, 0.4115, 0.7763, 0.2717, 0.3713, 0.3200, 0.5838, 0.5683], [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290], [0.6127, 0.4084, 0.8737, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397], [0.6203, 0.4020, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [0.6222, 0.4169, 0.8637, 0.5650, 0.4313, 0.4783, 0.5638, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.04430640529608354 step: 42 running loss: 0.001054914411811513 Train Steps: 42/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690], [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433], [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6063, 0.5617], [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400], [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633], [0.6087, 0.3976, 0.8337, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374], [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6467, 0.4283, 0.8193, 0.5591, 0.4115, 0.5895, 0.7145, 0.5837], [0.6037, 0.3921, 0.8437, 0.3812, 0.3790, 0.3736, 0.5460, 0.5361], [0.6682, 0.4433, 0.8867, 0.5180, 0.3720, 0.4951, 0.5859, 0.5464], [0.6078, 0.4041, 0.8402, 0.3575, 0.3668, 0.4851, 0.6394, 0.5646], [0.6845, 0.4671, 0.8428, 0.5293, 0.3853, 0.4209, 0.5545, 0.5740], [0.6057, 0.3982, 0.8355, 0.4236, 0.3830, 0.3535, 0.5858, 0.5416], [0.5911, 0.3794, 0.8517, 0.4044, 0.3831, 0.3585, 0.5884, 0.5192], [0.6004, 0.3727, 0.9194, 0.3995, 0.4778, 0.3034, 0.7175, 0.5427]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690], [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5412, 0.5433], [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6062, 0.5617], [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400], [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5638, 0.5633], [0.6087, 0.3976, 0.8338, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374], [0.6271, 0.4040, 0.9137, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.044957086036447436 step: 43 running loss: 0.0010455136287545914 Train Steps: 43/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4151, 0.8588, 0.5467, 0.3700, 0.3950, 0.5637, 0.5933], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389], [0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433], [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6263, 0.4233, 0.7924, 0.4626, 0.3788, 0.2883, 0.5573, 0.6047], [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367], [0.6102, 0.4020, 0.8638, 0.3717, 0.3625, 0.5017, 0.6038, 0.5500]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6396, 0.4144, 0.8675, 0.5608, 0.3827, 0.4323, 0.5664, 0.5795], [0.5833, 0.3797, 0.8016, 0.2433, 0.3960, 0.2990, 0.6173, 0.5316], [0.6888, 0.4475, 0.8880, 0.4995, 0.4760, 0.5530, 0.6200, 0.5388], [0.6149, 0.3904, 0.8058, 0.4255, 0.3589, 0.4215, 0.5320, 0.5407], [0.6182, 0.4064, 0.8864, 0.5289, 0.4396, 0.5638, 0.6034, 0.5243], [0.6576, 0.4455, 0.8199, 0.5035, 0.4000, 0.3319, 0.5725, 0.6120], [0.5834, 0.3808, 0.8157, 0.2679, 0.4248, 0.2818, 0.6214, 0.5289], [0.6289, 0.4154, 0.8847, 0.3929, 0.3923, 0.5196, 0.6435, 0.5359]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6202, 0.4151, 0.8587, 0.5467, 0.3700, 0.3950, 0.5638, 0.5933], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389], [0.6250, 0.4054, 0.8770, 0.4723, 0.4663, 0.5367, 0.6162, 0.5433], [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6263, 0.4232, 0.7924, 0.4626, 0.3787, 0.2883, 0.5573, 0.6047], [0.6204, 0.4110, 0.7912, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367], [0.6102, 0.4020, 0.8637, 0.3717, 0.3625, 0.5017, 0.6037, 0.5500]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.04550424014450982 step: 44 running loss: 0.0010341872760115868 Train Steps: 44/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167], [ nan, nan, 0.7725, 0.2611, 0.3675, 0.2733, 0.5413, 0.5167], [0.6127, 0.4115, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495], [0.6048, 0.3928, 0.8538, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719], [0.6275, 0.4157, 0.8337, 0.5800, 0.3763, 0.4200, 0.5547, 0.6125], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690], [0.6087, 0.3976, 0.8337, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6898, 0.4383, 0.9413, 0.4102, 0.4053, 0.3490, 0.6271, 0.5214], [0.0654, 0.0371, 0.8057, 0.2584, 0.3954, 0.3384, 0.5145, 0.5398], [0.5854, 0.3809, 0.7735, 0.2698, 0.3950, 0.3140, 0.5210, 0.5573], [0.7269, 0.4606, 0.8708, 0.5288, 0.4172, 0.5184, 0.5386, 0.5147], [0.7218, 0.4794, 0.8484, 0.5600, 0.4143, 0.4491, 0.5731, 0.6127], [0.6632, 0.4350, 0.8738, 0.5665, 0.3998, 0.5053, 0.6562, 0.5334], [0.6927, 0.4557, 0.8649, 0.5187, 0.4277, 0.5654, 0.7123, 0.5794], [0.6533, 0.4278, 0.8803, 0.3830, 0.3984, 0.3294, 0.5873, 0.5378]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6263, 0.5167], [0.0000, 0.0000, 0.7725, 0.2611, 0.3675, 0.2733, 0.5412, 0.5167], [0.6127, 0.4114, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495], [0.6048, 0.3928, 0.8537, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719], [0.6275, 0.4157, 0.8338, 0.5800, 0.3762, 0.4200, 0.5547, 0.6125], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690], [0.6087, 0.3976, 0.8338, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.04684005380840972 step: 45 running loss: 0.0010408900846313272 Train Steps: 45/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467], [0.6222, 0.3957, 0.8838, 0.5017, 0.3937, 0.4600, 0.5900, 0.5017], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6199, 0.4071, 0.7600, 0.2117, 0.4037, 0.2767, 0.6138, 0.5550], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550], [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.0948, 0.0649, 0.7369, 0.2828, 0.4063, 0.2761, 0.5456, 0.5900], [0.6272, 0.3999, 0.7922, 0.2179, 0.4729, 0.1858, 0.5854, 0.5116], [0.6376, 0.4229, 0.7174, 0.2835, 0.3986, 0.2754, 0.5660, 0.5660], [0.6890, 0.4232, 0.9184, 0.5697, 0.4056, 0.4773, 0.5971, 0.5330], [0.6022, 0.3880, 0.8296, 0.3167, 0.3761, 0.3408, 0.6021, 0.5541], [0.6351, 0.4145, 0.8118, 0.2516, 0.4369, 0.2929, 0.6219, 0.5716], [0.5690, 0.3789, 0.7400, 0.2806, 0.4204, 0.2384, 0.5463, 0.5773], [0.6463, 0.4115, 0.7827, 0.2099, 0.3979, 0.2565, 0.5745, 0.5275]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6162, 0.4134, 0.6700, 0.2467, 0.3963, 0.2533, 0.5738, 0.5467], [0.6222, 0.3957, 0.8838, 0.5017, 0.3938, 0.4600, 0.5900, 0.5017], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6199, 0.4071, 0.7600, 0.2117, 0.4038, 0.2767, 0.6137, 0.5550], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5362, 0.5550], [0.6165, 0.4106, 0.7575, 0.1733, 0.3837, 0.2650, 0.5680, 0.5116]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.04774374206317589 step: 46 running loss: 0.0010379074361559976 Train Steps: 46/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402], [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092], [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804], [0.6157, 0.4102, 0.8513, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890], [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926], [0.6325, 0.4165, 0.9000, 0.4617, 0.3813, 0.4900, 0.7485, 0.5447], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6132, 0.3930, 0.8448, 0.5366, 0.3979, 0.4417, 0.5856, 0.5404], [0.6065, 0.3869, 0.8564, 0.5038, 0.3883, 0.5033, 0.5811, 0.4995], [0.6180, 0.4011, 0.8389, 0.5640, 0.3996, 0.4710, 0.5505, 0.6195], [0.5635, 0.3521, 0.8894, 0.4405, 0.3446, 0.4432, 0.5997, 0.4692], [0.5423, 0.3544, 0.8496, 0.3760, 0.3585, 0.3441, 0.4819, 0.5728], [0.5672, 0.3738, 0.7683, 0.2288, 0.3706, 0.2683, 0.5741, 0.4807], [0.5929, 0.3905, 0.9018, 0.4519, 0.3811, 0.4729, 0.7004, 0.5482], [0.5863, 0.3843, 0.8827, 0.4588, 0.4425, 0.5560, 0.6066, 0.5455]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402], [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092], [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6263, 0.4038, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804], [0.6157, 0.4102, 0.8512, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890], [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926], [0.6325, 0.4165, 0.9000, 0.4617, 0.3812, 0.4900, 0.7485, 0.5447], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.04828619211912155 step: 47 running loss: 0.0010273657897685437 Train Steps: 47/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947], [0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494], [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6063, 0.5567], [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841], [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200], [0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183], [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5316, 0.3312, 0.7556, 0.1395, 0.4007, 0.2116, 0.6300, 0.4945], [0.5704, 0.3699, 0.8897, 0.3588, 0.4087, 0.3668, 0.6990, 0.5320], [0.5642, 0.3640, 0.8545, 0.4413, 0.4406, 0.5011, 0.5789, 0.5432], [0.5409, 0.3582, 0.8844, 0.4137, 0.3971, 0.3509, 0.6847, 0.5424], [0.5854, 0.3841, 0.8754, 0.4690, 0.3467, 0.4304, 0.5603, 0.5154], [0.5621, 0.3721, 0.7616, 0.2443, 0.3572, 0.2442, 0.5230, 0.5137], [0.5795, 0.3681, 0.8496, 0.5353, 0.3407, 0.3531, 0.5610, 0.4714], [0.5794, 0.3526, 0.7994, 0.5492, 0.3817, 0.4437, 0.5184, 0.5368]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947], [0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494], [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6062, 0.5567], [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841], [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200], [0.6204, 0.4049, 0.7975, 0.2700, 0.3938, 0.2567, 0.5700, 0.5183], [0.6272, 0.4071, 0.8737, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6222, 0.3937, 0.8350, 0.5617, 0.4137, 0.4600, 0.5800, 0.5233]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.04967287927865982 step: 48 running loss: 0.0010348516516387463 Train Steps: 48/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071], [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097], [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117], [0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136], [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114], [0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6262, 0.5367], [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5932, 0.3752, 0.7299, 0.1961, 0.3496, 0.2364, 0.5796, 0.4742], [0.5746, 0.3886, 0.8618, 0.4477, 0.4209, 0.5416, 0.5816, 0.5256], [0.5612, 0.3511, 0.7895, 0.2552, 0.3983, 0.1881, 0.5990, 0.5168], [0.5476, 0.3570, 0.7310, 0.1807, 0.3870, 0.1791, 0.5720, 0.4853], [0.5533, 0.3635, 0.7871, 0.3705, 0.3497, 0.4683, 0.5949, 0.5131], [0.5736, 0.3728, 0.8460, 0.5015, 0.4372, 0.4897, 0.6130, 0.5065], [0.5700, 0.3722, 0.8619, 0.4676, 0.4099, 0.5433, 0.6251, 0.5171], [0.5712, 0.3966, 0.8282, 0.5571, 0.3881, 0.3772, 0.5808, 0.5863]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6163, 0.4114, 0.7650, 0.2017, 0.3762, 0.2867, 0.5631, 0.5071], [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097], [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5788, 0.5117], [0.6132, 0.4118, 0.8200, 0.3633, 0.3562, 0.5400, 0.5787, 0.5136], [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114], [0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6263, 0.5367], [0.6212, 0.4159, 0.8675, 0.5783, 0.4087, 0.4317, 0.5612, 0.5917]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.05076544382609427 step: 49 running loss: 0.0010360294658386586 Train Steps: 49/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6058, 0.3986, 0.8324, 0.4626, 0.3838, 0.4983, 0.5147, 0.5466], [0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [0.6040, 0.4002, 0.7338, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778], [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6353, 0.4128, 0.8488, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5817, 0.3841, 0.8039, 0.4701, 0.3774, 0.5051, 0.5660, 0.5194], [0.4629, 0.3292, 0.7094, 0.2834, 0.4573, 0.1952, 0.5821, 0.5922], [0.5646, 0.3773, 0.7173, 0.1904, 0.3819, 0.2052, 0.5569, 0.4717], [0.5157, 0.3567, 0.7806, 0.2952, 0.3890, 0.2561, 0.5382, 0.5372], [0.6007, 0.4074, 0.8742, 0.4553, 0.4202, 0.5225, 0.6016, 0.5305], [0.4842, 0.3445, 0.8362, 0.2285, 0.5071, 0.1936, 0.6952, 0.5239], [0.5824, 0.3880, 0.8562, 0.4701, 0.3626, 0.5318, 0.6304, 0.4917], [0.5742, 0.3909, 0.7183, 0.1900, 0.3832, 0.2580, 0.6323, 0.5240]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6058, 0.3986, 0.8324, 0.4626, 0.3837, 0.4983, 0.5147, 0.5466], [0.6270, 0.4266, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [0.6040, 0.4002, 0.7337, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778], [0.6182, 0.4099, 0.7812, 0.3000, 0.3938, 0.2367, 0.5325, 0.5750], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6353, 0.4128, 0.8487, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483]]], device='cuda:0') loss_train_step before backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train: 0.05246755573898554 step: 50 running loss: 0.0010493511147797107 Train Steps: 50/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550], [0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823], [0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5563, 0.5533], [ nan, nan, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650], [0.6224, 0.4061, 0.8988, 0.4300, 0.3838, 0.4750, 0.6112, 0.5483], [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6535, 0.4426, 0.8615, 0.4605, 0.4753, 0.5020, 0.6464, 0.5205], [0.6618, 0.4507, 0.7415, 0.2485, 0.4715, 0.1504, 0.6295, 0.5336], [0.6118, 0.4104, 0.7157, 0.1890, 0.3934, 0.1675, 0.5420, 0.4826], [0.6650, 0.4469, 0.8542, 0.5318, 0.3817, 0.4565, 0.5792, 0.5371], [0.0827, 0.0845, 0.7178, 0.2095, 0.4269, 0.2542, 0.5514, 0.5404], [0.7079, 0.4768, 0.8986, 0.4412, 0.3717, 0.4759, 0.6466, 0.5368], [0.1929, 0.1621, 0.7625, 0.2185, 0.3761, 0.2935, 0.6065, 0.5570], [0.7233, 0.4964, 0.8791, 0.4836, 0.4390, 0.5540, 0.6191, 0.5469]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550], [0.6245, 0.4100, 0.7763, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6055, 0.4015, 0.7425, 0.2033, 0.4112, 0.1883, 0.5217, 0.4823], [0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5562, 0.5533], [0.0000, 0.0000, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650], [0.6224, 0.4061, 0.8988, 0.4300, 0.3837, 0.4750, 0.6112, 0.5483], [0.0000, 0.0000, 0.7525, 0.2291, 0.3837, 0.3017, 0.6050, 0.5667], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517]]], device='cuda:0') loss_train_step before backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train: 0.05460604024119675 step: 51 running loss: 0.0010707066713960148 Train Steps: 51/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6087, 0.3951, 0.8387, 0.5833, 0.4188, 0.4933, 0.5146, 0.4830], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810], [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533], [0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5737, 0.5700]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5814, 0.3786, 0.8285, 0.5403, 0.4182, 0.5051, 0.5368, 0.5026], [0.6062, 0.3988, 0.8697, 0.4950, 0.3917, 0.3794, 0.5435, 0.5671], [0.6641, 0.4272, 0.8921, 0.3767, 0.3677, 0.3942, 0.6647, 0.4933], [0.6181, 0.4069, 0.8375, 0.2639, 0.4191, 0.2653, 0.6480, 0.4718], [0.5773, 0.3799, 0.8746, 0.4581, 0.5086, 0.5067, 0.5284, 0.5730], [0.5787, 0.4009, 0.7336, 0.2481, 0.4412, 0.2274, 0.5866, 0.5746], [0.5987, 0.4005, 0.8599, 0.4030, 0.3679, 0.3575, 0.6200, 0.5459], [0.6027, 0.3978, 0.8663, 0.5080, 0.4044, 0.5205, 0.6093, 0.5446]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6087, 0.3951, 0.8388, 0.5833, 0.4187, 0.4933, 0.5146, 0.4830], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5412, 0.5683], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810], [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6252, 0.4162, 0.8662, 0.4467, 0.3625, 0.3567, 0.6037, 0.5533], [0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5738, 0.5700]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.0550856547197327 step: 52 running loss: 0.0010593395138410134 Train Steps: 52/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064], [ nan, nan, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702], [0.6115, 0.4005, 0.8838, 0.3867, 0.3763, 0.4700, 0.5800, 0.5550], [0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683], [0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967], [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5737, 0.5350], [0.6199, 0.4093, 0.7913, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700], [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6195, 0.4219, 0.8560, 0.4903, 0.5090, 0.5246, 0.5346, 0.5340], [0.0739, 0.0713, 0.8875, 0.3262, 0.5059, 0.2328, 0.6973, 0.5713], [0.6520, 0.4337, 0.8578, 0.3723, 0.4002, 0.4738, 0.5595, 0.5606], [0.6410, 0.4396, 0.8600, 0.3413, 0.3624, 0.3962, 0.5996, 0.5702], [0.6282, 0.4127, 0.8733, 0.4211, 0.4223, 0.2760, 0.6183, 0.5145], [0.6489, 0.4297, 0.8637, 0.4640, 0.4111, 0.3077, 0.5486, 0.5505], [0.6476, 0.4493, 0.7594, 0.2379, 0.4411, 0.2562, 0.5998, 0.5810], [0.6860, 0.4447, 0.8223, 0.4055, 0.3957, 0.4801, 0.5548, 0.5080]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6128, 0.4022, 0.8737, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064], [0.0000, 0.0000, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702], [0.6115, 0.4005, 0.8838, 0.3867, 0.3762, 0.4700, 0.5800, 0.5550], [0.6219, 0.4097, 0.8737, 0.3400, 0.3562, 0.4117, 0.5975, 0.5683], [0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967], [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5738, 0.5350], [0.6198, 0.4093, 0.7912, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700], [0.6084, 0.4005, 0.8400, 0.4317, 0.3762, 0.4750, 0.5476, 0.5058]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.05573676276253536 step: 53 running loss: 0.0010516370332553841 Train Steps: 53/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301], [0.6263, 0.4030, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771], [0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6075, 0.4007, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999], [0.6203, 0.4078, 0.8800, 0.5083, 0.3900, 0.5000, 0.6100, 0.5583]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6280, 0.4024, 0.9012, 0.5148, 0.4555, 0.4977, 0.6189, 0.5503], [0.5665, 0.3724, 0.9334, 0.4956, 0.4231, 0.5474, 0.6169, 0.5207], [0.6607, 0.4436, 0.7319, 0.2175, 0.4085, 0.2458, 0.5786, 0.5611], [0.6557, 0.4170, 0.9130, 0.4984, 0.4010, 0.4071, 0.5816, 0.5355], [0.6025, 0.3909, 0.8496, 0.5025, 0.4549, 0.5250, 0.5244, 0.5597], [0.6019, 0.3968, 0.8792, 0.5568, 0.4240, 0.4326, 0.5877, 0.5582], [0.6882, 0.4499, 0.8793, 0.4420, 0.4445, 0.5120, 0.5330, 0.5221], [0.6171, 0.4100, 0.9059, 0.5125, 0.4226, 0.5082, 0.5894, 0.5972]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301], [0.6263, 0.4029, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771], [0.6187, 0.4103, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6236, 0.3965, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6075, 0.4006, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6068, 0.3963, 0.8650, 0.4317, 0.4038, 0.5083, 0.5253, 0.4999], [0.6203, 0.4078, 0.8800, 0.5083, 0.3900, 0.5000, 0.6100, 0.5583]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.056453864090144634 step: 54 running loss: 0.001045441927595271 Train Steps: 54/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207], [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.6129, 0.4063, 0.8738, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874], [0.6152, 0.4131, 0.6863, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5476, 0.3369, 0.8878, 0.5874, 0.4094, 0.4444, 0.5650, 0.5610], [0.6466, 0.4018, 0.8539, 0.2829, 0.4648, 0.2358, 0.6455, 0.5369], [0.5906, 0.3769, 0.9650, 0.4317, 0.3911, 0.3305, 0.6028, 0.5531], [0.5458, 0.3493, 0.9107, 0.5839, 0.4247, 0.4551, 0.5193, 0.5251], [0.6134, 0.3893, 0.9136, 0.5638, 0.4533, 0.5064, 0.5083, 0.6175], [0.6576, 0.4241, 0.7325, 0.2693, 0.3812, 0.3485, 0.5727, 0.5774], [0.6544, 0.4179, 0.7269, 0.2816, 0.4004, 0.2486, 0.5374, 0.5922], [0.6109, 0.3914, 0.7628, 0.2722, 0.4018, 0.2721, 0.5324, 0.5373]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207], [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.6130, 0.4063, 0.8737, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874], [0.6152, 0.4131, 0.6862, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.057424455240834504 step: 55 running loss: 0.0010440810043788093 Train Steps: 55/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [ nan, nan, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983], [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926], [0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133], [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220], [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6488, 0.3997, 0.9324, 0.4416, 0.4320, 0.3891, 0.6941, 0.6133], [0.0758, 0.0263, 0.7338, 0.2581, 0.4481, 0.1622, 0.5430, 0.5465], [0.7226, 0.4630, 0.7617, 0.2752, 0.3950, 0.3043, 0.5740, 0.5084], [0.6153, 0.3819, 0.8446, 0.3057, 0.4062, 0.2827, 0.5835, 0.5420], [0.5881, 0.3800, 0.8620, 0.4081, 0.3653, 0.3207, 0.4853, 0.5845], [0.6604, 0.4283, 0.7799, 0.3036, 0.4455, 0.1807, 0.5666, 0.5509], [0.5751, 0.3707, 0.7140, 0.2635, 0.4521, 0.1633, 0.5350, 0.5606], [0.6710, 0.4345, 0.8234, 0.4160, 0.3584, 0.4337, 0.5220, 0.5283]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.0000, 0.0000, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983], [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926], [0.6205, 0.4015, 0.8350, 0.2717, 0.3988, 0.2550, 0.5788, 0.5133], [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220], [0.6076, 0.3952, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.05837335210526362 step: 56 running loss: 0.0010423812875939933 Train Steps: 56/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850], [0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748], [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5535, 0.3603, 0.9047, 0.4861, 0.4687, 0.5778, 0.5736, 0.5600], [0.5951, 0.3766, 0.8869, 0.5239, 0.3845, 0.3401, 0.6707, 0.5305], [0.6612, 0.4110, 0.8894, 0.3805, 0.3492, 0.4530, 0.6655, 0.4991], [0.6235, 0.4039, 0.6751, 0.3450, 0.3461, 0.2827, 0.4931, 0.5618], [0.5802, 0.3689, 0.7180, 0.2602, 0.3880, 0.2208, 0.5133, 0.5405], [0.6408, 0.4002, 0.8874, 0.3405, 0.3949, 0.4310, 0.6676, 0.5227], [0.5665, 0.3689, 0.8803, 0.5668, 0.3373, 0.3560, 0.5179, 0.5180], [0.5994, 0.3896, 0.8057, 0.2313, 0.4689, 0.1243, 0.6043, 0.4866]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6184, 0.4079, 0.8838, 0.4617, 0.4837, 0.5650, 0.6175, 0.5850], [0.6336, 0.4191, 0.8938, 0.5167, 0.3938, 0.3517, 0.7343, 0.5748], [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.059197709371801466 step: 57 running loss: 0.0010385563047684467 Train Steps: 57/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6581, 0.4206, 0.8478, 0.3237, 0.3369, 0.3658, 0.5894, 0.4957], [0.6343, 0.4063, 0.8662, 0.5718, 0.3784, 0.4163, 0.5256, 0.5283], [0.6484, 0.4191, 0.8517, 0.5362, 0.3857, 0.5657, 0.6739, 0.5407], [0.6023, 0.3653, 0.8979, 0.5239, 0.3713, 0.5414, 0.6793, 0.5288], [0.6106, 0.3916, 0.8149, 0.5506, 0.3792, 0.4707, 0.6510, 0.5134], [0.6310, 0.4057, 0.8812, 0.4616, 0.4065, 0.4958, 0.5455, 0.5182], [0.6720, 0.4215, 0.7415, 0.2647, 0.3884, 0.1843, 0.5366, 0.5280], [0.6216, 0.3953, 0.8320, 0.5320, 0.3525, 0.4859, 0.6587, 0.5009]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6193, 0.4050, 0.7312, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6289, 0.4019, 0.8112, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.05994822469074279 step: 58 running loss: 0.0010335900808748756 Train Steps: 58/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6261, 0.4029, 0.8720, 0.3364, 0.3665, 0.3753, 0.6531, 0.5183], [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083], [0.6199, 0.4093, 0.7913, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700], [0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799], [0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233], [0.6259, 0.4133, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019], [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6293, 0.4008, 0.8419, 0.3521, 0.3527, 0.3801, 0.6642, 0.4922], [0.6315, 0.4160, 0.8540, 0.4946, 0.3421, 0.4459, 0.6039, 0.5588], [0.5937, 0.4044, 0.7634, 0.2575, 0.3904, 0.2473, 0.6448, 0.5504], [0.6096, 0.3997, 0.8519, 0.3804, 0.3719, 0.3151, 0.6080, 0.5556], [0.5817, 0.3542, 0.8235, 0.5666, 0.3866, 0.4522, 0.6036, 0.5165], [0.6386, 0.4336, 0.8146, 0.2436, 0.4702, 0.1479, 0.6958, 0.4712], [0.5744, 0.3767, 0.8577, 0.4240, 0.3493, 0.5482, 0.5868, 0.4898], [0.6189, 0.3986, 0.8186, 0.3587, 0.3306, 0.2866, 0.5438, 0.5318]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6261, 0.4029, 0.8720, 0.3364, 0.3665, 0.3753, 0.6531, 0.5183], [0.6186, 0.4060, 0.8750, 0.5050, 0.3537, 0.4367, 0.5813, 0.6083], [0.6198, 0.4093, 0.7912, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700], [0.6250, 0.4236, 0.8637, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799], [0.6222, 0.3937, 0.8350, 0.5617, 0.4137, 0.4600, 0.5800, 0.5233], [0.6259, 0.4132, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6161, 0.4099, 0.8737, 0.4383, 0.3787, 0.5483, 0.5605, 0.5019], [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.060535903030540794 step: 59 running loss: 0.0010260322547549288 Train Steps: 59/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4102, 0.7288, 0.2417, 0.4150, 0.2383, 0.6100, 0.5500], [0.6115, 0.4081, 0.6725, 0.2433, 0.4088, 0.1933, 0.5167, 0.5544], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583], [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367], [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6403, 0.4280, 0.7446, 0.2610, 0.3933, 0.2408, 0.6553, 0.5600], [0.6543, 0.4360, 0.6836, 0.2661, 0.4094, 0.1976, 0.5499, 0.5387], [0.6499, 0.4227, 0.7998, 0.2759, 0.4333, 0.2275, 0.6982, 0.5251], [0.5878, 0.3954, 0.7473, 0.3859, 0.3282, 0.3870, 0.5789, 0.5672], [0.5764, 0.3836, 0.8764, 0.5084, 0.3964, 0.4931, 0.5974, 0.5081], [0.5695, 0.3758, 0.8933, 0.4360, 0.3550, 0.4061, 0.6115, 0.5394], [0.6436, 0.4339, 0.7857, 0.2777, 0.3960, 0.2413, 0.6778, 0.5352], [0.5729, 0.3854, 0.8123, 0.3846, 0.3252, 0.4032, 0.5814, 0.4950]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6201, 0.4102, 0.7287, 0.2417, 0.4150, 0.2383, 0.6100, 0.5500], [0.6115, 0.4081, 0.6725, 0.2433, 0.4087, 0.1933, 0.5167, 0.5544], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6170, 0.4102, 0.7468, 0.3695, 0.3462, 0.3767, 0.5238, 0.5823], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6212, 0.4033, 0.8938, 0.4167, 0.3812, 0.4267, 0.5612, 0.5583], [0.6204, 0.4110, 0.7912, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367], [0.6076, 0.3952, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.061192915483843535 step: 60 running loss: 0.0010198819247307256 Train Steps: 60/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494], [0.6250, 0.4146, 0.8838, 0.3933, 0.3588, 0.4283, 0.6162, 0.5367], [0.6087, 0.3976, 0.8337, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300], [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407], [0.6289, 0.4081, 0.8720, 0.3487, 0.3900, 0.3183, 0.6703, 0.5376]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6098, 0.4095, 0.8948, 0.3430, 0.4369, 0.3946, 0.7409, 0.5581], [0.5593, 0.3761, 0.8257, 0.3682, 0.3557, 0.4102, 0.6401, 0.5402], [0.5868, 0.4028, 0.8061, 0.3714, 0.3707, 0.3365, 0.6015, 0.5382], [0.6645, 0.4425, 0.7266, 0.1562, 0.3846, 0.2444, 0.5894, 0.5023], [0.5956, 0.4105, 0.8264, 0.5500, 0.3945, 0.4470, 0.5775, 0.5710], [0.6154, 0.4200, 0.7679, 0.2885, 0.4765, 0.1840, 0.6045, 0.5576], [0.5763, 0.3963, 0.8310, 0.3714, 0.3726, 0.3122, 0.5121, 0.5635], [0.5984, 0.3993, 0.8475, 0.3177, 0.3888, 0.2935, 0.6982, 0.5648]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494], [0.6250, 0.4146, 0.8838, 0.3933, 0.3587, 0.4283, 0.6162, 0.5367], [0.6087, 0.3976, 0.8338, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300], [0.6165, 0.4106, 0.7575, 0.1733, 0.3837, 0.2650, 0.5680, 0.5116], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6229, 0.4107, 0.8138, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6133, 0.4094, 0.8495, 0.4028, 0.3587, 0.3200, 0.5003, 0.5407], [0.6289, 0.4081, 0.8720, 0.3487, 0.3900, 0.3183, 0.6703, 0.5376]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.06174909329274669 step: 61 running loss: 0.0010122802179138803 Train Steps: 61/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833], [0.6200, 0.4070, 0.8938, 0.4183, 0.3538, 0.4567, 0.6175, 0.5400], [ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933], [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6732, 0.4678, 0.8815, 0.4820, 0.3907, 0.4533, 0.6017, 0.6037], [0.6412, 0.4282, 0.7319, 0.2063, 0.4344, 0.2330, 0.5760, 0.5694], [0.6649, 0.4414, 0.8627, 0.5376, 0.4606, 0.4971, 0.6258, 0.5521], [0.7022, 0.4611, 0.8948, 0.4707, 0.3917, 0.5046, 0.6166, 0.4972], [0.7338, 0.4807, 0.8952, 0.3996, 0.3814, 0.4844, 0.6488, 0.5259], [0.0807, 0.0648, 0.6899, 0.2234, 0.4330, 0.2029, 0.5665, 0.5873], [0.6507, 0.4482, 0.7274, 0.2220, 0.3836, 0.3474, 0.6153, 0.5847], [0.7269, 0.4700, 0.9090, 0.4028, 0.3918, 0.3560, 0.6513, 0.5766]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6186, 0.4060, 0.8750, 0.5050, 0.3537, 0.4367, 0.5813, 0.6083], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6229, 0.4066, 0.8512, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6037, 0.4833], [0.6200, 0.4070, 0.8938, 0.4183, 0.3537, 0.4567, 0.6175, 0.5400], [0.0000, 0.0000, 0.6688, 0.2513, 0.4112, 0.2117, 0.5193, 0.5933], [0.6132, 0.4066, 0.7259, 0.2402, 0.3587, 0.3300, 0.6000, 0.5600], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.06322128738975152 step: 62 running loss: 0.0010196981837056697 Train Steps: 62/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528], [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133], [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6324, 0.4039, 0.8685, 0.4872, 0.4011, 0.4443, 0.4893, 0.5842], [0.6219, 0.4002, 0.7847, 0.1992, 0.4461, 0.2610, 0.6476, 0.5553], [0.6258, 0.4004, 0.8205, 0.4090, 0.3847, 0.4614, 0.5261, 0.5988], [0.6113, 0.3853, 0.8718, 0.4939, 0.4130, 0.5536, 0.7076, 0.5905], [0.6323, 0.4041, 0.8950, 0.3352, 0.4317, 0.3551, 0.7193, 0.5770], [0.6194, 0.4062, 0.8558, 0.5051, 0.3982, 0.4593, 0.6008, 0.5352], [0.6342, 0.4045, 0.8060, 0.2035, 0.4183, 0.2695, 0.5965, 0.5183], [0.6079, 0.3940, 0.8847, 0.4094, 0.4375, 0.3078, 0.6545, 0.5843]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6332, 0.4165, 0.9100, 0.3350, 0.4187, 0.3683, 0.7438, 0.5528], [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133], [0.6257, 0.4034, 0.8288, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.06367821450112388 step: 63 running loss: 0.0010107653095416489 Train Steps: 63/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6126, 0.4039, 0.8237, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138], [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986], [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6179, 0.4008, 0.8600, 0.4015, 0.3932, 0.2515, 0.5711, 0.5438], [0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550], [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114], [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6023, 0.3895, 0.8220, 0.3872, 0.3741, 0.3751, 0.5697, 0.6120], [0.5667, 0.3706, 0.7275, 0.2095, 0.4766, 0.2035, 0.5421, 0.6021], [0.5960, 0.3922, 0.9036, 0.3543, 0.4618, 0.2125, 0.6223, 0.5303], [0.6050, 0.3791, 0.8262, 0.2050, 0.4125, 0.2759, 0.6123, 0.5004], [0.5788, 0.3780, 0.8298, 0.3318, 0.4147, 0.2553, 0.5634, 0.5648], [0.5880, 0.3790, 0.8971, 0.4535, 0.3892, 0.4829, 0.5863, 0.5748], [0.5920, 0.3777, 0.8626, 0.4950, 0.4768, 0.5564, 0.5932, 0.5270], [0.5899, 0.3935, 0.7271, 0.1832, 0.4383, 0.2355, 0.6159, 0.5777]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6126, 0.4038, 0.8238, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138], [0.6218, 0.4185, 0.7337, 0.2650, 0.4625, 0.1950, 0.5688, 0.5800], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986], [0.6257, 0.4034, 0.8288, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6179, 0.4008, 0.8600, 0.4015, 0.3932, 0.2515, 0.5711, 0.5438], [0.6197, 0.4091, 0.8800, 0.4783, 0.3537, 0.4767, 0.5950, 0.5550], [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114], [0.6200, 0.4071, 0.7337, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.06423556891968474 step: 64 running loss: 0.001003680764370074 Train Steps: 64/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118], [0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617], [0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.0176, 0.0051, 0.7073, 0.2451, 0.3940, 0.2041, 0.5673, 0.5674], [0.7240, 0.4591, 0.8006, 0.2585, 0.3777, 0.3049, 0.6131, 0.5358], [0.6416, 0.4175, 0.8055, 0.3550, 0.3874, 0.3014, 0.5880, 0.5704], [0.6756, 0.4380, 0.8862, 0.3885, 0.3845, 0.4889, 0.5990, 0.5338], [0.6557, 0.4121, 0.8220, 0.2495, 0.4030, 0.2284, 0.5946, 0.5053], [0.6400, 0.4365, 0.7531, 0.3356, 0.4336, 0.2066, 0.5468, 0.6130], [0.6865, 0.4423, 0.8561, 0.5533, 0.4550, 0.4492, 0.5814, 0.5617], [0.6215, 0.4045, 0.7008, 0.1796, 0.4043, 0.2497, 0.5549, 0.5109]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6212, 0.4171, 0.7875, 0.3633, 0.3812, 0.2933, 0.5675, 0.5700], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118], [0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617], [0.6043, 0.4022, 0.6888, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.06490114069310948 step: 65 running loss: 0.0009984790875862998 Train Steps: 65/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6115, 0.4081, 0.6725, 0.2433, 0.4088, 0.1933, 0.5167, 0.5544], [ nan, nan, 0.8888, 0.3100, 0.5262, 0.2817, 0.7145, 0.6003], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6208, 0.4082, 0.8538, 0.3067, 0.3588, 0.3717, 0.6112, 0.5517], [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162], [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6413, 0.4297, 0.6962, 0.2717, 0.4131, 0.2019, 0.4723, 0.5475], [0.0419, 0.0319, 0.9071, 0.3396, 0.4849, 0.2528, 0.7129, 0.5819], [0.6728, 0.4302, 0.8161, 0.2820, 0.4344, 0.2315, 0.6006, 0.5320], [0.6691, 0.4475, 0.8545, 0.3254, 0.3570, 0.3591, 0.5835, 0.5511], [0.6772, 0.4555, 0.8836, 0.5338, 0.3839, 0.5459, 0.7248, 0.5521], [0.6453, 0.4206, 0.9209, 0.4473, 0.3538, 0.4860, 0.6285, 0.5191], [0.6039, 0.3915, 0.7655, 0.3256, 0.3388, 0.3197, 0.5228, 0.5558], [0.5965, 0.4058, 0.7044, 0.2233, 0.4040, 0.1886, 0.5195, 0.5606]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6115, 0.4081, 0.6725, 0.2433, 0.4087, 0.1933, 0.5167, 0.5544], [0.0000, 0.0000, 0.8888, 0.3100, 0.5263, 0.2817, 0.7145, 0.6003], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6208, 0.4082, 0.8537, 0.3067, 0.3587, 0.3717, 0.6112, 0.5517], [0.6339, 0.4123, 0.8637, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162], [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.06546252308180556 step: 66 running loss: 0.0009918564103303872 Train Steps: 66/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166], [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519], [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5988, 0.5700], [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373], [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967], [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097], [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6000, 0.3890, 0.8598, 0.5231, 0.3860, 0.4836, 0.6604, 0.5219], [0.5418, 0.3529, 0.8517, 0.5108, 0.4181, 0.4930, 0.5155, 0.5245], [0.5073, 0.3391, 0.7839, 0.2833, 0.3466, 0.3270, 0.5728, 0.5695], [0.6115, 0.4022, 0.8242, 0.5736, 0.3559, 0.4262, 0.5729, 0.5374], [0.6262, 0.4260, 0.8762, 0.5047, 0.3387, 0.2864, 0.6100, 0.5022], [0.5922, 0.3897, 0.8660, 0.4663, 0.4117, 0.4919, 0.5157, 0.4979], [0.5938, 0.3825, 0.8493, 0.2560, 0.4352, 0.2088, 0.6199, 0.4986], [0.4856, 0.3118, 0.8906, 0.3083, 0.3727, 0.3904, 0.6850, 0.5458]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166], [0.6100, 0.4016, 0.8600, 0.5067, 0.4613, 0.5233, 0.5086, 0.5519], [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5987, 0.5700], [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373], [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967], [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097], [0.6275, 0.4111, 0.8462, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.06678534619277343 step: 67 running loss: 0.0009967962118324393 Train Steps: 67/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6136, 0.4117, 0.8700, 0.5167, 0.4188, 0.5083, 0.5147, 0.5495], [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167], [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500], [0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300], [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431], [ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383], [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261], [0.6099, 0.4030, 0.8638, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6511, 0.4257, 0.8758, 0.5555, 0.3867, 0.4742, 0.5685, 0.5145], [0.6889, 0.4593, 0.7985, 0.3512, 0.3446, 0.3632, 0.6308, 0.5870], [0.6273, 0.4013, 0.8833, 0.4678, 0.3919, 0.5013, 0.6203, 0.5304], [0.5450, 0.3481, 0.7664, 0.2362, 0.3711, 0.2955, 0.6387, 0.5304], [0.6303, 0.4170, 0.8735, 0.5181, 0.4061, 0.4715, 0.5485, 0.4984], [0.0494, 0.0384, 0.7164, 0.2535, 0.3838, 0.1804, 0.5373, 0.5331], [0.6281, 0.4168, 0.8878, 0.5096, 0.3740, 0.4657, 0.5807, 0.5154], [0.6367, 0.4216, 0.8746, 0.5444, 0.4802, 0.4629, 0.5561, 0.5284]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6136, 0.4117, 0.8700, 0.5167, 0.4187, 0.5083, 0.5147, 0.5495], [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6037, 0.6167], [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5838, 0.5500], [0.6207, 0.4081, 0.7663, 0.2067, 0.3963, 0.3200, 0.6313, 0.5300], [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431], [0.0000, 0.0000, 0.7268, 0.2333, 0.4125, 0.1933, 0.5113, 0.5383], [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261], [0.6098, 0.4030, 0.8637, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.06771578977350146 step: 68 running loss: 0.0009958204378456097 Train Steps: 68/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413], [0.6160, 0.4093, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808], [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533], [0.6164, 0.4102, 0.8850, 0.4867, 0.4213, 0.5983, 0.5609, 0.5038], [0.6201, 0.4151, 0.8588, 0.5467, 0.3700, 0.3950, 0.5637, 0.5933]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5082, 0.3207, 0.9109, 0.3394, 0.5157, 0.2867, 0.7239, 0.5230], [0.5908, 0.3902, 0.8460, 0.4669, 0.3775, 0.4759, 0.5441, 0.5613], [0.6483, 0.4339, 0.8936, 0.5021, 0.3721, 0.4222, 0.5393, 0.5726], [0.5676, 0.3806, 0.8563, 0.3392, 0.3757, 0.3146, 0.5798, 0.5350], [0.5690, 0.3795, 0.6955, 0.2171, 0.4206, 0.2327, 0.5407, 0.5215], [0.5241, 0.3457, 0.7194, 0.2207, 0.3837, 0.2938, 0.5650, 0.5470], [0.5168, 0.3465, 0.8857, 0.5028, 0.4674, 0.6062, 0.6085, 0.5349], [0.6198, 0.4226, 0.8568, 0.5667, 0.3677, 0.4148, 0.5821, 0.5672]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413], [0.6160, 0.4092, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808], [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6054, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6114, 0.4018, 0.7212, 0.1967, 0.3762, 0.2700, 0.5875, 0.5533], [0.6164, 0.4102, 0.8850, 0.4867, 0.4212, 0.5983, 0.5609, 0.5038], [0.6202, 0.4151, 0.8587, 0.5467, 0.3700, 0.3950, 0.5638, 0.5933]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.06889769760891795 step: 69 running loss: 0.0009985173566509848 Train Steps: 69/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6185, 0.4042, 0.7700, 0.2250, 0.4062, 0.2117, 0.5763, 0.5150], [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6063, 0.6083], [0.6250, 0.4146, 0.8838, 0.3933, 0.3588, 0.4283, 0.6162, 0.5367], [ nan, nan, 0.7192, 0.2346, 0.4037, 0.2050, 0.5138, 0.5650], [0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5887, 0.5017], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5883, 0.3851, 0.8589, 0.3032, 0.5168, 0.2417, 0.6563, 0.5515], [0.5297, 0.3508, 0.7547, 0.2512, 0.4262, 0.2454, 0.5404, 0.5347], [0.6445, 0.4419, 0.8571, 0.3767, 0.3818, 0.4626, 0.6121, 0.5993], [0.6430, 0.4185, 0.8573, 0.4084, 0.3759, 0.4354, 0.6020, 0.5308], [0.0117, 0.0080, 0.7005, 0.2403, 0.4421, 0.2541, 0.4979, 0.5550], [0.6239, 0.4059, 0.8724, 0.4794, 0.3975, 0.5280, 0.5794, 0.5216], [0.6547, 0.4375, 0.7912, 0.3228, 0.3815, 0.4045, 0.5427, 0.5234], [0.6379, 0.4056, 0.8605, 0.3202, 0.4096, 0.3092, 0.6162, 0.5334]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6184, 0.4042, 0.7700, 0.2250, 0.4062, 0.2117, 0.5763, 0.5150], [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6062, 0.6083], [0.6250, 0.4146, 0.8838, 0.3933, 0.3587, 0.4283, 0.6162, 0.5367], [0.0000, 0.0000, 0.7192, 0.2346, 0.4038, 0.2050, 0.5138, 0.5650], [0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5888, 0.5017], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.06951731926528737 step: 70 running loss: 0.0009931045609326767 Train Steps: 70/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533], [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773], [0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471], [0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164], [ nan, nan, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183], [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.1622, 0.1066, 0.7874, 0.3014, 0.3781, 0.3310, 0.5317, 0.5481], [0.6755, 0.4376, 0.8550, 0.4936, 0.4995, 0.5143, 0.4808, 0.5582], [0.6905, 0.4685, 0.8566, 0.4401, 0.3896, 0.4463, 0.4683, 0.5373], [0.7136, 0.4800, 0.8401, 0.5447, 0.4031, 0.3674, 0.5622, 0.5945], [0.0378, 0.0286, 0.8293, 0.2527, 0.5296, 0.2630, 0.6928, 0.5580], [0.6766, 0.4412, 0.8797, 0.4461, 0.4491, 0.5980, 0.6137, 0.5414], [0.6949, 0.4499, 0.8717, 0.4481, 0.4151, 0.5993, 0.6216, 0.5185], [0.6602, 0.4530, 0.9019, 0.4224, 0.4236, 0.3748, 0.6915, 0.5616]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533], [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773], [0.6127, 0.4084, 0.8700, 0.4467, 0.3988, 0.4317, 0.5013, 0.5471], [0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164], [0.0000, 0.0000, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183], [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.07116618793224916 step: 71 running loss: 0.0010023406751021007 Train Steps: 71/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412], [0.6224, 0.4061, 0.8988, 0.4300, 0.3838, 0.4750, 0.6112, 0.5483], [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882], [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804], [0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967], [0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5637, 0.5083], [0.6199, 0.4071, 0.7600, 0.2117, 0.4037, 0.2767, 0.6138, 0.5550]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6020, 0.3793, 0.8858, 0.3604, 0.3895, 0.2816, 0.5661, 0.5512], [0.5487, 0.3592, 0.8974, 0.4165, 0.3968, 0.4994, 0.6111, 0.5781], [0.5426, 0.3551, 0.7878, 0.2664, 0.3672, 0.4156, 0.6022, 0.5935], [0.5907, 0.3904, 0.8766, 0.3900, 0.3722, 0.4003, 0.5070, 0.5364], [0.5848, 0.3734, 0.8911, 0.3657, 0.3688, 0.3688, 0.6141, 0.5162], [0.5836, 0.3809, 0.8993, 0.4411, 0.4129, 0.2957, 0.6122, 0.5266], [0.5586, 0.3550, 0.8771, 0.5170, 0.4175, 0.4976, 0.5237, 0.5390], [0.6279, 0.4028, 0.7860, 0.2222, 0.4409, 0.3015, 0.6051, 0.5704]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412], [0.6224, 0.4061, 0.8988, 0.4300, 0.3837, 0.4750, 0.6112, 0.5483], [0.6197, 0.4051, 0.7812, 0.2650, 0.3512, 0.4050, 0.6112, 0.5500], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882], [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804], [0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967], [0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5638, 0.5083], [0.6199, 0.4071, 0.7600, 0.2117, 0.4038, 0.2767, 0.6137, 0.5550]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.07192522671539336 step: 72 running loss: 0.000998961482158241 Train Steps: 72/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750], [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6224, 0.4061, 0.8988, 0.4300, 0.3838, 0.4750, 0.6112, 0.5483], [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6208, 0.4082, 0.8538, 0.3067, 0.3588, 0.3717, 0.6112, 0.5517], [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600], [0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5656, 0.3701, 0.8129, 0.3049, 0.4055, 0.2572, 0.5526, 0.5638], [0.5737, 0.3879, 0.9114, 0.5173, 0.3807, 0.4555, 0.5800, 0.5659], [0.5693, 0.3739, 0.9095, 0.4160, 0.3864, 0.4838, 0.6312, 0.5468], [0.5836, 0.3736, 0.8785, 0.4841, 0.4992, 0.5042, 0.5113, 0.5590], [0.5832, 0.3820, 0.8750, 0.5604, 0.4101, 0.4439, 0.5613, 0.5650], [0.5447, 0.3638, 0.8757, 0.3092, 0.3734, 0.3732, 0.6159, 0.5459], [0.5780, 0.3907, 0.7475, 0.2412, 0.3644, 0.3376, 0.6089, 0.5725], [0.6014, 0.3843, 0.9054, 0.4120, 0.3852, 0.4092, 0.5635, 0.5452]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6182, 0.4099, 0.7812, 0.3000, 0.3938, 0.2367, 0.5325, 0.5750], [0.6222, 0.4171, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6224, 0.4061, 0.8988, 0.4300, 0.3837, 0.4750, 0.6112, 0.5483], [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6208, 0.4082, 0.8537, 0.3067, 0.3587, 0.3717, 0.6112, 0.5517], [0.6132, 0.4066, 0.7259, 0.2402, 0.3587, 0.3300, 0.6000, 0.5600], [0.6212, 0.4033, 0.8938, 0.4167, 0.3812, 0.4267, 0.5612, 0.5583]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.07242716127075255 step: 73 running loss: 0.000992152894119898 Train Steps: 73/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6199, 0.4071, 0.7600, 0.2117, 0.4037, 0.2767, 0.6138, 0.5550], [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400], [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350], [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133], [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717], [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533], [ nan, nan, 0.6793, 0.2110, 0.4012, 0.2167, 0.5112, 0.5583], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6857, 0.4347, 0.8096, 0.2185, 0.4308, 0.2854, 0.6269, 0.5558], [ 0.6357, 0.4159, 0.9101, 0.4135, 0.3712, 0.4606, 0.6289, 0.5638], [ 0.6290, 0.4377, 0.9140, 0.4249, 0.3806, 0.4106, 0.5434, 0.5583], [ 0.5919, 0.3861, 0.8664, 0.2422, 0.4864, 0.1704, 0.6385, 0.5383], [ 0.5907, 0.3928, 0.7141, 0.2533, 0.3866, 0.2558, 0.5609, 0.5914], [ 0.6507, 0.4208, 0.7482, 0.1980, 0.3761, 0.2632, 0.5899, 0.5728], [-0.0138, -0.0199, 0.7143, 0.2253, 0.4229, 0.2101, 0.5316, 0.5606], [ 0.6328, 0.3930, 0.9018, 0.4740, 0.3764, 0.5228, 0.6135, 0.5193]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6199, 0.4071, 0.7600, 0.2117, 0.4038, 0.2767, 0.6137, 0.5550], [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6137, 0.5400], [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350], [0.6266, 0.4065, 0.8313, 0.2467, 0.4787, 0.1733, 0.6313, 0.5133], [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717], [0.6114, 0.4018, 0.7212, 0.1967, 0.3762, 0.2700, 0.5875, 0.5533], [0.0000, 0.0000, 0.6793, 0.2110, 0.4013, 0.2167, 0.5113, 0.5583], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.07284825743408874 step: 74 running loss: 0.0009844359112714694 Train Steps: 74/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921], [0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583], [0.6289, 0.4032, 0.8419, 0.5446, 0.4075, 0.5017, 0.6312, 0.5117], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6102, 0.3897, 0.8551, 0.2125, 0.3776, 0.2651, 0.6252, 0.5094], [0.6052, 0.3984, 0.9014, 0.4016, 0.3618, 0.3818, 0.5351, 0.5663], [0.6176, 0.4032, 0.8555, 0.5277, 0.3850, 0.4669, 0.6429, 0.5278], [0.5970, 0.3958, 0.8931, 0.4394, 0.4216, 0.5530, 0.6155, 0.5644], [0.5029, 0.3541, 0.7187, 0.2376, 0.3972, 0.2121, 0.5850, 0.5618], [0.5635, 0.3682, 0.8465, 0.2279, 0.4235, 0.1639, 0.6123, 0.5153], [0.5826, 0.3937, 0.8821, 0.4890, 0.4012, 0.4987, 0.5960, 0.5400], [0.5946, 0.4059, 0.8974, 0.4093, 0.4117, 0.4826, 0.5654, 0.5658]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921], [0.6212, 0.4033, 0.8938, 0.4167, 0.3812, 0.4267, 0.5612, 0.5583], [0.6289, 0.4031, 0.8419, 0.5446, 0.4075, 0.5017, 0.6313, 0.5117], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.0736210917821154 step: 75 running loss: 0.000981614557094872 Train Steps: 75/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626], [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.6202, 0.4079, 0.8025, 0.2500, 0.3763, 0.3217, 0.6125, 0.5533], [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550], [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064], [0.6332, 0.4118, 0.9238, 0.4267, 0.4012, 0.4733, 0.7525, 0.5436], [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6487, 0.4125, 0.8374, 0.5678, 0.3638, 0.4479, 0.6072, 0.4584], [0.5668, 0.3660, 0.8805, 0.4362, 0.3959, 0.5112, 0.6293, 0.5503], [0.5917, 0.3874, 0.7891, 0.2344, 0.3570, 0.2849, 0.6054, 0.5305], [0.5951, 0.3889, 0.8693, 0.4499, 0.3407, 0.3709, 0.5155, 0.5738], [0.6648, 0.4452, 0.7050, 0.2302, 0.3687, 0.1957, 0.5489, 0.5349], [0.5661, 0.3774, 0.8431, 0.4716, 0.4545, 0.4936, 0.5221, 0.5009], [0.6120, 0.4011, 0.9029, 0.4008, 0.3748, 0.4457, 0.7348, 0.5422], [0.6432, 0.4102, 0.8724, 0.3048, 0.3337, 0.3145, 0.6142, 0.4937]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626], [0.6241, 0.4142, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.6202, 0.4079, 0.8025, 0.2500, 0.3762, 0.3217, 0.6125, 0.5533], [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5362, 0.5550], [0.6102, 0.4005, 0.8687, 0.5100, 0.4812, 0.5400, 0.5404, 0.5064], [0.6332, 0.4118, 0.9237, 0.4267, 0.4013, 0.4733, 0.7525, 0.5436], [0.6255, 0.4017, 0.8687, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.07427169638685882 step: 76 running loss: 0.0009772591629849845 Train Steps: 76/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6125, 0.4035, 0.7825, 0.3100, 0.3463, 0.4900, 0.5832, 0.5637], [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083], [0.6275, 0.4048, 0.8488, 0.2883, 0.4463, 0.2033, 0.6321, 0.5155], [0.6259, 0.4133, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983], [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6216, 0.4167, 0.8588, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6554, 0.4331, 0.7440, 0.2241, 0.3653, 0.3191, 0.6177, 0.5078], [0.5756, 0.3785, 0.7637, 0.3179, 0.3499, 0.4842, 0.5989, 0.5165], [0.6087, 0.3967, 0.8508, 0.5047, 0.3572, 0.4255, 0.5865, 0.5805], [0.6427, 0.4099, 0.8366, 0.2696, 0.4381, 0.2028, 0.6484, 0.4932], [0.6039, 0.3984, 0.8158, 0.2238, 0.4992, 0.1749, 0.6391, 0.4684], [0.6422, 0.4182, 0.8247, 0.5406, 0.3554, 0.3843, 0.5606, 0.5869], [0.6590, 0.4183, 0.8570, 0.3301, 0.3534, 0.3471, 0.6224, 0.4780], [0.6645, 0.4232, 0.8241, 0.5647, 0.3989, 0.5272, 0.6056, 0.5310]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6125, 0.4035, 0.7825, 0.3100, 0.3462, 0.4900, 0.5832, 0.5637], [0.6186, 0.4060, 0.8750, 0.5050, 0.3537, 0.4367, 0.5813, 0.6083], [0.6275, 0.4048, 0.8487, 0.2883, 0.4462, 0.2033, 0.6321, 0.5155], [0.6259, 0.4132, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6260, 0.4214, 0.8537, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983], [0.6255, 0.4017, 0.8687, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6216, 0.4167, 0.8587, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.07463139062747359 step: 77 running loss: 0.0009692388393178388 Train Steps: 77/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6214, 0.4112, 0.7838, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733], [0.6199, 0.4065, 0.7598, 0.2385, 0.4317, 0.1981, 0.5933, 0.5221], [0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100], [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6023, 0.3914, 0.8483, 0.4306, 0.3720, 0.3634, 0.5348, 0.5365], [0.6116, 0.4021, 0.8334, 0.4468, 0.4512, 0.2645, 0.5703, 0.5711], [0.6597, 0.4230, 0.7794, 0.1988, 0.3646, 0.2922, 0.6104, 0.4579], [0.5780, 0.3787, 0.7765, 0.3631, 0.3513, 0.4057, 0.5427, 0.5166], [0.6164, 0.3919, 0.8649, 0.4805, 0.3788, 0.4427, 0.6133, 0.5166], [0.6889, 0.4362, 0.7362, 0.2125, 0.4312, 0.2173, 0.6049, 0.4742], [0.5940, 0.3791, 0.8297, 0.5513, 0.3694, 0.4709, 0.6086, 0.5311], [0.6342, 0.4165, 0.7445, 0.2720, 0.4020, 0.2731, 0.6232, 0.5852]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6214, 0.4112, 0.7837, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733], [0.6199, 0.4065, 0.7598, 0.2385, 0.4317, 0.1981, 0.5933, 0.5221], [0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100], [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.07542235916480422 step: 78 running loss: 0.0009669533226256951 Train Steps: 78/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411], [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100], [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550], [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6350, 0.4043, 0.8738, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950], [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6322, 0.4329, 0.8176, 0.2788, 0.4258, 0.2343, 0.5572, 0.5423], [0.6476, 0.4297, 0.8261, 0.5259, 0.4003, 0.5108, 0.5924, 0.4821], [0.6028, 0.4122, 0.8890, 0.4161, 0.4090, 0.2621, 0.5544, 0.5396], [0.6388, 0.4294, 0.8767, 0.4735, 0.3778, 0.4834, 0.5753, 0.5458], [0.6263, 0.4230, 0.8026, 0.5230, 0.3972, 0.5063, 0.6677, 0.5665], [0.6590, 0.4498, 0.7379, 0.1787, 0.4037, 0.2600, 0.6136, 0.5496], [0.6252, 0.4112, 0.8502, 0.5726, 0.3917, 0.4984, 0.6056, 0.4986], [0.6329, 0.4150, 0.8400, 0.5292, 0.4007, 0.4738, 0.5812, 0.5399]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411], [0.6275, 0.4024, 0.8500, 0.5383, 0.3913, 0.4883, 0.6288, 0.5100], [0.6226, 0.4098, 0.8913, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6197, 0.4091, 0.8800, 0.4783, 0.3537, 0.4767, 0.5950, 0.5550], [0.6346, 0.4086, 0.7937, 0.5500, 0.3963, 0.4867, 0.7343, 0.5702], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6350, 0.4043, 0.8737, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950], [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.07581857420154847 step: 79 running loss: 0.000959728787361373 Train Steps: 79/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633], [0.6276, 0.4120, 0.8738, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892], [0.6353, 0.4128, 0.9138, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991], [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6265, 0.4088, 0.8025, 0.1850, 0.4163, 0.2500, 0.6290, 0.4947], [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6582, 0.4217, 0.8559, 0.5431, 0.3977, 0.5892, 0.6823, 0.5507], [0.5421, 0.3751, 0.8435, 0.3560, 0.4198, 0.2311, 0.5940, 0.4788], [0.5280, 0.3593, 0.8839, 0.3768, 0.4569, 0.3085, 0.7006, 0.6056], [0.6138, 0.4281, 0.6970, 0.2340, 0.3655, 0.2705, 0.5795, 0.5592], [0.6089, 0.4218, 0.8466, 0.4879, 0.4248, 0.5155, 0.5631, 0.5667], [0.6077, 0.4128, 0.7871, 0.4091, 0.4669, 0.2647, 0.5149, 0.6082], [0.6323, 0.4149, 0.7712, 0.2223, 0.4029, 0.2477, 0.6125, 0.5030], [0.6510, 0.4502, 0.7419, 0.2712, 0.4474, 0.1928, 0.5697, 0.5589]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6293, 0.4024, 0.8750, 0.5000, 0.4013, 0.5733, 0.7121, 0.5633], [0.6276, 0.4120, 0.8737, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892], [0.6353, 0.4128, 0.9137, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991], [0.6114, 0.4018, 0.7212, 0.1967, 0.3762, 0.2700, 0.5875, 0.5533], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6299, 0.4303, 0.7962, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6265, 0.4088, 0.8025, 0.1850, 0.4162, 0.2500, 0.6290, 0.4947], [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.07664816980832256 step: 80 running loss: 0.000958102122604032 Train Steps: 80/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131], [0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136], [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274], [0.6166, 0.4008, 0.8563, 0.5667, 0.4388, 0.4933, 0.5575, 0.5567]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6689, 0.4372, 0.9093, 0.4226, 0.4324, 0.3600, 0.7135, 0.6139], [0.6498, 0.4387, 0.8441, 0.5649, 0.3814, 0.4846, 0.7158, 0.5975], [0.6646, 0.4442, 0.8671, 0.4075, 0.3482, 0.3767, 0.5877, 0.5507], [0.6307, 0.4162, 0.8190, 0.3802, 0.3695, 0.5152, 0.5928, 0.5423], [0.6437, 0.4392, 0.6882, 0.2361, 0.4377, 0.1644, 0.5355, 0.5713], [0.6287, 0.4257, 0.8331, 0.5270, 0.4388, 0.5462, 0.5233, 0.5611], [0.6295, 0.4227, 0.8565, 0.3296, 0.4674, 0.2036, 0.6249, 0.5424], [0.5897, 0.3983, 0.8491, 0.5866, 0.4412, 0.4888, 0.5794, 0.5939]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131], [0.6132, 0.4118, 0.8200, 0.3633, 0.3562, 0.5400, 0.5787, 0.5136], [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274], [0.6166, 0.4008, 0.8562, 0.5667, 0.4387, 0.4933, 0.5575, 0.5567]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.07702205292298459 step: 81 running loss: 0.0009508895422590689 Train Steps: 81/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6263, 0.4030, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771], [ nan, nan, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702], [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748], [0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393], [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400], [0.6250, 0.4106, 0.8700, 0.3717, 0.3588, 0.4967, 0.6038, 0.5167], [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7931, 0.5002, 0.9014, 0.4972, 0.3986, 0.5360, 0.6156, 0.5124], [0.2200, 0.1395, 0.9035, 0.3557, 0.5113, 0.2155, 0.6968, 0.6014], [0.7179, 0.4789, 0.8840, 0.4460, 0.4199, 0.2446, 0.5541, 0.5812], [0.7196, 0.4776, 0.8282, 0.5575, 0.4153, 0.4038, 0.6555, 0.5852], [0.7195, 0.4702, 0.8308, 0.2887, 0.4164, 0.2733, 0.6438, 0.5775], [0.7504, 0.4889, 0.8567, 0.3720, 0.3693, 0.4779, 0.6205, 0.5722], [0.7627, 0.4916, 0.8883, 0.4007, 0.3814, 0.4807, 0.6480, 0.5714], [0.1268, 0.0987, 0.7463, 0.2425, 0.3991, 0.2815, 0.5947, 0.6059]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6263, 0.4029, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771], [0.0000, 0.0000, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702], [0.6226, 0.4098, 0.8913, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748], [0.6286, 0.4086, 0.8408, 0.2801, 0.4162, 0.2800, 0.6725, 0.5393], [0.6168, 0.4029, 0.8523, 0.3417, 0.3587, 0.5000, 0.6125, 0.5400], [0.6250, 0.4105, 0.8700, 0.3717, 0.3587, 0.4967, 0.6037, 0.5167], [0.0000, 0.0000, 0.7525, 0.2291, 0.3837, 0.3017, 0.6050, 0.5667]]], device='cuda:0') loss_train_step before backward: tensor(0.0039, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0039, device='cuda:0', grad_fn=) loss_train: 0.08094305740087293 step: 82 running loss: 0.0009871104561082066 Train Steps: 82/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750], [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267], [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320], [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938], [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167], [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100], [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6775, 0.4409, 0.8038, 0.2923, 0.4102, 0.2379, 0.5431, 0.5955], [0.7007, 0.4416, 0.8367, 0.5775, 0.4175, 0.4591, 0.5865, 0.5586], [0.6926, 0.4455, 0.8810, 0.4902, 0.4014, 0.5026, 0.7283, 0.5516], [0.6904, 0.4220, 0.8995, 0.5639, 0.4153, 0.4616, 0.6263, 0.5324], [0.6587, 0.4276, 0.8119, 0.3081, 0.3876, 0.3840, 0.6036, 0.6393], [0.1182, 0.0820, 0.8003, 0.2678, 0.4247, 0.2427, 0.5935, 0.6027], [0.6364, 0.4076, 0.9295, 0.4042, 0.3742, 0.3750, 0.6706, 0.5503], [0.6407, 0.4136, 0.8899, 0.5239, 0.3889, 0.4709, 0.6358, 0.5303]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6182, 0.4099, 0.7812, 0.3000, 0.3938, 0.2367, 0.5325, 0.5750], [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267], [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320], [0.6334, 0.4037, 0.8637, 0.5733, 0.4013, 0.4717, 0.6369, 0.4938], [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6037, 0.6167], [0.0000, 0.0000, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100], [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.0819609404716175 step: 83 running loss: 0.0009874812105014157 Train Steps: 83/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6179, 0.4082, 0.6688, 0.2667, 0.3588, 0.3317, 0.5750, 0.5783], [0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019], [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200], [0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467], [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528], [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6161, 0.3941, 0.7660, 0.3402, 0.4910, 0.1819, 0.5801, 0.6160], [0.6550, 0.4167, 0.7156, 0.2698, 0.3404, 0.3261, 0.6140, 0.5784], [0.5628, 0.3644, 0.9160, 0.4328, 0.3729, 0.5294, 0.5967, 0.5198], [0.5325, 0.3280, 0.8888, 0.3057, 0.4375, 0.2048, 0.6688, 0.5315], [0.5760, 0.3584, 0.9019, 0.4981, 0.4276, 0.5066, 0.6589, 0.5600], [0.6087, 0.3825, 0.9356, 0.3558, 0.4076, 0.3620, 0.7728, 0.5539], [0.5557, 0.3471, 0.8858, 0.5116, 0.4644, 0.5172, 0.5522, 0.5472], [0.5297, 0.3323, 0.8645, 0.5822, 0.4484, 0.5120, 0.5551, 0.5111]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6179, 0.4082, 0.6687, 0.2667, 0.3587, 0.3317, 0.5750, 0.5783], [0.6161, 0.4099, 0.8737, 0.4383, 0.3787, 0.5483, 0.5605, 0.5019], [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200], [0.6199, 0.4059, 0.8700, 0.4900, 0.4162, 0.5000, 0.6162, 0.5467], [0.6332, 0.4165, 0.9100, 0.3350, 0.4187, 0.3683, 0.7438, 0.5528], [0.6100, 0.4016, 0.8600, 0.5067, 0.4613, 0.5233, 0.5086, 0.5519], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.08303061101469211 step: 84 running loss: 0.0009884596549368109 Train Steps: 84/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400], [0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748], [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285], [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459], [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350], [0.6201, 0.4151, 0.8588, 0.5467, 0.3700, 0.3950, 0.5637, 0.5933]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5600, 0.3587, 0.8868, 0.3793, 0.3792, 0.4846, 0.6134, 0.5291], [0.6271, 0.4055, 0.9044, 0.4859, 0.4086, 0.3777, 0.7089, 0.5524], [0.5926, 0.3717, 0.8923, 0.5424, 0.3704, 0.4701, 0.6231, 0.5157], [0.6262, 0.3905, 0.9014, 0.4869, 0.3972, 0.5699, 0.7311, 0.5323], [0.5637, 0.3538, 0.9013, 0.3600, 0.4607, 0.2612, 0.6273, 0.5327], [0.5475, 0.3581, 0.8491, 0.5177, 0.4012, 0.4949, 0.6715, 0.5227], [0.4972, 0.3167, 0.8166, 0.2413, 0.4067, 0.2753, 0.5941, 0.5354], [0.5164, 0.3416, 0.8693, 0.5236, 0.3773, 0.4062, 0.5634, 0.5922]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6137, 0.5400], [0.6336, 0.4191, 0.8938, 0.5167, 0.3938, 0.3517, 0.7343, 0.5748], [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285], [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459], [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350], [0.6202, 0.4151, 0.8587, 0.5467, 0.3700, 0.3950, 0.5638, 0.5933]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.08437361448886804 step: 85 running loss: 0.0009926307586925651 Train Steps: 85/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733], [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167], [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933], [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183], [0.6222, 0.4169, 0.8638, 0.5650, 0.4313, 0.4783, 0.5637, 0.5633], [0.6185, 0.4042, 0.7700, 0.2250, 0.4062, 0.2117, 0.5763, 0.5150], [0.6176, 0.3911, 0.8738, 0.4217, 0.3488, 0.4033, 0.6025, 0.4817]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.0196, 0.0066, 0.7102, 0.2163, 0.4524, 0.2058, 0.6032, 0.5641], [0.6906, 0.4361, 0.8928, 0.5351, 0.4137, 0.5438, 0.6143, 0.5024], [0.6659, 0.4039, 0.9054, 0.3828, 0.3644, 0.4686, 0.7225, 0.4885], [0.0653, 0.0219, 0.6876, 0.2563, 0.4154, 0.2180, 0.5575, 0.5544], [0.6370, 0.3989, 0.7716, 0.3058, 0.3992, 0.3338, 0.6395, 0.5891], [0.6990, 0.4554, 0.8908, 0.5944, 0.4268, 0.5075, 0.5973, 0.5411], [0.6624, 0.4194, 0.7643, 0.2401, 0.4091, 0.2334, 0.6098, 0.5052], [0.6629, 0.4042, 0.8942, 0.4275, 0.3569, 0.4162, 0.6152, 0.4653]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733], [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167], [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [0.0000, 0.0000, 0.6688, 0.2513, 0.4112, 0.2117, 0.5193, 0.5933], [0.6196, 0.4094, 0.7563, 0.2817, 0.3938, 0.3183, 0.6012, 0.6183], [0.6222, 0.4169, 0.8637, 0.5650, 0.4313, 0.4783, 0.5638, 0.5633], [0.6184, 0.4042, 0.7700, 0.2250, 0.4062, 0.2117, 0.5763, 0.5150], [0.6176, 0.3911, 0.8737, 0.4217, 0.3487, 0.4033, 0.6025, 0.4817]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.08524726060568355 step: 86 running loss: 0.0009912472163451575 Train Steps: 86/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540], [0.6273, 0.4105, 0.8988, 0.4517, 0.3912, 0.2550, 0.5894, 0.4811], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510], [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064], [0.6133, 0.4066, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358], [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350], [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5647, 0.3489, 0.9124, 0.4385, 0.4083, 0.3114, 0.7018, 0.5291], [0.5665, 0.3577, 0.9090, 0.4379, 0.3979, 0.2554, 0.6149, 0.4734], [0.5564, 0.3416, 0.8504, 0.4353, 0.3480, 0.4655, 0.5663, 0.5339], [0.4692, 0.2868, 0.6970, 0.2790, 0.3601, 0.3093, 0.5470, 0.5174], [0.5477, 0.3552, 0.8623, 0.5110, 0.4692, 0.5417, 0.5568, 0.4848], [0.5594, 0.3586, 0.6936, 0.2703, 0.3643, 0.2353, 0.5577, 0.4833], [0.6098, 0.3628, 0.8656, 0.5068, 0.4168, 0.4914, 0.5540, 0.5095], [0.5030, 0.3005, 0.8339, 0.5956, 0.3856, 0.4719, 0.5939, 0.5110]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540], [0.6273, 0.4105, 0.8988, 0.4517, 0.3913, 0.2550, 0.5894, 0.4811], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510], [0.6102, 0.4005, 0.8687, 0.5100, 0.4812, 0.5400, 0.5404, 0.5064], [0.6133, 0.4065, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358], [0.6126, 0.3954, 0.8537, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350], [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.08683439713786356 step: 87 running loss: 0.000998096518826018 Train Steps: 87/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6198, 0.4115, 0.7762, 0.2717, 0.3713, 0.3200, 0.5837, 0.5683], [0.6104, 0.4029, 0.8738, 0.4900, 0.4088, 0.4533, 0.5070, 0.5510], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090], [0.6256, 0.4199, 0.8638, 0.5800, 0.3987, 0.4383, 0.5600, 0.5950], [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864], [0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5745, 0.3609, 0.8605, 0.4861, 0.3773, 0.5457, 0.6321, 0.4751], [0.6006, 0.3970, 0.7590, 0.2772, 0.3567, 0.3082, 0.5580, 0.5534], [0.5465, 0.3792, 0.8374, 0.4923, 0.3945, 0.4651, 0.4942, 0.5230], [0.5758, 0.3915, 0.8660, 0.4992, 0.3439, 0.4264, 0.6091, 0.4961], [0.5833, 0.3887, 0.8603, 0.3826, 0.3814, 0.2886, 0.6262, 0.4685], [0.5718, 0.3936, 0.8292, 0.5865, 0.3843, 0.4386, 0.5741, 0.5793], [0.5968, 0.3960, 0.8556, 0.5090, 0.3664, 0.4894, 0.6865, 0.5578], [0.5435, 0.3496, 0.8422, 0.5232, 0.3791, 0.5010, 0.6271, 0.4993]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6198, 0.4115, 0.7763, 0.2717, 0.3713, 0.3200, 0.5838, 0.5683], [0.6104, 0.4029, 0.8737, 0.4900, 0.4087, 0.4533, 0.5070, 0.5510], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090], [0.6256, 0.4199, 0.8637, 0.5800, 0.3988, 0.4383, 0.5600, 0.5950], [0.6314, 0.4107, 0.8750, 0.5100, 0.3787, 0.4900, 0.7121, 0.5864], [0.6308, 0.3990, 0.8687, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.0875556360988412 step: 88 running loss: 0.0009949504102141045 Train Steps: 88/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750], [0.6117, 0.4019, 0.8538, 0.4067, 0.3513, 0.3583, 0.5663, 0.5133], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090], [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650], [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207], [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5855, 0.3954, 0.6825, 0.2724, 0.3647, 0.2915, 0.5967, 0.5634], [0.5919, 0.3891, 0.8399, 0.4297, 0.3236, 0.3514, 0.5381, 0.5206], [0.5814, 0.3986, 0.8560, 0.4971, 0.4336, 0.5638, 0.6024, 0.5450], [0.5850, 0.3971, 0.8536, 0.3945, 0.3774, 0.2862, 0.6237, 0.4889], [0.5602, 0.3785, 0.8340, 0.4917, 0.4045, 0.5007, 0.5293, 0.5463], [0.5367, 0.3529, 0.7761, 0.2756, 0.4170, 0.2268, 0.6246, 0.5053], [0.6315, 0.4231, 0.8110, 0.2677, 0.4568, 0.2181, 0.6119, 0.4950], [0.5476, 0.3665, 0.8404, 0.5231, 0.3812, 0.4663, 0.5407, 0.5821]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750], [0.6116, 0.4019, 0.8537, 0.4067, 0.3512, 0.3583, 0.5663, 0.5133], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090], [0.6098, 0.3991, 0.8637, 0.4717, 0.4263, 0.4967, 0.5213, 0.5650], [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207], [0.6275, 0.4111, 0.8462, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6163, 0.4001, 0.8788, 0.5033, 0.4013, 0.4633, 0.5337, 0.5767]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.08827467539231293 step: 89 running loss: 0.000991850285306887 Train Steps: 89/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6284, 0.4093, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167], [0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693], [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284], [0.6185, 0.4042, 0.7700, 0.2250, 0.4062, 0.2117, 0.5763, 0.5150], [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5839, 0.3950, 0.8307, 0.5161, 0.3785, 0.4998, 0.6994, 0.5390], [0.5677, 0.3877, 0.7637, 0.2760, 0.3595, 0.3222, 0.5726, 0.5647], [0.6210, 0.4307, 0.8743, 0.4635, 0.3645, 0.3866, 0.5997, 0.5272], [0.5621, 0.3761, 0.8384, 0.4889, 0.3639, 0.4513, 0.4973, 0.5228], [0.5775, 0.3924, 0.7376, 0.2344, 0.4520, 0.1773, 0.5464, 0.5549], [0.5597, 0.3808, 0.7354, 0.2337, 0.4123, 0.2073, 0.5410, 0.5483], [0.6110, 0.4146, 0.8000, 0.2184, 0.4810, 0.1910, 0.5711, 0.5012], [0.5807, 0.3846, 0.8845, 0.4835, 0.3652, 0.4919, 0.6218, 0.5216]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6284, 0.4092, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167], [0.6038, 0.3946, 0.8413, 0.4883, 0.3562, 0.4550, 0.5266, 0.4693], [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284], [0.6184, 0.4042, 0.7700, 0.2250, 0.4062, 0.2117, 0.5763, 0.5150], [0.6276, 0.4095, 0.8238, 0.2250, 0.4663, 0.1783, 0.6171, 0.4869], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.08886227456969209 step: 90 running loss: 0.000987358606329912 Valid Steps: 10/10 Loss: nan 9.3309 -------------------------------------------------- Epoch: 6 Train Loss: 0.0010 Valid Loss: nan -------------------------------------------------- size of train loader is: 90 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090], [0.6275, 0.4081, 0.8063, 0.2017, 0.4825, 0.1583, 0.6156, 0.4869], [0.6145, 0.4007, 0.8775, 0.4533, 0.4562, 0.5533, 0.6088, 0.5533], [ nan, nan, 0.6512, 0.1717, 0.4100, 0.1983, 0.5253, 0.5240], [0.6200, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913], [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183], [0.6277, 0.4083, 0.8350, 0.2717, 0.4562, 0.1800, 0.5918, 0.4878], [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6453, 0.4394, 0.8678, 0.3829, 0.3862, 0.2651, 0.6238, 0.5143], [ 0.6626, 0.4516, 0.7857, 0.2041, 0.4790, 0.1266, 0.6018, 0.5084], [ 0.5989, 0.4005, 0.8487, 0.4353, 0.4402, 0.5201, 0.5815, 0.5510], [-0.0018, -0.0064, 0.6547, 0.1877, 0.4051, 0.1848, 0.5274, 0.5574], [ 0.6113, 0.3946, 0.8557, 0.5194, 0.3778, 0.4875, 0.5826, 0.4969], [ 0.6051, 0.4128, 0.7403, 0.2913, 0.3818, 0.2822, 0.5789, 0.6328], [ 0.6751, 0.4404, 0.8000, 0.2561, 0.4524, 0.1953, 0.5601, 0.5146], [ 0.6628, 0.4574, 0.8744, 0.4788, 0.3909, 0.4544, 0.5270, 0.5527]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090], [0.6275, 0.4081, 0.8062, 0.2017, 0.4825, 0.1583, 0.6156, 0.4869], [0.6145, 0.4007, 0.8775, 0.4533, 0.4563, 0.5533, 0.6087, 0.5533], [0.0000, 0.0000, 0.6513, 0.1717, 0.4100, 0.1983, 0.5253, 0.5240], [0.6199, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913], [0.6196, 0.4094, 0.7563, 0.2817, 0.3938, 0.3183, 0.6012, 0.6183], [0.6277, 0.4083, 0.8350, 0.2717, 0.4563, 0.1800, 0.5918, 0.4878], [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.0004884650697931647 step: 1 running loss: 0.0004884650697931647 Train Steps: 1/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6261, 0.4066, 0.8325, 0.2150, 0.4763, 0.2667, 0.7002, 0.5633], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864], [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567], [0.6276, 0.4120, 0.8738, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892], [0.6286, 0.4055, 0.9000, 0.4717, 0.3763, 0.4683, 0.7018, 0.5494]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6640, 0.4397, 0.8230, 0.2081, 0.4813, 0.2426, 0.6619, 0.5524], [0.5629, 0.3849, 0.7947, 0.1877, 0.4817, 0.1585, 0.5632, 0.4943], [0.6622, 0.4429, 0.8487, 0.2691, 0.4924, 0.1942, 0.6363, 0.5429], [0.5763, 0.4041, 0.6908, 0.2247, 0.4289, 0.1865, 0.4956, 0.5708], [0.6442, 0.4362, 0.8877, 0.5054, 0.3816, 0.4920, 0.6745, 0.5856], [0.5904, 0.4002, 0.8679, 0.4434, 0.3575, 0.4028, 0.5257, 0.5667], [0.6361, 0.4374, 0.8812, 0.3180, 0.4441, 0.2118, 0.5942, 0.4950], [0.6376, 0.4275, 0.9151, 0.4626, 0.3637, 0.4694, 0.6836, 0.5516]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6261, 0.4066, 0.8325, 0.2150, 0.4762, 0.2667, 0.7002, 0.5633], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6154, 0.4112, 0.7038, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6314, 0.4107, 0.8750, 0.5100, 0.3787, 0.4900, 0.7121, 0.5864], [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567], [0.6276, 0.4120, 0.8737, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892], [0.6286, 0.4055, 0.9000, 0.4717, 0.3762, 0.4683, 0.7018, 0.5494]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.000938043842324987 step: 2 running loss: 0.0004690219211624935 Train Steps: 2/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6200, 0.3993, 0.8519, 0.4923, 0.3962, 0.4717, 0.6013, 0.5433], [0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183], [0.6284, 0.4093, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167], [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5563, 0.5533]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5811, 0.3845, 0.7509, 0.2356, 0.4348, 0.2075, 0.5513, 0.5607], [0.6369, 0.4152, 0.8912, 0.4861, 0.3891, 0.4627, 0.6156, 0.5446], [0.7136, 0.4810, 0.7813, 0.2562, 0.4903, 0.1406, 0.6018, 0.5508], [0.6665, 0.4429, 0.9152, 0.4684, 0.4712, 0.5621, 0.6198, 0.5209], [0.7010, 0.4706, 0.9237, 0.4485, 0.3770, 0.3864, 0.6339, 0.5018], [0.6253, 0.4285, 0.7250, 0.2029, 0.3956, 0.2703, 0.5921, 0.5407], [0.6760, 0.4365, 0.9321, 0.4231, 0.3687, 0.3669, 0.6178, 0.5479], [0.6130, 0.4115, 0.8999, 0.5236, 0.4055, 0.4553, 0.5712, 0.5576]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6193, 0.4050, 0.7312, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6200, 0.3993, 0.8519, 0.4923, 0.3963, 0.4717, 0.6012, 0.5433], [0.6245, 0.4100, 0.7763, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6251, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6087, 0.5183], [0.6284, 0.4092, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167], [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6087, 0.5583], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5562, 0.5533]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.001644053525524214 step: 3 running loss: 0.0005480178418414047 Train Steps: 3/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450], [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947], [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133], [0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967], [0.6261, 0.4029, 0.8720, 0.3364, 0.3665, 0.3753, 0.6531, 0.5183]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6809, 0.4596, 0.8557, 0.4188, 0.4484, 0.2379, 0.5578, 0.6116], [0.6355, 0.4162, 0.8816, 0.5361, 0.4015, 0.4226, 0.5648, 0.5624], [0.6737, 0.4352, 0.8612, 0.5365, 0.4400, 0.4808, 0.6048, 0.5333], [0.6598, 0.4225, 0.8887, 0.4296, 0.3856, 0.4134, 0.6230, 0.5433], [0.6424, 0.3978, 0.7910, 0.1386, 0.4325, 0.2133, 0.6732, 0.5020], [0.6612, 0.4249, 0.8707, 0.4846, 0.4547, 0.4941, 0.5900, 0.5281], [0.6900, 0.4489, 0.9203, 0.4105, 0.4064, 0.2570, 0.6465, 0.5050], [0.6496, 0.4127, 0.8628, 0.3158, 0.3697, 0.3593, 0.6484, 0.5083]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6229, 0.4066, 0.8512, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450], [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947], [0.6202, 0.4053, 0.8637, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133], [0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967], [0.6261, 0.4029, 0.8720, 0.3364, 0.3665, 0.3753, 0.6531, 0.5183]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.0021857938554603606 step: 4 running loss: 0.0005464484638650902 Train Steps: 4/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.7335, 0.2569, 0.3788, 0.2667, 0.5066, 0.5578], [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494], [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702], [0.6263, 0.4030, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771], [0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.1280, 0.0777, 0.7659, 0.2281, 0.4004, 0.2490, 0.5271, 0.5553], [0.7083, 0.4421, 0.8876, 0.5065, 0.4055, 0.4180, 0.7057, 0.5312], [0.7266, 0.4623, 0.9024, 0.4195, 0.3830, 0.3276, 0.6195, 0.5389], [0.7253, 0.4551, 0.9119, 0.4579, 0.4334, 0.4596, 0.5622, 0.5170], [0.6726, 0.4309, 0.7633, 0.2221, 0.4598, 0.2383, 0.6116, 0.6030], [0.7279, 0.4571, 0.8371, 0.5180, 0.4143, 0.4637, 0.7164, 0.5570], [0.7040, 0.4440, 0.9334, 0.4408, 0.3999, 0.5113, 0.6502, 0.4685], [0.7033, 0.4410, 0.8890, 0.5672, 0.3969, 0.4621, 0.6445, 0.4977]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.7335, 0.2569, 0.3787, 0.2667, 0.5066, 0.5578], [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494], [0.6252, 0.4162, 0.8662, 0.4467, 0.3625, 0.3567, 0.6037, 0.5533], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6193, 0.4078, 0.7287, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6346, 0.4086, 0.7937, 0.5500, 0.3963, 0.4867, 0.7343, 0.5702], [0.6263, 0.4029, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771], [0.6310, 0.4017, 0.8562, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.004074420110555366 step: 5 running loss: 0.0008148840221110732 Train Steps: 5/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6296, 0.4045, 0.9138, 0.4100, 0.4232, 0.4242, 0.7422, 0.5297], [0.6126, 0.4039, 0.8237, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810], [0.6164, 0.4102, 0.8850, 0.4867, 0.4213, 0.5983, 0.5609, 0.5038], [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301], [0.6144, 0.4032, 0.8563, 0.3283, 0.3525, 0.4200, 0.5775, 0.5583], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7257, 0.4661, 0.8940, 0.4321, 0.4246, 0.4201, 0.7156, 0.5389], [0.6664, 0.4251, 0.8172, 0.4186, 0.3816, 0.3503, 0.6246, 0.6174], [0.6524, 0.4117, 0.8400, 0.3153, 0.4190, 0.2505, 0.6523, 0.4660], [0.6420, 0.4179, 0.8749, 0.5114, 0.4585, 0.5876, 0.6144, 0.5262], [0.6775, 0.4307, 0.8955, 0.3628, 0.3980, 0.2863, 0.6482, 0.5436], [0.6652, 0.4166, 0.8540, 0.5090, 0.4253, 0.5043, 0.6642, 0.5239], [0.6302, 0.3965, 0.8610, 0.3333, 0.3732, 0.4108, 0.5898, 0.5453], [0.6933, 0.4378, 0.8574, 0.4011, 0.3623, 0.3861, 0.6253, 0.5153]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6296, 0.4045, 0.9137, 0.4100, 0.4232, 0.4242, 0.7422, 0.5297], [0.6126, 0.4038, 0.8238, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810], [0.6164, 0.4102, 0.8850, 0.4867, 0.4212, 0.5983, 0.5609, 0.5038], [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301], [0.6144, 0.4032, 0.8562, 0.3283, 0.3525, 0.4200, 0.5775, 0.5583], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.004839719942538068 step: 6 running loss: 0.0008066199904230112 Train Steps: 6/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5988, 0.5700], [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600], [0.6216, 0.4167, 0.8588, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667], [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033], [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412], [0.6083, 0.3957, 0.8638, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6245, 0.4115, 0.8700, 0.4883, 0.4625, 0.5517, 0.6100, 0.5217]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5816, 0.3644, 0.7977, 0.2874, 0.3627, 0.3423, 0.6396, 0.5712], [0.6387, 0.4050, 0.7193, 0.2647, 0.3624, 0.3305, 0.6420, 0.5596], [0.6548, 0.4103, 0.8759, 0.5895, 0.4044, 0.4977, 0.6551, 0.5572], [0.6460, 0.4036, 0.8800, 0.5074, 0.4045, 0.5317, 0.6356, 0.5149], [0.6181, 0.4096, 0.8668, 0.5840, 0.4459, 0.4698, 0.5891, 0.5323], [0.6339, 0.3948, 0.8712, 0.4996, 0.4313, 0.5057, 0.5855, 0.4824], [0.6447, 0.4080, 0.8645, 0.4941, 0.4178, 0.4753, 0.5918, 0.5320], [0.6328, 0.4041, 0.8715, 0.4879, 0.4285, 0.5262, 0.6690, 0.5281]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5987, 0.5700], [0.6132, 0.4066, 0.7259, 0.2402, 0.3587, 0.3300, 0.6000, 0.5600], [0.6216, 0.4167, 0.8587, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667], [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033], [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412], [0.6083, 0.3957, 0.8637, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6245, 0.4115, 0.8700, 0.4883, 0.4625, 0.5517, 0.6100, 0.5217]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.005535328615223989 step: 7 running loss: 0.0007907612307462841 Train Steps: 7/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8850, 0.2817, 0.5112, 0.2183, 0.7184, 0.5436], [ nan, nan, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6122, 0.3993, 0.8738, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927], [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083], [0.6361, 0.4076, 0.8862, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297], [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332], [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.2315, 0.1313, 0.8693, 0.2834, 0.5037, 0.2446, 0.6931, 0.5315], [0.1525, 0.0825, 0.9006, 0.3641, 0.5002, 0.2595, 0.7082, 0.5658], [0.6859, 0.4555, 0.8358, 0.3334, 0.3620, 0.3177, 0.5499, 0.5456], [0.7301, 0.4657, 0.8544, 0.5141, 0.4441, 0.5523, 0.5109, 0.4873], [0.7582, 0.4908, 0.8610, 0.5299, 0.3585, 0.4687, 0.5936, 0.6015], [0.7582, 0.4922, 0.8566, 0.5872, 0.3775, 0.5032, 0.6848, 0.5286], [0.7481, 0.4672, 0.7994, 0.5595, 0.3788, 0.5282, 0.7132, 0.5261], [0.1357, 0.0802, 0.7567, 0.2895, 0.4016, 0.2786, 0.5403, 0.5677]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8850, 0.2817, 0.5113, 0.2183, 0.7184, 0.5436], [0.0000, 0.0000, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6122, 0.3993, 0.8737, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927], [0.6186, 0.4060, 0.8750, 0.5050, 0.3537, 0.4367, 0.5813, 0.6083], [0.6361, 0.4076, 0.8863, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297], [0.6289, 0.4019, 0.8112, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332], [0.0000, 0.0000, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600]]], device='cuda:0') loss_train_step before backward: tensor(0.0038, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0038, device='cuda:0', grad_fn=) loss_train: 0.009357340022688732 step: 8 running loss: 0.0011696675028360914 Train Steps: 8/90 Loss: 0.0012 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183], [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6063, 0.5617], [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6214, 0.4112, 0.7838, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882], [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5513, 0.3499, 0.8754, 0.3885, 0.4092, 0.2980, 0.6123, 0.5509], [0.5079, 0.3261, 0.8825, 0.5033, 0.3551, 0.4708, 0.5870, 0.5716], [0.5486, 0.3592, 0.8733, 0.5777, 0.3856, 0.4378, 0.5609, 0.5814], [0.5862, 0.3672, 0.8805, 0.3626, 0.4499, 0.3485, 0.7259, 0.5326], [0.5220, 0.3297, 0.7695, 0.2367, 0.3591, 0.3185, 0.5644, 0.5184], [0.5648, 0.3485, 0.8498, 0.4890, 0.3768, 0.5428, 0.5952, 0.5136], [0.5259, 0.3511, 0.8384, 0.5438, 0.3694, 0.4067, 0.5304, 0.5357], [0.5168, 0.3309, 0.8624, 0.4739, 0.4763, 0.5202, 0.5775, 0.5674]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183], [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6062, 0.5617], [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6214, 0.4112, 0.7837, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882], [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.01096438625245355 step: 9 running loss: 0.0012182651391615057 Train Steps: 9/90 Loss: 0.0012 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246], [0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300], [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6124, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402], [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[-0.1154, -0.0725, 0.7042, 0.2203, 0.4047, 0.2516, 0.5119, 0.5439], [ 0.6155, 0.3898, 0.7578, 0.2185, 0.3942, 0.3477, 0.6270, 0.5514], [ 0.5186, 0.3507, 0.8849, 0.4107, 0.3595, 0.4197, 0.5105, 0.5245], [ 0.5673, 0.3583, 0.8317, 0.5371, 0.3788, 0.4971, 0.5513, 0.5711], [ 0.5857, 0.3667, 0.8425, 0.5602, 0.4127, 0.5175, 0.6205, 0.5132], [ 0.5985, 0.3863, 0.7517, 0.3606, 0.5034, 0.2179, 0.5441, 0.6325], [ 0.5683, 0.3477, 0.8672, 0.3361, 0.3666, 0.3888, 0.6350, 0.5105], [ 0.5255, 0.3333, 0.8658, 0.5461, 0.4960, 0.5110, 0.5175, 0.5537]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.6900, 0.1917, 0.3938, 0.2367, 0.5240, 0.5246], [0.6207, 0.4081, 0.7663, 0.2067, 0.3963, 0.3200, 0.6313, 0.5300], [0.6082, 0.4024, 0.8737, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6123, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402], [0.6275, 0.4024, 0.8500, 0.5383, 0.3913, 0.4883, 0.6288, 0.5100], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6255, 0.4017, 0.8687, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6084, 0.4008, 0.8587, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.0121056787611451 step: 10 running loss: 0.00121056787611451 Train Steps: 10/90 Loss: 0.0012 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082], [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317], [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4749, 0.3249, 0.6890, 0.2968, 0.3606, 0.3122, 0.4926, 0.5845], [0.5149, 0.3481, 0.9194, 0.4068, 0.4093, 0.2657, 0.5357, 0.5565], [0.5039, 0.3380, 0.8925, 0.4838, 0.3632, 0.4430, 0.5818, 0.5308], [0.5325, 0.3546, 0.7897, 0.2229, 0.4606, 0.2118, 0.5699, 0.5353], [0.5349, 0.3446, 0.8689, 0.5037, 0.4380, 0.5839, 0.5417, 0.5008], [0.5464, 0.3667, 0.8537, 0.5286, 0.3979, 0.4906, 0.6903, 0.5944], [0.4846, 0.3226, 0.8902, 0.5299, 0.3904, 0.3915, 0.5313, 0.5475], [0.4769, 0.3094, 0.8922, 0.4655, 0.4012, 0.5251, 0.5598, 0.4930]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6226, 0.4098, 0.8913, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082], [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317], [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791]]], device='cuda:0') loss_train_step before backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0028, device='cuda:0', grad_fn=) loss_train: 0.0149084952136036 step: 11 running loss: 0.0013553177466912364 Train Steps: 11/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6152, 0.4131, 0.6863, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567], [ nan, nan, 0.7648, 0.2722, 0.3962, 0.2183, 0.5060, 0.5422], [0.6346, 0.4165, 0.9138, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471], [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901], [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633], [0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748], [ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5501, 0.3820, 0.7020, 0.2429, 0.3641, 0.3376, 0.5606, 0.5479], [0.6051, 0.4123, 0.8787, 0.4856, 0.4827, 0.4924, 0.5378, 0.5501], [0.0207, 0.0367, 0.7621, 0.2604, 0.3706, 0.2205, 0.4753, 0.5109], [0.6254, 0.4297, 0.9316, 0.4102, 0.3995, 0.4381, 0.7016, 0.5453], [0.5953, 0.4058, 0.8951, 0.4503, 0.3976, 0.4549, 0.4712, 0.5083], [0.6086, 0.3952, 0.8804, 0.4871, 0.3942, 0.5839, 0.6646, 0.5367], [0.6103, 0.4185, 0.8988, 0.5163, 0.3965, 0.3625, 0.7024, 0.5593], [0.0516, 0.0381, 0.7138, 0.2252, 0.4206, 0.1837, 0.4850, 0.5502]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6152, 0.4131, 0.6862, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [0.6112, 0.4029, 0.8637, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567], [0.0000, 0.0000, 0.7648, 0.2722, 0.3963, 0.2183, 0.5060, 0.5422], [0.6346, 0.4165, 0.9137, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471], [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901], [0.6293, 0.4024, 0.8750, 0.5000, 0.4013, 0.5733, 0.7121, 0.5633], [0.6336, 0.4191, 0.8938, 0.5167, 0.3938, 0.3517, 0.7343, 0.5748], [0.0000, 0.0000, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.015397945273434743 step: 12 running loss: 0.001283162106119562 Train Steps: 12/90 Loss: 0.0013 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012], [0.6197, 0.3986, 0.8800, 0.4617, 0.4188, 0.4783, 0.5687, 0.5550], [0.6226, 0.4103, 0.8575, 0.3450, 0.4388, 0.2067, 0.5787, 0.5383], [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621], [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426], [0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.2244, 0.1580, 0.7232, 0.2552, 0.4145, 0.2532, 0.5267, 0.5611], [0.6021, 0.4020, 0.8655, 0.5411, 0.3562, 0.4406, 0.6404, 0.4828], [0.5888, 0.3883, 0.8618, 0.4325, 0.3983, 0.4777, 0.5589, 0.5230], [0.6022, 0.4384, 0.8692, 0.3581, 0.4352, 0.2404, 0.5637, 0.5412], [0.0772, 0.0743, 0.8547, 0.2651, 0.5129, 0.2014, 0.6702, 0.5527], [0.5748, 0.3815, 0.8642, 0.5016, 0.3821, 0.5142, 0.5752, 0.4859], [0.6070, 0.4108, 0.8904, 0.4143, 0.3503, 0.3365, 0.5993, 0.5210], [0.5782, 0.4060, 0.8511, 0.4246, 0.3702, 0.4451, 0.4878, 0.5240]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012], [0.6197, 0.3986, 0.8800, 0.4617, 0.4187, 0.4783, 0.5688, 0.5550], [0.6226, 0.4103, 0.8575, 0.3450, 0.4387, 0.2067, 0.5788, 0.5383], [0.0000, 0.0000, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621], [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426], [0.6127, 0.4084, 0.8700, 0.4467, 0.3988, 0.4317, 0.5013, 0.5471]]], device='cuda:0') loss_train_step before backward: tensor(0.0039, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0039, device='cuda:0', grad_fn=) loss_train: 0.019262570276623592 step: 13 running loss: 0.0014817361751248916 Train Steps: 13/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117], [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567], [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200], [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519], [0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550], [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726], [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5710, 0.3831, 0.7021, 0.1966, 0.3722, 0.2206, 0.5852, 0.5107], [0.5573, 0.3777, 0.8761, 0.3221, 0.3587, 0.2827, 0.6023, 0.4967], [0.5786, 0.3829, 0.8698, 0.4437, 0.3681, 0.3612, 0.5885, 0.5433], [0.5425, 0.3700, 0.9020, 0.4852, 0.3674, 0.4227, 0.6205, 0.5100], [0.5838, 0.3936, 0.8620, 0.5194, 0.4552, 0.4883, 0.5314, 0.5250], [0.5672, 0.3824, 0.8751, 0.4618, 0.3647, 0.4477, 0.6069, 0.5339], [0.5511, 0.3646, 0.7593, 0.2733, 0.3573, 0.2556, 0.5577, 0.4689], [0.4920, 0.3405, 0.7663, 0.2715, 0.4157, 0.2283, 0.5999, 0.5700]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6043, 0.4022, 0.6888, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117], [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567], [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200], [0.6100, 0.4016, 0.8600, 0.5067, 0.4613, 0.5233, 0.5086, 0.5519], [0.6197, 0.4091, 0.8800, 0.4783, 0.3537, 0.4767, 0.5950, 0.5550], [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726], [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5688, 0.5567]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.020266804698621854 step: 14 running loss: 0.0014476289070444182 Train Steps: 14/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183], [0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650], [0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019], [0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5906, 0.4013, 0.8613, 0.4550, 0.4265, 0.5389, 0.6329, 0.5143], [0.5986, 0.4023, 0.8560, 0.2970, 0.4190, 0.2048, 0.6304, 0.5300], [0.5941, 0.4060, 0.7585, 0.2281, 0.4382, 0.1613, 0.6238, 0.5178], [0.5804, 0.3693, 0.7585, 0.2230, 0.4211, 0.1526, 0.6025, 0.5110], [0.5636, 0.3875, 0.6671, 0.2289, 0.3848, 0.1903, 0.5313, 0.5479], [0.5481, 0.3683, 0.8799, 0.4000, 0.3423, 0.3808, 0.6015, 0.5648], [0.5518, 0.3761, 0.8546, 0.4256, 0.3682, 0.4922, 0.5651, 0.4971], [0.5714, 0.3834, 0.8223, 0.2560, 0.3741, 0.2333, 0.5960, 0.5108]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6250, 0.4131, 0.8687, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.6213, 0.4001, 0.7713, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183], [0.6149, 0.4054, 0.6712, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6203, 0.4096, 0.8863, 0.4267, 0.3537, 0.4117, 0.6025, 0.5650], [0.6161, 0.4099, 0.8737, 0.4383, 0.3787, 0.5483, 0.5605, 0.5019], [0.6205, 0.4015, 0.8350, 0.2717, 0.3988, 0.2550, 0.5788, 0.5133]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.020893012144370005 step: 15 running loss: 0.0013928674762913337 Train Steps: 15/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400], [0.6353, 0.4128, 0.8488, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268], [0.6272, 0.4045, 0.8538, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6218, 0.4098, 0.7238, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350], [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6354, 0.4278, 0.7517, 0.2153, 0.4245, 0.1671, 0.5917, 0.5210], [0.5721, 0.3818, 0.8357, 0.3290, 0.3228, 0.4648, 0.6028, 0.5294], [0.6360, 0.4356, 0.8251, 0.2208, 0.5036, 0.1596, 0.6466, 0.5631], [0.5783, 0.3829, 0.8533, 0.4649, 0.3959, 0.4808, 0.5663, 0.5193], [0.6522, 0.4228, 0.8267, 0.5780, 0.3619, 0.4118, 0.5654, 0.4704], [0.6159, 0.4050, 0.7258, 0.1741, 0.4098, 0.2255, 0.6184, 0.5360], [0.6335, 0.4194, 0.8854, 0.4637, 0.3663, 0.3228, 0.6904, 0.5412], [0.5684, 0.3690, 0.8838, 0.3206, 0.4286, 0.3061, 0.7137, 0.5155]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.6168, 0.4029, 0.8523, 0.3417, 0.3587, 0.5000, 0.6125, 0.5400], [0.6353, 0.4128, 0.8487, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268], [0.6271, 0.4045, 0.8537, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6218, 0.4098, 0.7237, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350], [0.6307, 0.4029, 0.8988, 0.4817, 0.3938, 0.3500, 0.7311, 0.5378], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.02157996202004142 step: 16 running loss: 0.0013487476262525888 Train Steps: 16/90 Loss: 0.0013 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.6277, 0.4083, 0.8350, 0.2717, 0.4562, 0.1800, 0.5918, 0.4878], [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083], [0.6087, 0.3976, 0.8337, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300], [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552], [0.6198, 0.4101, 0.8838, 0.5283, 0.3763, 0.5267, 0.5913, 0.5567], [0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7122, 0.4747, 0.8445, 0.4124, 0.3622, 0.3767, 0.5372, 0.5509], [0.7166, 0.4586, 0.8083, 0.2608, 0.4620, 0.2070, 0.5982, 0.4951], [0.5601, 0.3597, 0.7098, 0.2516, 0.4317, 0.2329, 0.5504, 0.5738], [0.6958, 0.4550, 0.8660, 0.4839, 0.3600, 0.4225, 0.6120, 0.5877], [0.6833, 0.4521, 0.8362, 0.3760, 0.3806, 0.3086, 0.6095, 0.5258], [0.0834, 0.0515, 0.8465, 0.2422, 0.5094, 0.2428, 0.7514, 0.5423], [0.7125, 0.4723, 0.8479, 0.5150, 0.3856, 0.5142, 0.6055, 0.5365], [0.7223, 0.4807, 0.8598, 0.5261, 0.3942, 0.4611, 0.6115, 0.5069]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6151, 0.4125, 0.8737, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.6277, 0.4083, 0.8350, 0.2717, 0.4563, 0.1800, 0.5918, 0.4878], [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6186, 0.4060, 0.8750, 0.5050, 0.3537, 0.4367, 0.5813, 0.6083], [0.6087, 0.3976, 0.8338, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300], [0.0000, 0.0000, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552], [0.6198, 0.4101, 0.8838, 0.5283, 0.3762, 0.5267, 0.5913, 0.5567], [0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.0230987589166034 step: 17 running loss: 0.0013587505245060824 Train Steps: 17/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131], [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5837, 0.5583], [ nan, nan, 0.8850, 0.3000, 0.5363, 0.2250, 0.7343, 0.5771], [0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6948, 0.4342, 0.8274, 0.3955, 0.3516, 0.4189, 0.5985, 0.5181], [0.6616, 0.4353, 0.6757, 0.2311, 0.4291, 0.1741, 0.5695, 0.5668], [0.7638, 0.4901, 0.8623, 0.5660, 0.4073, 0.4236, 0.5623, 0.5521], [0.7171, 0.4646, 0.8997, 0.4317, 0.4204, 0.2874, 0.6812, 0.5608], [0.7199, 0.4767, 0.8423, 0.4943, 0.4293, 0.5077, 0.5502, 0.5317], [0.7093, 0.4723, 0.8679, 0.4347, 0.4737, 0.5264, 0.5791, 0.5443], [0.1111, 0.0722, 0.8653, 0.2842, 0.5132, 0.2146, 0.7645, 0.5478], [0.7002, 0.4582, 0.8673, 0.4589, 0.3723, 0.4649, 0.6140, 0.5413]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131], [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5512, 0.5750], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5838, 0.5583], [0.0000, 0.0000, 0.8850, 0.3000, 0.5362, 0.2250, 0.7343, 0.5771], [0.6197, 0.4091, 0.8800, 0.4783, 0.3537, 0.4767, 0.5950, 0.5550]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.02501470936113037 step: 18 running loss: 0.001389706075618354 Train Steps: 18/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374], [0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006], [0.6083, 0.3957, 0.8638, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6171, 0.4127, 0.8900, 0.4800, 0.4325, 0.5783, 0.5769, 0.5090], [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6704, 0.4291, 0.7901, 0.3458, 0.3656, 0.3822, 0.5583, 0.5874], [0.6621, 0.4222, 0.8653, 0.3725, 0.3861, 0.3233, 0.6159, 0.5628], [0.7010, 0.4444, 0.8489, 0.5905, 0.3913, 0.4867, 0.6440, 0.5469], [0.6631, 0.4274, 0.8679, 0.4871, 0.4668, 0.5191, 0.5627, 0.5301], [0.6619, 0.4045, 0.9055, 0.4651, 0.3786, 0.3689, 0.6459, 0.5426], [0.7109, 0.4619, 0.8841, 0.5022, 0.3762, 0.4304, 0.6520, 0.5508], [0.6863, 0.4545, 0.8942, 0.4778, 0.4615, 0.5887, 0.5857, 0.5413], [0.7541, 0.4829, 0.7754, 0.2297, 0.4743, 0.1854, 0.6195, 0.5590]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6102, 0.4001, 0.7738, 0.3583, 0.3462, 0.3800, 0.5524, 0.5689], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374], [0.6310, 0.4017, 0.8562, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006], [0.6083, 0.3957, 0.8637, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6171, 0.4127, 0.8900, 0.4800, 0.4325, 0.5783, 0.5769, 0.5090], [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.026208211231278256 step: 19 running loss: 0.0013793795384883293 Train Steps: 19/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343], [0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517], [0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297], [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6256, 0.4199, 0.8638, 0.5800, 0.3987, 0.4383, 0.5600, 0.5950], [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391], [0.6147, 0.4107, 0.8137, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412], [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6235, 0.3860, 0.8834, 0.3867, 0.4348, 0.3253, 0.7166, 0.5325], [0.6541, 0.4193, 0.8757, 0.4302, 0.4235, 0.5741, 0.5998, 0.5390], [0.6558, 0.4154, 0.8989, 0.4283, 0.4241, 0.3357, 0.6506, 0.5496], [0.7330, 0.4808, 0.7978, 0.3939, 0.4796, 0.2567, 0.5489, 0.6241], [0.7382, 0.4679, 0.8532, 0.5744, 0.3992, 0.4674, 0.5695, 0.5939], [0.6836, 0.4243, 0.8089, 0.2868, 0.4234, 0.2646, 0.6214, 0.5413], [0.6716, 0.4293, 0.7856, 0.3307, 0.3658, 0.2910, 0.4913, 0.5589], [0.0623, 0.0216, 0.8508, 0.2625, 0.5168, 0.2670, 0.7309, 0.5608]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6275, 0.4003, 0.9100, 0.3783, 0.4387, 0.3133, 0.7058, 0.5343], [0.6122, 0.4006, 0.8850, 0.4217, 0.4087, 0.5517, 0.6062, 0.5517], [0.6364, 0.4165, 0.9087, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297], [0.6299, 0.4303, 0.7962, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6256, 0.4199, 0.8637, 0.5800, 0.3988, 0.4383, 0.5600, 0.5950], [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391], [0.6147, 0.4107, 0.8138, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412], [0.0000, 0.0000, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.027047252719057724 step: 20 running loss: 0.0013523626359528862 Train Steps: 20/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8525, 0.2217, 0.5413, 0.2367, 0.7367, 0.5482], [0.6160, 0.4093, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808], [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124], [0.6148, 0.3996, 0.8488, 0.3867, 0.3488, 0.4067, 0.5863, 0.5000], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552], [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6275, 0.4048, 0.8488, 0.2883, 0.4463, 0.2033, 0.6321, 0.5155]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.1991, 0.1185, 0.8436, 0.2488, 0.5410, 0.2496, 0.7251, 0.5584], [0.6871, 0.4371, 0.8549, 0.5069, 0.3750, 0.4871, 0.5453, 0.5909], [0.7278, 0.4729, 0.8267, 0.3110, 0.4081, 0.2709, 0.5883, 0.5380], [0.6961, 0.4302, 0.8692, 0.4282, 0.3687, 0.4330, 0.5951, 0.5323], [0.6967, 0.4432, 0.7616, 0.2722, 0.3858, 0.3439, 0.6094, 0.5486], [0.0834, 0.0346, 0.8687, 0.2925, 0.5354, 0.2652, 0.7239, 0.5716], [0.7481, 0.4821, 0.8982, 0.5283, 0.3879, 0.4561, 0.6644, 0.5748], [0.7045, 0.4447, 0.8591, 0.3253, 0.4729, 0.2292, 0.6392, 0.5443]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8525, 0.2217, 0.5412, 0.2367, 0.7367, 0.5482], [0.6160, 0.4092, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808], [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124], [0.6148, 0.3996, 0.8487, 0.3867, 0.3487, 0.4067, 0.5863, 0.5000], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.0000, 0.0000, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552], [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6275, 0.4048, 0.8487, 0.2883, 0.4462, 0.2033, 0.6321, 0.5155]]], device='cuda:0') loss_train_step before backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train: 0.029495812923414633 step: 21 running loss: 0.0014045625201626016 Train Steps: 21/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116], [ nan, nan, 0.8488, 0.2300, 0.5563, 0.2100, 0.7390, 0.5679], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407], [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5837, 0.5583]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6522, 0.4116, 0.7930, 0.2094, 0.4086, 0.2687, 0.5724, 0.5104], [0.1545, 0.0771, 0.8851, 0.2805, 0.5526, 0.2638, 0.7352, 0.5587], [0.5950, 0.3730, 0.9092, 0.4708, 0.4053, 0.4660, 0.5343, 0.5407], [0.6627, 0.4341, 0.7680, 0.3104, 0.4542, 0.2209, 0.5618, 0.5739], [0.6252, 0.3891, 0.8365, 0.2923, 0.4193, 0.2608, 0.6167, 0.5603], [0.5939, 0.3655, 0.8846, 0.5923, 0.4072, 0.5317, 0.6605, 0.5606], [0.6327, 0.4021, 0.8764, 0.4260, 0.3743, 0.3276, 0.5130, 0.5581], [0.5707, 0.3681, 0.9099, 0.4746, 0.4918, 0.5689, 0.5726, 0.5683]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6165, 0.4106, 0.7575, 0.1733, 0.3837, 0.2650, 0.5680, 0.5116], [0.0000, 0.0000, 0.8487, 0.2300, 0.5562, 0.2100, 0.7390, 0.5679], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6204, 0.4110, 0.7912, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6133, 0.4094, 0.8495, 0.4028, 0.3587, 0.3200, 0.5003, 0.5407], [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5838, 0.5583]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.030567696288926527 step: 22 running loss: 0.0013894407404057513 Train Steps: 22/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6336, 0.4086, 0.8900, 0.3950, 0.3900, 0.2950, 0.6504, 0.5066], [ nan, nan, 0.9088, 0.3783, 0.4562, 0.2617, 0.6741, 0.5575], [ nan, nan, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650], [ nan, nan, 0.7725, 0.2611, 0.3675, 0.2733, 0.5413, 0.5167], [0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960], [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6392, 0.4052, 0.8825, 0.5237, 0.4866, 0.5350, 0.5361, 0.5516], [0.7080, 0.4577, 0.7620, 0.3528, 0.5123, 0.1878, 0.5503, 0.6081], [0.6843, 0.4237, 0.9446, 0.3873, 0.4054, 0.3337, 0.6932, 0.5064], [0.1743, 0.1102, 0.9388, 0.3837, 0.4724, 0.2862, 0.6806, 0.5682], [0.1335, 0.0885, 0.7111, 0.2123, 0.4566, 0.2587, 0.5505, 0.5482], [0.1014, 0.0485, 0.7536, 0.2699, 0.3843, 0.2945, 0.5392, 0.5497], [0.5646, 0.3630, 0.9077, 0.3126, 0.4761, 0.2239, 0.6187, 0.5062], [0.6201, 0.4074, 0.8433, 0.3348, 0.3664, 0.3087, 0.5544, 0.5099]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6336, 0.4086, 0.8900, 0.3950, 0.3900, 0.2950, 0.6504, 0.5066], [0.0000, 0.0000, 0.9087, 0.3783, 0.4563, 0.2617, 0.6741, 0.5575], [0.0000, 0.0000, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650], [0.0000, 0.0000, 0.7725, 0.2611, 0.3675, 0.2733, 0.5412, 0.5167], [0.6259, 0.4156, 0.8813, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960], [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.03241958349826746 step: 23 running loss: 0.0014095471086203243 Train Steps: 23/90 Loss: 0.0014 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332], [0.6092, 0.4001, 0.8638, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986], [0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6171, 0.4127, 0.8900, 0.4800, 0.4325, 0.5783, 0.5769, 0.5090], [0.6153, 0.4119, 0.8463, 0.3833, 0.3600, 0.3200, 0.5106, 0.5563], [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082], [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5196, 0.3046, 0.9327, 0.4675, 0.3851, 0.3874, 0.6833, 0.5238], [0.5047, 0.3282, 0.8866, 0.4704, 0.4349, 0.5383, 0.5316, 0.5102], [0.4951, 0.3208, 0.9140, 0.3828, 0.4615, 0.2056, 0.6317, 0.5085], [0.5170, 0.3447, 0.8880, 0.4490, 0.3800, 0.3987, 0.5487, 0.5116], [0.4440, 0.2850, 0.9048, 0.4702, 0.4556, 0.5799, 0.5876, 0.5062], [0.4837, 0.3263, 0.8488, 0.3854, 0.3605, 0.3197, 0.5185, 0.5563], [0.4881, 0.3055, 0.8886, 0.5065, 0.4360, 0.5546, 0.6133, 0.5081], [0.5354, 0.3664, 0.8931, 0.5447, 0.4107, 0.4721, 0.5752, 0.5636]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332], [0.6092, 0.4001, 0.8637, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986], [0.6108, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6171, 0.4127, 0.8900, 0.4800, 0.4325, 0.5783, 0.5769, 0.5090], [0.6153, 0.4119, 0.8462, 0.3833, 0.3600, 0.3200, 0.5106, 0.5563], [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082], [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600]]], device='cuda:0') loss_train_step before backward: tensor(0.0030, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0030, device='cuda:0', grad_fn=) loss_train: 0.035387905983952805 step: 24 running loss: 0.0014744960826647002 Train Steps: 24/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841], [0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [ nan, nan, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583], [0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.5617, 0.3751, 0.8700, 0.3838, 0.3548, 0.3857, 0.5578, 0.5258], [ 0.5048, 0.3418, 0.8775, 0.3221, 0.4436, 0.2397, 0.6185, 0.4976], [ 0.5276, 0.3512, 0.9314, 0.4296, 0.4197, 0.3713, 0.7200, 0.5474], [ 0.5901, 0.4103, 0.7223, 0.2898, 0.4692, 0.1786, 0.5616, 0.5901], [-0.0880, -0.0387, 0.7203, 0.2246, 0.4522, 0.1743, 0.5261, 0.5415], [ 0.4960, 0.3342, 0.8938, 0.4103, 0.3636, 0.5101, 0.6043, 0.5562], [ 0.4895, 0.3157, 0.9209, 0.5233, 0.3707, 0.3770, 0.6103, 0.4452], [ 0.4501, 0.3161, 0.8708, 0.4482, 0.4058, 0.5794, 0.5658, 0.5021]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6250, 0.4131, 0.8687, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841], [0.6270, 0.4266, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [0.0000, 0.0000, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583], [0.6200, 0.4112, 0.8863, 0.4100, 0.3638, 0.4917, 0.6087, 0.6050], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6064, 0.4019, 0.8650, 0.4517, 0.4038, 0.5367, 0.5703, 0.5609]]], device='cuda:0') loss_train_step before backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train: 0.03780612416449003 step: 25 running loss: 0.0015122449665796011 Train Steps: 25/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283], [ nan, nan, 0.7192, 0.2346, 0.4037, 0.2050, 0.5138, 0.5650], [0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167], [0.6226, 0.4103, 0.8575, 0.3450, 0.4388, 0.2067, 0.5787, 0.5383], [0.6222, 0.3957, 0.8838, 0.5017, 0.3937, 0.4600, 0.5900, 0.5017], [0.6346, 0.4165, 0.9138, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.4933, 0.3324, 0.9031, 0.3576, 0.3816, 0.2557, 0.6110, 0.5010], [-0.1040, -0.0676, 0.7089, 0.2206, 0.4172, 0.1880, 0.5342, 0.5466], [ 0.5285, 0.3772, 0.8592, 0.3481, 0.3227, 0.3968, 0.5869, 0.5273], [ 0.5383, 0.3741, 0.7581, 0.3609, 0.3283, 0.4108, 0.5078, 0.5228], [ 0.5499, 0.3637, 0.7490, 0.2379, 0.4342, 0.1393, 0.5775, 0.4974], [ 0.4884, 0.3450, 0.8489, 0.3635, 0.4307, 0.2122, 0.5531, 0.5142], [ 0.4173, 0.2685, 0.8786, 0.4987, 0.3810, 0.4578, 0.5729, 0.4654], [ 0.4876, 0.3254, 0.9143, 0.3999, 0.3792, 0.4312, 0.7038, 0.5270]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6187, 0.5283], [0.0000, 0.0000, 0.7192, 0.2346, 0.4038, 0.2050, 0.5138, 0.5650], [0.6219, 0.4097, 0.8737, 0.3400, 0.3562, 0.4117, 0.5975, 0.5683], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6204, 0.4007, 0.7837, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167], [0.6226, 0.4103, 0.8575, 0.3450, 0.4387, 0.2067, 0.5788, 0.5383], [0.6222, 0.3957, 0.8838, 0.5017, 0.3938, 0.4600, 0.5900, 0.5017], [0.6346, 0.4165, 0.9137, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471]]], device='cuda:0') loss_train_step before backward: tensor(0.0029, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0029, device='cuda:0', grad_fn=) loss_train: 0.040753726119874045 step: 26 running loss: 0.0015674510046105401 Train Steps: 26/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268], [0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019], [0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783], [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261], [0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6095, 0.3970, 0.8688, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4985, 0.3262, 0.8613, 0.4730, 0.4176, 0.4879, 0.6350, 0.5325], [0.5425, 0.3775, 0.8711, 0.4228, 0.3583, 0.5446, 0.5575, 0.5107], [0.5347, 0.3735, 0.8484, 0.4161, 0.3385, 0.3609, 0.5067, 0.5492], [0.5321, 0.3510, 0.8759, 0.4189, 0.3166, 0.3310, 0.6180, 0.4916], [0.5305, 0.3616, 0.8636, 0.4562, 0.3478, 0.3919, 0.5319, 0.5672], [0.5022, 0.3362, 0.8598, 0.4538, 0.3929, 0.4576, 0.5421, 0.5218], [0.5116, 0.3473, 0.8895, 0.3635, 0.3684, 0.2352, 0.6363, 0.5174], [0.4949, 0.3282, 0.8744, 0.4593, 0.4581, 0.4634, 0.5353, 0.5341]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268], [0.6161, 0.4099, 0.8737, 0.4383, 0.3787, 0.5483, 0.5605, 0.5019], [0.6151, 0.4125, 0.8737, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783], [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261], [0.6277, 0.4117, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6095, 0.3970, 0.8687, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940]]], device='cuda:0') loss_train_step before backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train: 0.042814259621081874 step: 27 running loss: 0.0015857133192993286 Train Steps: 27/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110], [0.6250, 0.4013, 0.8525, 0.5417, 0.4037, 0.5117, 0.6325, 0.5017], [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609], [ nan, nan, 0.7335, 0.2569, 0.3788, 0.2667, 0.5066, 0.5578], [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383], [0.6173, 0.4114, 0.7325, 0.2500, 0.4213, 0.1917, 0.5338, 0.5700], [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320], [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.5983, 0.4167, 0.8870, 0.4386, 0.3617, 0.5374, 0.5467, 0.5090], [ 0.6104, 0.4011, 0.8272, 0.5336, 0.3790, 0.4975, 0.5823, 0.4976], [ 0.5249, 0.3588, 0.8783, 0.3062, 0.3926, 0.2847, 0.6929, 0.5565], [-0.1273, -0.0658, 0.7397, 0.2481, 0.3711, 0.2330, 0.5285, 0.5775], [ 0.5797, 0.4041, 0.7436, 0.2373, 0.4315, 0.1334, 0.5688, 0.5258], [ 0.6172, 0.4352, 0.7226, 0.2292, 0.3858, 0.1790, 0.5193, 0.5609], [ 0.6064, 0.3998, 0.8377, 0.5035, 0.3568, 0.5040, 0.6954, 0.5307], [ 0.6284, 0.4079, 0.8532, 0.4978, 0.3948, 0.4959, 0.5576, 0.4846]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6186, 0.4129, 0.8900, 0.4567, 0.3938, 0.5417, 0.5734, 0.5110], [0.6250, 0.4013, 0.8525, 0.5417, 0.4038, 0.5117, 0.6325, 0.5017], [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609], [0.0000, 0.0000, 0.7335, 0.2569, 0.3787, 0.2667, 0.5066, 0.5578], [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383], [0.6173, 0.4114, 0.7325, 0.2500, 0.4212, 0.1917, 0.5337, 0.5700], [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320], [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.043800648621981964 step: 28 running loss: 0.0015643088793564988 Train Steps: 28/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6147, 0.4107, 0.8137, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412], [0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272], [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567], [0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164], [0.6111, 0.4033, 0.8300, 0.3267, 0.3588, 0.3333, 0.5444, 0.5637], [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5070, 0.3404, 0.7796, 0.3037, 0.3528, 0.2542, 0.5178, 0.5303], [0.5961, 0.3965, 0.8401, 0.3414, 0.3790, 0.2841, 0.5402, 0.5174], [0.6057, 0.3977, 0.8470, 0.4220, 0.3571, 0.3361, 0.6071, 0.5284], [0.5779, 0.3969, 0.7766, 0.3082, 0.3443, 0.3543, 0.5923, 0.5151], [0.6144, 0.4063, 0.8500, 0.4293, 0.3710, 0.4007, 0.5790, 0.5648], [0.5379, 0.3636, 0.8434, 0.5175, 0.3964, 0.3437, 0.5801, 0.5920], [0.6287, 0.4280, 0.8185, 0.3094, 0.3471, 0.3301, 0.5489, 0.5375], [0.6084, 0.4077, 0.8355, 0.4366, 0.3513, 0.3734, 0.5412, 0.5617]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6147, 0.4107, 0.8138, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412], [0.6186, 0.4080, 0.8625, 0.3483, 0.3787, 0.2650, 0.5320, 0.5272], [0.6252, 0.4162, 0.8662, 0.4467, 0.3625, 0.3567, 0.6037, 0.5533], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567], [0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164], [0.6111, 0.4033, 0.8300, 0.3267, 0.3587, 0.3333, 0.5444, 0.5637], [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.044581994734471664 step: 29 running loss: 0.0015373101632576436 Train Steps: 29/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748], [0.6084, 0.3981, 0.8588, 0.5233, 0.4600, 0.5367, 0.5680, 0.5006], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575], [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6040, 0.4002, 0.7338, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5954, 0.4118, 0.8543, 0.4456, 0.4525, 0.4905, 0.5458, 0.5619], [0.6419, 0.4508, 0.8598, 0.4282, 0.4053, 0.5013, 0.5462, 0.5402], [0.6981, 0.4758, 0.8232, 0.5364, 0.3724, 0.3886, 0.6626, 0.5528], [0.6648, 0.4608, 0.8242, 0.5233, 0.4306, 0.5186, 0.5066, 0.5186], [0.6749, 0.4485, 0.8571, 0.3060, 0.3899, 0.3924, 0.6725, 0.5735], [0.6380, 0.4405, 0.8437, 0.5172, 0.3811, 0.5291, 0.7002, 0.5559], [0.6293, 0.4254, 0.7027, 0.2040, 0.3756, 0.1766, 0.5020, 0.5047], [0.6604, 0.4439, 0.8687, 0.4279, 0.3489, 0.3185, 0.5489, 0.5276]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748], [0.6084, 0.3981, 0.8587, 0.5233, 0.4600, 0.5367, 0.5680, 0.5006], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575], [0.6339, 0.4123, 0.8637, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6040, 0.4002, 0.7337, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.04559074839926325 step: 30 running loss: 0.001519691613308775 Train Steps: 30/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6060, 0.3924, 0.8450, 0.5717, 0.4200, 0.5217, 0.5253, 0.4752], [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117], [0.6147, 0.4112, 0.7988, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550], [0.6167, 0.4048, 0.6831, 0.3639, 0.3763, 0.3017, 0.5700, 0.5883], [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092], [0.6110, 0.3984, 0.8750, 0.4933, 0.4625, 0.4950, 0.5578, 0.5676], [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284], [0.6332, 0.4118, 0.9238, 0.4267, 0.4012, 0.4733, 0.7525, 0.5436]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6383, 0.4051, 0.8426, 0.5858, 0.4593, 0.5259, 0.5491, 0.5228], [0.7065, 0.4534, 0.8905, 0.4899, 0.3710, 0.4799, 0.6445, 0.5080], [0.6552, 0.4527, 0.7933, 0.3352, 0.3711, 0.2825, 0.5235, 0.5575], [0.7061, 0.4787, 0.7113, 0.3436, 0.3689, 0.3042, 0.5457, 0.5738], [0.6846, 0.4498, 0.8709, 0.5279, 0.3860, 0.5291, 0.6037, 0.5104], [0.6956, 0.4875, 0.8825, 0.5053, 0.4610, 0.5115, 0.5537, 0.5858], [0.5936, 0.3968, 0.7494, 0.2417, 0.4437, 0.1887, 0.5902, 0.5428], [0.7195, 0.4881, 0.8953, 0.4497, 0.4147, 0.4874, 0.7514, 0.5634]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6060, 0.3924, 0.8450, 0.5717, 0.4200, 0.5217, 0.5253, 0.4752], [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117], [0.6147, 0.4112, 0.7987, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550], [0.6167, 0.4048, 0.6831, 0.3639, 0.3762, 0.3017, 0.5700, 0.5883], [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092], [0.6110, 0.3984, 0.8750, 0.4933, 0.4625, 0.4950, 0.5578, 0.5676], [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284], [0.6332, 0.4118, 0.9237, 0.4267, 0.4013, 0.4733, 0.7525, 0.5436]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.04676109258434735 step: 31 running loss: 0.0015084223414305598 Train Steps: 31/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5988, 0.5667], [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617], [0.6222, 0.3957, 0.8838, 0.5017, 0.3937, 0.4600, 0.5900, 0.5017], [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5837, 0.5317], [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378], [0.6059, 0.4002, 0.7562, 0.2767, 0.3538, 0.3033, 0.5529, 0.5455], [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6692, 0.4374, 0.8151, 0.2765, 0.4514, 0.2167, 0.6261, 0.5162], [0.5933, 0.3883, 0.7637, 0.2549, 0.4119, 0.2749, 0.5830, 0.6016], [0.6993, 0.4590, 0.8689, 0.5085, 0.4934, 0.4897, 0.5562, 0.5860], [0.6885, 0.4309, 0.8752, 0.5305, 0.4021, 0.4740, 0.5839, 0.5138], [0.7271, 0.4863, 0.8403, 0.3473, 0.3716, 0.3249, 0.5934, 0.5531], [0.7110, 0.4639, 0.8905, 0.4920, 0.4118, 0.3469, 0.7221, 0.5454], [0.7115, 0.4821, 0.7279, 0.3006, 0.3740, 0.3376, 0.5692, 0.5613], [0.7465, 0.4876, 0.8552, 0.5547, 0.3968, 0.4857, 0.6460, 0.5347]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6277, 0.4056, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5987, 0.5667], [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617], [0.6222, 0.3957, 0.8838, 0.5017, 0.3938, 0.4600, 0.5900, 0.5017], [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5838, 0.5317], [0.6307, 0.4029, 0.8988, 0.4817, 0.3938, 0.3500, 0.7311, 0.5378], [0.6059, 0.4002, 0.7563, 0.2767, 0.3537, 0.3033, 0.5529, 0.5455], [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.048350871045840904 step: 32 running loss: 0.0015109647201825283 Train Steps: 32/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852], [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366], [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567], [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7183, 0.4435, 0.7780, 0.2519, 0.4622, 0.1673, 0.6079, 0.5254], [0.6890, 0.4253, 0.8122, 0.2988, 0.4185, 0.2408, 0.6214, 0.5200], [0.7704, 0.4847, 0.8624, 0.4061, 0.3801, 0.4079, 0.6020, 0.5743], [0.7668, 0.4961, 0.9040, 0.5122, 0.4858, 0.4915, 0.5727, 0.5860], [0.7760, 0.4867, 0.8250, 0.5689, 0.4193, 0.4985, 0.7223, 0.5521], [0.1626, 0.0791, 0.7773, 0.2831, 0.4128, 0.2344, 0.5432, 0.5718], [0.7430, 0.4662, 0.8972, 0.3785, 0.4033, 0.4556, 0.7315, 0.5490], [0.7862, 0.4972, 0.8749, 0.5722, 0.4226, 0.4457, 0.5695, 0.4978]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6204, 0.4007, 0.7837, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6140, 0.4070, 0.8700, 0.5000, 0.4613, 0.4900, 0.5260, 0.5852], [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366], [0.0000, 0.0000, 0.7515, 0.2708, 0.3988, 0.2267, 0.5163, 0.5567], [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869]]], device='cuda:0') loss_train_step before backward: tensor(0.0034, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0034, device='cuda:0', grad_fn=) loss_train: 0.0517900716222357 step: 33 running loss: 0.0015693961097647182 Train Steps: 33/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400], [0.6200, 0.4049, 0.8638, 0.5617, 0.4125, 0.5100, 0.6013, 0.5317], [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5837, 0.5317], [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367], [0.6224, 0.4097, 0.7438, 0.2267, 0.3850, 0.2850, 0.5988, 0.5250], [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611], [0.6129, 0.4063, 0.8738, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6262, 0.4034, 0.8379, 0.2548, 0.4757, 0.2024, 0.6381, 0.5267], [0.6725, 0.4191, 0.8749, 0.5841, 0.4385, 0.5090, 0.6299, 0.5375], [0.6988, 0.4461, 0.8549, 0.3476, 0.3818, 0.3153, 0.6010, 0.5361], [0.6572, 0.4037, 0.8795, 0.3257, 0.4055, 0.2845, 0.6564, 0.5286], [0.6318, 0.4080, 0.7481, 0.2524, 0.4223, 0.3030, 0.6058, 0.5469], [0.6642, 0.4098, 0.8865, 0.4767, 0.4578, 0.4683, 0.5849, 0.5601], [0.6951, 0.4315, 0.8775, 0.5485, 0.4589, 0.4654, 0.5631, 0.5894], [0.6888, 0.4206, 0.9079, 0.5414, 0.3968, 0.3758, 0.6470, 0.4819]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400], [0.6199, 0.4049, 0.8637, 0.5617, 0.4125, 0.5100, 0.6012, 0.5317], [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5838, 0.5317], [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367], [0.6224, 0.4097, 0.7437, 0.2267, 0.3850, 0.2850, 0.5987, 0.5250], [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611], [0.6130, 0.4063, 0.8737, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.052406656759558246 step: 34 running loss: 0.001541372257634066 Train Steps: 34/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350], [0.6268, 0.4029, 0.8500, 0.2683, 0.3937, 0.3500, 0.6860, 0.5297], [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791], [0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748], [0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.6250, 0.4106, 0.8700, 0.3717, 0.3588, 0.4967, 0.6038, 0.5167], [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6202, 0.3780, 0.7768, 0.2914, 0.4400, 0.2355, 0.5699, 0.5551], [0.6527, 0.4156, 0.8333, 0.2886, 0.3895, 0.3419, 0.6861, 0.5263], [0.7031, 0.4200, 0.8915, 0.5075, 0.4197, 0.4965, 0.6249, 0.4885], [0.5822, 0.3480, 0.9264, 0.4856, 0.4590, 0.2685, 0.7117, 0.5417], [0.7206, 0.4528, 0.9131, 0.4676, 0.4242, 0.5323, 0.5750, 0.5227], [0.6428, 0.3892, 0.8874, 0.4650, 0.4004, 0.4453, 0.5611, 0.5443], [0.6559, 0.4030, 0.8732, 0.3825, 0.3912, 0.4849, 0.6456, 0.5597], [0.6766, 0.4138, 0.8838, 0.5775, 0.4509, 0.5091, 0.5833, 0.5711]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6229, 0.4066, 0.7613, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350], [0.6268, 0.4029, 0.8500, 0.2683, 0.3938, 0.3500, 0.6860, 0.5297], [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791], [0.6329, 0.4196, 0.9237, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748], [0.6186, 0.4129, 0.8900, 0.4567, 0.3938, 0.5417, 0.5734, 0.5110], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.6250, 0.4105, 0.8700, 0.3717, 0.3587, 0.4967, 0.6037, 0.5167], [0.6192, 0.4128, 0.8512, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.05321587107027881 step: 35 running loss: 0.001520453459150823 Train Steps: 35/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6188, 0.5433], [0.6127, 0.4118, 0.8650, 0.5083, 0.4088, 0.5367, 0.5300, 0.5456], [0.6261, 0.4066, 0.8325, 0.2150, 0.4763, 0.2667, 0.7002, 0.5633], [ nan, nan, 0.8888, 0.3100, 0.5262, 0.2817, 0.7145, 0.6003], [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312], [0.6317, 0.4038, 0.8287, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986], [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6875, 0.4320, 0.8564, 0.5744, 0.4058, 0.4555, 0.5670, 0.5928], [0.6644, 0.4154, 0.9198, 0.4740, 0.3834, 0.5058, 0.6488, 0.5174], [0.6649, 0.4253, 0.8915, 0.5074, 0.4305, 0.5249, 0.5541, 0.5414], [0.6130, 0.3686, 0.8652, 0.2302, 0.4823, 0.2487, 0.7005, 0.5395], [0.0716, 0.0118, 0.9327, 0.3118, 0.5047, 0.2719, 0.7061, 0.5590], [0.7000, 0.4239, 0.8849, 0.3953, 0.3745, 0.4427, 0.6062, 0.5183], [0.6816, 0.4226, 0.8694, 0.5829, 0.3913, 0.4535, 0.6265, 0.4815], [0.7006, 0.4562, 0.7253, 0.2361, 0.4113, 0.2272, 0.5532, 0.5245]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6187, 0.5433], [0.6127, 0.4118, 0.8650, 0.5083, 0.4087, 0.5367, 0.5300, 0.5456], [0.6261, 0.4066, 0.8325, 0.2150, 0.4762, 0.2667, 0.7002, 0.5633], [0.0000, 0.0000, 0.8888, 0.3100, 0.5263, 0.2817, 0.7145, 0.6003], [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312], [0.6317, 0.4038, 0.8288, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986], [0.6072, 0.4029, 0.7038, 0.2150, 0.3913, 0.2267, 0.5516, 0.5507]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.05410699962521903 step: 36 running loss: 0.0015029722118116398 Train Steps: 36/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.7225, 0.2167, 0.3987, 0.2283, 0.5427, 0.5181], [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197], [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733], [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.1819, 0.1039, 0.7380, 0.1940, 0.3809, 0.2428, 0.5503, 0.5216], [0.5895, 0.3619, 0.9011, 0.5202, 0.3940, 0.4769, 0.5646, 0.5651], [0.6023, 0.3690, 0.8773, 0.3985, 0.3527, 0.3939, 0.6104, 0.5608], [0.5871, 0.3863, 0.8467, 0.2964, 0.4725, 0.2073, 0.5912, 0.5331], [0.5822, 0.3717, 0.7326, 0.2561, 0.4252, 0.2154, 0.5554, 0.5833], [0.5647, 0.3731, 0.8998, 0.5127, 0.4519, 0.5512, 0.5964, 0.5058], [0.6267, 0.3978, 0.9258, 0.5146, 0.3502, 0.4465, 0.6199, 0.5489], [0.6408, 0.3909, 0.8375, 0.2682, 0.4460, 0.2356, 0.6727, 0.5104]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.7225, 0.2167, 0.3988, 0.2283, 0.5427, 0.5181], [0.6163, 0.4001, 0.8788, 0.5033, 0.4013, 0.4633, 0.5337, 0.5767], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6229, 0.4107, 0.8138, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197], [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733], [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.05532921847770922 step: 37 running loss: 0.0014953842831813302 Train Steps: 37/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [ nan, nan, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737], [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400], [0.6268, 0.4029, 0.8500, 0.2683, 0.3937, 0.3500, 0.6860, 0.5297], [0.6203, 0.4021, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [0.6207, 0.4110, 0.8738, 0.5000, 0.4800, 0.5633, 0.6300, 0.5433]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.5584, 0.3814, 0.7396, 0.3220, 0.4831, 0.1942, 0.5172, 0.6068], [ 0.5950, 0.3975, 0.7396, 0.1990, 0.3960, 0.2508, 0.5917, 0.5550], [ 0.5619, 0.3668, 0.7915, 0.3295, 0.3272, 0.3418, 0.5757, 0.5112], [-0.0712, -0.0459, 0.8614, 0.2210, 0.4886, 0.2268, 0.6629, 0.5561], [ 0.5797, 0.3691, 0.8868, 0.4916, 0.4381, 0.5639, 0.5886, 0.5283], [ 0.5762, 0.3851, 0.8199, 0.2775, 0.3516, 0.3389, 0.6678, 0.5093], [ 0.6372, 0.4073, 0.8711, 0.5256, 0.3495, 0.3846, 0.5695, 0.5189], [ 0.6062, 0.3832, 0.8912, 0.4883, 0.4486, 0.5551, 0.5764, 0.5444]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6200, 0.4071, 0.7337, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.0000, 0.0000, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737], [0.6185, 0.4098, 0.8838, 0.4900, 0.4538, 0.5800, 0.6288, 0.5400], [0.6268, 0.4029, 0.8500, 0.2683, 0.3938, 0.3500, 0.6860, 0.5297], [0.6203, 0.4020, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [0.6207, 0.4110, 0.8737, 0.5000, 0.4800, 0.5633, 0.6300, 0.5433]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.056172026408603415 step: 38 running loss: 0.0014782112212790373 Train Steps: 38/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575], [0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133], [0.6022, 0.3994, 0.8025, 0.3350, 0.3350, 0.4400, 0.5565, 0.5025], [0.6115, 0.4081, 0.6725, 0.2433, 0.4088, 0.1933, 0.5167, 0.5544], [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4933, 0.3400, 0.8663, 0.5160, 0.4407, 0.5056, 0.5874, 0.5225], [0.4956, 0.3527, 0.7297, 0.2796, 0.4278, 0.2104, 0.5411, 0.5783], [0.4876, 0.3190, 0.8714, 0.2921, 0.3974, 0.3867, 0.7012, 0.5721], [0.5360, 0.3603, 0.8380, 0.2608, 0.3976, 0.2755, 0.5699, 0.5331], [0.4918, 0.3276, 0.7912, 0.3296, 0.3477, 0.4230, 0.5811, 0.5216], [0.4919, 0.3439, 0.6967, 0.2405, 0.4097, 0.2051, 0.5022, 0.5646], [0.4976, 0.3403, 0.8864, 0.4499, 0.4459, 0.5219, 0.5894, 0.5640], [0.5277, 0.3468, 0.8753, 0.4974, 0.3895, 0.5026, 0.5835, 0.5277]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575], [0.6205, 0.4015, 0.8350, 0.2717, 0.3988, 0.2550, 0.5788, 0.5133], [0.6022, 0.3994, 0.8025, 0.3350, 0.3350, 0.4400, 0.5565, 0.5025], [0.6115, 0.4081, 0.6725, 0.2433, 0.4087, 0.1933, 0.5167, 0.5544], [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033]]], device='cuda:0') loss_train_step before backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0024, device='cuda:0', grad_fn=) loss_train: 0.05856394310831092 step: 39 running loss: 0.0015016395668797672 Train Steps: 39/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400], [0.6198, 0.4101, 0.8838, 0.5283, 0.3763, 0.5267, 0.5913, 0.5567], [0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378], [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650], [0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485], [0.6250, 0.4106, 0.8700, 0.3717, 0.3588, 0.4967, 0.6038, 0.5167], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550], [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4798, 0.3501, 0.8590, 0.4827, 0.4369, 0.5005, 0.6019, 0.5522], [0.5323, 0.3763, 0.8803, 0.5342, 0.3938, 0.5222, 0.5869, 0.5534], [0.5625, 0.3739, 0.8506, 0.5645, 0.3774, 0.4322, 0.5961, 0.5285], [0.4954, 0.3495, 0.8718, 0.4683, 0.4258, 0.4900, 0.5436, 0.5614], [0.5050, 0.3452, 0.7314, 0.2752, 0.3728, 0.3034, 0.5197, 0.5486], [0.5076, 0.3545, 0.8631, 0.3498, 0.3658, 0.4916, 0.6479, 0.5467], [0.5609, 0.3935, 0.7184, 0.2431, 0.4048, 0.2111, 0.5341, 0.5568], [0.5422, 0.3743, 0.7302, 0.2188, 0.3692, 0.3188, 0.6069, 0.5689]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6201, 0.3970, 0.8413, 0.4950, 0.4412, 0.5183, 0.6087, 0.5400], [0.6198, 0.4101, 0.8838, 0.5283, 0.3762, 0.5267, 0.5913, 0.5567], [0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378], [0.6098, 0.3991, 0.8637, 0.4717, 0.4263, 0.4967, 0.5213, 0.5650], [0.6123, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485], [0.6250, 0.4105, 0.8700, 0.3717, 0.3587, 0.4967, 0.6037, 0.5167], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5362, 0.5550], [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.060194525722181424 step: 40 running loss: 0.0015048631430545356 Train Steps: 40/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667], [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332], [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915], [0.6118, 0.4052, 0.8463, 0.3917, 0.3538, 0.3450, 0.5053, 0.5593], [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869], [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5737, 0.5350], [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5218, 0.3743, 0.8242, 0.3354, 0.3818, 0.3548, 0.5813, 0.5783], [0.5036, 0.3452, 0.7850, 0.4899, 0.3735, 0.4991, 0.6932, 0.5218], [0.5335, 0.3687, 0.8543, 0.4282, 0.3633, 0.3501, 0.5961, 0.5151], [0.5147, 0.3757, 0.8075, 0.3539, 0.3500, 0.3448, 0.4714, 0.5646], [0.5429, 0.3846, 0.7967, 0.2209, 0.4756, 0.2256, 0.6027, 0.5078], [0.5002, 0.3524, 0.8492, 0.4275, 0.4547, 0.5480, 0.5696, 0.5613], [0.5504, 0.4018, 0.8314, 0.4504, 0.3838, 0.3229, 0.5412, 0.5547], [0.5075, 0.3631, 0.7072, 0.2028, 0.3909, 0.2826, 0.5505, 0.5508]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5987, 0.5667], [0.6289, 0.4019, 0.8112, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332], [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915], [0.6118, 0.4052, 0.8462, 0.3917, 0.3537, 0.3450, 0.5053, 0.5593], [0.6276, 0.4095, 0.8238, 0.2250, 0.4663, 0.1783, 0.6171, 0.4869], [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5738, 0.5350], [0.6114, 0.4018, 0.7212, 0.1967, 0.3762, 0.2700, 0.5875, 0.5533]]], device='cuda:0') loss_train_step before backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train: 0.062190487602492794 step: 41 running loss: 0.0015168411610364096 Train Steps: 41/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6167, 0.4048, 0.6831, 0.3639, 0.3763, 0.3017, 0.5700, 0.5883], [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058], [0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6268, 0.4061, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300], [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683], [0.6166, 0.4008, 0.8563, 0.5667, 0.4388, 0.4933, 0.5575, 0.5567]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5534, 0.3963, 0.7433, 0.2502, 0.4639, 0.1631, 0.5861, 0.5406], [0.4998, 0.3601, 0.7076, 0.3060, 0.3642, 0.3174, 0.5446, 0.5721], [0.5338, 0.3740, 0.8254, 0.3796, 0.3466, 0.4556, 0.5302, 0.5044], [0.5335, 0.3762, 0.8138, 0.5281, 0.4133, 0.4841, 0.5833, 0.5231], [0.5301, 0.3713, 0.8326, 0.2195, 0.4565, 0.2315, 0.6110, 0.5118], [0.5860, 0.4042, 0.8220, 0.4887, 0.3692, 0.5435, 0.6988, 0.5536], [0.5799, 0.4024, 0.8479, 0.4923, 0.3590, 0.3807, 0.5321, 0.5592], [0.5488, 0.3756, 0.8268, 0.5125, 0.4210, 0.4862, 0.5559, 0.5352]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6245, 0.4100, 0.7763, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6167, 0.4048, 0.6831, 0.3639, 0.3762, 0.3017, 0.5700, 0.5883], [0.6084, 0.4005, 0.8400, 0.4317, 0.3762, 0.4750, 0.5476, 0.5058], [0.6229, 0.4066, 0.8512, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6268, 0.4060, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300], [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5412, 0.5683], [0.6166, 0.4008, 0.8562, 0.5667, 0.4387, 0.4933, 0.5575, 0.5567]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.06355495491879992 step: 42 running loss: 0.0015132132123523792 Train Steps: 42/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6284, 0.4127, 0.8538, 0.5867, 0.4363, 0.5083, 0.6038, 0.5433], [0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6173, 0.4114, 0.7325, 0.2500, 0.4213, 0.1917, 0.5338, 0.5700], [ nan, nan, 0.8888, 0.3100, 0.5262, 0.2817, 0.7145, 0.6003], [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519], [0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378], [0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110], [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6403, 0.4367, 0.8090, 0.5827, 0.4012, 0.4859, 0.5567, 0.5294], [0.6944, 0.4701, 0.7308, 0.2138, 0.4012, 0.2153, 0.5763, 0.5016], [0.6507, 0.4582, 0.7047, 0.2664, 0.4054, 0.1817, 0.5152, 0.5649], [0.0979, 0.0755, 0.8524, 0.3215, 0.4901, 0.2879, 0.6982, 0.5806], [0.6130, 0.4266, 0.8024, 0.5167, 0.4311, 0.5133, 0.5184, 0.5435], [0.5819, 0.3945, 0.8331, 0.2693, 0.4923, 0.2523, 0.7011, 0.5474], [0.6537, 0.4636, 0.8480, 0.4701, 0.3679, 0.5577, 0.5616, 0.5122], [0.5964, 0.4112, 0.8555, 0.3988, 0.3533, 0.4023, 0.6510, 0.5319]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6284, 0.4127, 0.8537, 0.5867, 0.4363, 0.5083, 0.6037, 0.5433], [0.6202, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6173, 0.4114, 0.7325, 0.2500, 0.4212, 0.1917, 0.5337, 0.5700], [0.0000, 0.0000, 0.8888, 0.3100, 0.5263, 0.2817, 0.7145, 0.6003], [0.6100, 0.4016, 0.8600, 0.5067, 0.4613, 0.5233, 0.5086, 0.5519], [0.6293, 0.4097, 0.8800, 0.2517, 0.5263, 0.2600, 0.7430, 0.5378], [0.6186, 0.4129, 0.8900, 0.4567, 0.3938, 0.5417, 0.5734, 0.5110], [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.0646070629300084 step: 43 running loss: 0.0015024898355815905 Train Steps: 43/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6063, 0.6083], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817], [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567], [0.6357, 0.4118, 0.8400, 0.2500, 0.5413, 0.1633, 0.6725, 0.5586], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6535, 0.4311, 0.8405, 0.3824, 0.3650, 0.4421, 0.6013, 0.5688], [0.6359, 0.4276, 0.8393, 0.3346, 0.3727, 0.3086, 0.5698, 0.5271], [0.6506, 0.4239, 0.8290, 0.5575, 0.3890, 0.4996, 0.7215, 0.5566], [0.5780, 0.3946, 0.8061, 0.3165, 0.4428, 0.2544, 0.5824, 0.5658], [0.6421, 0.4263, 0.8572, 0.4694, 0.4742, 0.5226, 0.5898, 0.5518], [0.5819, 0.3656, 0.8441, 0.2752, 0.5485, 0.1852, 0.6449, 0.5419], [0.6521, 0.4286, 0.7027, 0.2786, 0.4204, 0.2309, 0.5696, 0.5364], [0.6317, 0.4114, 0.8712, 0.4954, 0.4499, 0.5450, 0.6113, 0.5097]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6062, 0.6083], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817], [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567], [0.6357, 0.4118, 0.8400, 0.2500, 0.5412, 0.1633, 0.6725, 0.5586], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.6250, 0.4054, 0.8770, 0.4723, 0.4663, 0.5367, 0.6162, 0.5433]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.06501127360388637 step: 44 running loss: 0.001477528945542872 Train Steps: 44/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117], [ nan, nan, 0.8463, 0.2550, 0.5850, 0.2133, 0.7129, 0.6072], [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600], [ nan, nan, 0.7335, 0.2569, 0.3788, 0.2667, 0.5066, 0.5578], [0.6153, 0.4117, 0.8688, 0.5167, 0.4895, 0.5647, 0.5524, 0.5136], [0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6216, 0.4099, 0.7225, 0.2033, 0.4188, 0.2217, 0.5975, 0.5283], [0.6268, 0.4061, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7726, 0.4826, 0.7481, 0.2335, 0.4231, 0.2256, 0.5740, 0.5187], [0.3963, 0.2433, 0.8701, 0.3023, 0.5606, 0.2781, 0.7299, 0.5774], [0.7517, 0.4843, 0.8652, 0.5935, 0.4058, 0.5193, 0.5969, 0.5574], [0.2212, 0.1395, 0.7287, 0.2680, 0.3992, 0.2929, 0.5313, 0.5597], [0.6715, 0.4485, 0.8737, 0.5333, 0.4792, 0.5533, 0.5826, 0.5285], [0.7009, 0.4446, 0.7609, 0.2963, 0.4872, 0.1567, 0.6087, 0.5468], [0.7345, 0.4826, 0.7151, 0.2483, 0.4275, 0.2223, 0.5629, 0.5467], [0.7022, 0.4410, 0.8480, 0.2651, 0.4787, 0.2276, 0.6324, 0.5189]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5788, 0.5117], [0.0000, 0.0000, 0.8462, 0.2550, 0.5850, 0.2133, 0.7129, 0.6072], [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600], [0.0000, 0.0000, 0.7335, 0.2569, 0.3787, 0.2667, 0.5066, 0.5578], [0.6154, 0.4117, 0.8687, 0.5167, 0.4895, 0.5647, 0.5524, 0.5136], [0.6245, 0.4100, 0.7763, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6216, 0.4099, 0.7225, 0.2033, 0.4187, 0.2217, 0.5975, 0.5283], [0.6268, 0.4060, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300]]], device='cuda:0') loss_train_step before backward: tensor(0.0062, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0062, device='cuda:0', grad_fn=) loss_train: 0.07122712582349777 step: 45 running loss: 0.0015828250182999505 Train Steps: 45/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6268, 0.4029, 0.8500, 0.2683, 0.3937, 0.3500, 0.6860, 0.5297], [0.6346, 0.4092, 0.7712, 0.5917, 0.4037, 0.4767, 0.7343, 0.5725], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097], [0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378], [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7331, 0.4592, 0.7735, 0.2241, 0.4503, 0.1924, 0.5746, 0.5400], [0.6419, 0.4064, 0.8766, 0.4838, 0.4706, 0.4968, 0.5743, 0.5954], [0.7422, 0.4753, 0.8519, 0.2872, 0.4188, 0.3272, 0.6793, 0.5440], [0.6730, 0.4350, 0.8086, 0.5327, 0.4291, 0.4629, 0.6900, 0.5999], [0.6489, 0.4106, 0.8866, 0.4799, 0.4025, 0.5063, 0.6117, 0.5371], [0.6957, 0.4588, 0.8931, 0.4718, 0.4490, 0.5881, 0.5687, 0.5560], [0.6324, 0.4016, 0.9098, 0.2668, 0.5592, 0.2225, 0.7408, 0.5735], [0.6425, 0.4179, 0.9084, 0.5664, 0.4185, 0.4261, 0.5618, 0.6009]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5788, 0.5117], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6268, 0.4029, 0.8500, 0.2683, 0.3938, 0.3500, 0.6860, 0.5297], [0.6346, 0.4092, 0.7713, 0.5917, 0.4038, 0.4767, 0.7343, 0.5725], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097], [0.6293, 0.4097, 0.8800, 0.2517, 0.5263, 0.2600, 0.7430, 0.5378], [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.07247095857746899 step: 46 running loss: 0.001575455621249326 Train Steps: 46/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6117, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103], [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6252, 0.4158, 0.8988, 0.4083, 0.3788, 0.4783, 0.6225, 0.5633], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285], [0.6307, 0.4029, 0.8650, 0.5200, 0.3763, 0.4017, 0.7311, 0.5366], [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6574, 0.4007, 0.7161, 0.2415, 0.4372, 0.2182, 0.5540, 0.5274], [0.7188, 0.4395, 0.9126, 0.3551, 0.4240, 0.3446, 0.6664, 0.5151], [0.6449, 0.4115, 0.9352, 0.4280, 0.4575, 0.4946, 0.6595, 0.5656], [0.6935, 0.4357, 0.7239, 0.3226, 0.4177, 0.2775, 0.5561, 0.5865], [0.6732, 0.4144, 0.7784, 0.2303, 0.4354, 0.2535, 0.6162, 0.5620], [0.6671, 0.4141, 0.8405, 0.4290, 0.4107, 0.4286, 0.5424, 0.5497], [0.6732, 0.4279, 0.9269, 0.5229, 0.4582, 0.3990, 0.7307, 0.5606], [0.6479, 0.4105, 0.8007, 0.3074, 0.4547, 0.2792, 0.6077, 0.6354]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6116, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103], [0.6255, 0.4017, 0.8687, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6252, 0.4158, 0.8988, 0.4083, 0.3787, 0.4783, 0.6225, 0.5633], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6114, 0.4018, 0.7212, 0.1967, 0.3762, 0.2700, 0.5875, 0.5533], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285], [0.6307, 0.4029, 0.8650, 0.5200, 0.3762, 0.4017, 0.7311, 0.5366], [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.07407605787739158 step: 47 running loss: 0.001576086337816842 Train Steps: 47/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748], [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550], [0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103], [0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071], [0.6075, 0.4007, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6368, 0.3935, 0.8968, 0.4891, 0.4338, 0.3506, 0.7530, 0.5938], [0.6039, 0.3732, 0.8718, 0.3862, 0.3848, 0.3704, 0.5681, 0.5286], [0.6556, 0.4214, 0.8814, 0.4510, 0.5159, 0.4949, 0.6190, 0.5738], [0.6569, 0.4089, 0.9095, 0.3692, 0.4205, 0.2534, 0.6525, 0.5490], [0.6146, 0.3732, 0.8748, 0.4358, 0.4161, 0.4537, 0.5730, 0.5365], [0.7169, 0.4363, 0.7584, 0.1976, 0.4032, 0.2690, 0.6160, 0.5167], [0.6273, 0.3904, 0.8306, 0.4688, 0.4328, 0.4919, 0.5570, 0.5485], [0.6928, 0.4262, 0.8915, 0.5248, 0.4117, 0.3709, 0.5685, 0.5948]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6336, 0.4191, 0.8938, 0.5167, 0.3938, 0.3517, 0.7343, 0.5748], [0.6082, 0.4024, 0.8737, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550], [0.6277, 0.4117, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103], [0.6163, 0.4114, 0.7650, 0.2017, 0.3762, 0.2867, 0.5631, 0.5071], [0.6075, 0.4006, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5412, 0.5683]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.07483209832571447 step: 48 running loss: 0.0015590020484523848 Train Steps: 48/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6197, 0.3986, 0.8800, 0.4617, 0.4188, 0.4783, 0.5687, 0.5550], [0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980], [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567], [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377], [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.1510, 0.0802, 0.8048, 0.3031, 0.3931, 0.2843, 0.6019, 0.5653], [0.7519, 0.4651, 0.8721, 0.4573, 0.4482, 0.5135, 0.6305, 0.5623], [0.7314, 0.4637, 0.8811, 0.5660, 0.4295, 0.4587, 0.5821, 0.5647], [0.7187, 0.4577, 0.8832, 0.4578, 0.4222, 0.4926, 0.6105, 0.5489], [0.7343, 0.4611, 0.7655, 0.2794, 0.3611, 0.2828, 0.5849, 0.4956], [0.0396, 0.0147, 0.7697, 0.2914, 0.4217, 0.2391, 0.5808, 0.5600], [0.7838, 0.4756, 0.8688, 0.5498, 0.4112, 0.4997, 0.6354, 0.5302], [0.7763, 0.4855, 0.8959, 0.4841, 0.3739, 0.4266, 0.6179, 0.5216]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6197, 0.3986, 0.8800, 0.4617, 0.4187, 0.4783, 0.5688, 0.5550], [0.6079, 0.3964, 0.7420, 0.2958, 0.3562, 0.2917, 0.5351, 0.4980], [0.0000, 0.0000, 0.7515, 0.2708, 0.3988, 0.2267, 0.5163, 0.5567], [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377], [0.6236, 0.3965, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183]]], device='cuda:0') loss_train_step before backward: tensor(0.0030, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0030, device='cuda:0', grad_fn=) loss_train: 0.07787081459537148 step: 49 running loss: 0.0015892002978647242 Train Steps: 49/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.7225, 0.2167, 0.3987, 0.2283, 0.5427, 0.5181], [0.6250, 0.4146, 0.8838, 0.3933, 0.3588, 0.4283, 0.6162, 0.5367], [0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609], [0.6245, 0.4115, 0.8700, 0.4883, 0.4625, 0.5517, 0.6100, 0.5217], [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467], [0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250], [0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[-0.0325, -0.0318, 0.7265, 0.2437, 0.4054, 0.2141, 0.5640, 0.5116], [ 0.5875, 0.3782, 0.8540, 0.4057, 0.3558, 0.4153, 0.6077, 0.5356], [ 0.6916, 0.4526, 0.8666, 0.4584, 0.4000, 0.5510, 0.5852, 0.5289], [ 0.6534, 0.4066, 0.8812, 0.5019, 0.4452, 0.5662, 0.6000, 0.5215], [ 0.6641, 0.4348, 0.6917, 0.2722, 0.3654, 0.2443, 0.5700, 0.5324], [ 0.6990, 0.4371, 0.8753, 0.3732, 0.3904, 0.2664, 0.6206, 0.4665], [ 0.6395, 0.3998, 0.9076, 0.4593, 0.3602, 0.4819, 0.6314, 0.5227], [ 0.5846, 0.3655, 0.8533, 0.2823, 0.4021, 0.2868, 0.7068, 0.5403]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.7225, 0.2167, 0.3988, 0.2283, 0.5427, 0.5181], [0.6250, 0.4146, 0.8838, 0.3933, 0.3587, 0.4283, 0.6162, 0.5367], [0.6064, 0.4019, 0.8650, 0.4517, 0.4038, 0.5367, 0.5703, 0.5609], [0.6245, 0.4115, 0.8700, 0.4883, 0.4625, 0.5517, 0.6100, 0.5217], [0.6162, 0.4134, 0.6700, 0.2467, 0.3963, 0.2533, 0.5738, 0.5467], [0.6277, 0.4036, 0.8687, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6263, 0.4065, 0.9038, 0.4317, 0.3587, 0.4550, 0.6325, 0.5250], [0.6286, 0.4086, 0.8408, 0.2801, 0.4162, 0.2800, 0.6725, 0.5393]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.07848756696330383 step: 50 running loss: 0.0015697513392660767 Train Steps: 50/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391], [0.6222, 0.4169, 0.8638, 0.5650, 0.4313, 0.4783, 0.5637, 0.5633], [ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235], [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284], [ nan, nan, 0.8300, 0.3150, 0.3588, 0.3383, 0.5208, 0.5194], [0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748], [0.6064, 0.3953, 0.8738, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6583, 0.4061, 0.7852, 0.2628, 0.3901, 0.2268, 0.5825, 0.5011], [0.6119, 0.4043, 0.8460, 0.5443, 0.3911, 0.4834, 0.5273, 0.5403], [0.0739, 0.0346, 0.7801, 0.3044, 0.3294, 0.3241, 0.5190, 0.5207], [0.6127, 0.3999, 0.8732, 0.4730, 0.3982, 0.5596, 0.6111, 0.5051], [0.6407, 0.4011, 0.7635, 0.2083, 0.4314, 0.1839, 0.5845, 0.4947], [0.1436, 0.0885, 0.7909, 0.3158, 0.3326, 0.3305, 0.5361, 0.5008], [0.6194, 0.4075, 0.8671, 0.4859, 0.3701, 0.3710, 0.7046, 0.5399], [0.6698, 0.4239, 0.8582, 0.4529, 0.3303, 0.4754, 0.5364, 0.4863]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391], [0.6222, 0.4169, 0.8637, 0.5650, 0.4313, 0.4783, 0.5638, 0.5633], [0.0000, 0.0000, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235], [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284], [0.0000, 0.0000, 0.8300, 0.3150, 0.3587, 0.3383, 0.5208, 0.5194], [0.6336, 0.4191, 0.8938, 0.5167, 0.3938, 0.3517, 0.7343, 0.5748], [0.6064, 0.3952, 0.8737, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.07955434458563104 step: 51 running loss: 0.0015598891095221772 Train Steps: 51/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995], [0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167], [0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123], [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750], [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598], [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510], [ nan, nan, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6183, 0.4076, 0.8838, 0.4517, 0.3813, 0.4483, 0.5775, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.5704, 0.3827, 0.7387, 0.2036, 0.3578, 0.2297, 0.5657, 0.4697], [ 0.5516, 0.3512, 0.7640, 0.2236, 0.4117, 0.1581, 0.5666, 0.4943], [ 0.5503, 0.3709, 0.8578, 0.3682, 0.3118, 0.4334, 0.5395, 0.4950], [ 0.5605, 0.3689, 0.8584, 0.5159, 0.3863, 0.4964, 0.5315, 0.5135], [ 0.5507, 0.3585, 0.8238, 0.5576, 0.4316, 0.4736, 0.5286, 0.5450], [ 0.5587, 0.3720, 0.6769, 0.2708, 0.3132, 0.3024, 0.4931, 0.5178], [-0.1807, -0.1097, 0.8726, 0.3411, 0.4731, 0.2189, 0.6576, 0.5594], [ 0.5047, 0.3208, 0.8690, 0.4605, 0.3415, 0.4591, 0.5479, 0.5489]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995], [0.6204, 0.4007, 0.7837, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167], [0.6213, 0.4131, 0.8438, 0.3550, 0.3512, 0.4400, 0.5716, 0.5123], [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750], [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598], [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510], [0.0000, 0.0000, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6183, 0.4076, 0.8838, 0.4517, 0.3812, 0.4483, 0.5775, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.08149451989447698 step: 52 running loss: 0.001567202305663019 Train Steps: 52/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377], [0.6205, 0.4081, 0.8950, 0.4017, 0.3788, 0.4700, 0.5963, 0.5667], [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283], [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954], [ nan, nan, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6065, 0.3959, 0.8548, 0.5664, 0.3848, 0.4831, 0.5509, 0.5286], [ 0.6251, 0.4124, 0.8550, 0.3978, 0.3658, 0.4721, 0.5616, 0.5464], [-0.0914, -0.0326, 0.6882, 0.2071, 0.3913, 0.2299, 0.5179, 0.5272], [ 0.5802, 0.3960, 0.7924, 0.1981, 0.4634, 0.1556, 0.6002, 0.4862], [ 0.5282, 0.3601, 0.8603, 0.5014, 0.3412, 0.3819, 0.5508, 0.5523], [ 0.5425, 0.3861, 0.8650, 0.4160, 0.3282, 0.3782, 0.5710, 0.5371], [ 0.6114, 0.4252, 0.8005, 0.3769, 0.3246, 0.4035, 0.5220, 0.4932], [-0.0446, -0.0088, 0.6980, 0.1995, 0.4239, 0.2405, 0.5390, 0.5463]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377], [0.6205, 0.4081, 0.8950, 0.4017, 0.3787, 0.4700, 0.5962, 0.5667], [0.0000, 0.0000, 0.6900, 0.1917, 0.3938, 0.2367, 0.5240, 0.5246], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6262, 0.4163, 0.8850, 0.5183, 0.3762, 0.4150, 0.6025, 0.5500], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283], [0.6076, 0.3952, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954], [0.0000, 0.0000, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.08231787505792454 step: 53 running loss: 0.0015531674539231044 Train Steps: 53/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150], [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431], [0.6110, 0.4047, 0.8700, 0.4483, 0.3713, 0.3967, 0.5088, 0.5517], [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583], [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4633, 0.3053, 0.7813, 0.2637, 0.4218, 0.2074, 0.5650, 0.5414], [0.4423, 0.3266, 0.8592, 0.4890, 0.4347, 0.4827, 0.4844, 0.5379], [0.4234, 0.3148, 0.8460, 0.4503, 0.3582, 0.4114, 0.4690, 0.5458], [0.4643, 0.3248, 0.8627, 0.5094, 0.4220, 0.4801, 0.5468, 0.5422], [0.4713, 0.3208, 0.8873, 0.5206, 0.4017, 0.5500, 0.6978, 0.5498], [0.4593, 0.3276, 0.8906, 0.4396, 0.3651, 0.4203, 0.5323, 0.5475], [0.4902, 0.3599, 0.8177, 0.3682, 0.3393, 0.4011, 0.5317, 0.5457], [0.5154, 0.3562, 0.7241, 0.1829, 0.3850, 0.2528, 0.5751, 0.5372]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150], [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431], [0.6110, 0.4047, 0.8700, 0.4483, 0.3713, 0.3967, 0.5088, 0.5517], [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6212, 0.4033, 0.8938, 0.4167, 0.3812, 0.4267, 0.5612, 0.5583], [0.6058, 0.3978, 0.8288, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483]]], device='cuda:0') loss_train_step before backward: tensor(0.0039, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0039, device='cuda:0', grad_fn=) loss_train: 0.08622756163822487 step: 54 running loss: 0.0015968066970041643 Train Steps: 54/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6284, 0.4127, 0.8538, 0.5867, 0.4363, 0.5083, 0.6038, 0.5433], [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323], [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600], [ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533], [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.4549, 0.3279, 0.8062, 0.3474, 0.4819, 0.2351, 0.5100, 0.6041], [ 0.4452, 0.3079, 0.8588, 0.5672, 0.4430, 0.4954, 0.5422, 0.5496], [ 0.5699, 0.3944, 0.7603, 0.1899, 0.4359, 0.2076, 0.5866, 0.5161], [ 0.5667, 0.3971, 0.7179, 0.2186, 0.3601, 0.3267, 0.5661, 0.5567], [-0.0764, -0.0384, 0.7836, 0.2590, 0.3608, 0.2841, 0.4919, 0.5454], [ 0.4562, 0.3183, 0.9138, 0.3884, 0.4177, 0.3604, 0.6740, 0.5921], [ 0.5548, 0.4049, 0.7312, 0.2623, 0.4146, 0.2199, 0.5294, 0.5516], [ 0.5124, 0.3712, 0.8831, 0.3997, 0.3388, 0.3850, 0.5597, 0.5397]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6299, 0.4303, 0.7962, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6284, 0.4127, 0.8537, 0.5867, 0.4363, 0.5083, 0.6037, 0.5433], [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323], [0.6132, 0.4066, 0.7259, 0.2402, 0.3587, 0.3300, 0.6000, 0.5600], [0.0000, 0.0000, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533], [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283]]], device='cuda:0') loss_train_step before backward: tensor(0.0029, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0029, device='cuda:0', grad_fn=) loss_train: 0.0891112532117404 step: 55 running loss: 0.0016202046038498255 Train Steps: 55/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6353, 0.4128, 0.9138, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991], [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050], [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064], [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938], [0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391], [0.6133, 0.4066, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.2841, 0.2073, 0.8971, 0.3288, 0.4746, 0.3006, 0.6896, 0.6023], [0.4716, 0.3356, 0.9035, 0.3814, 0.4286, 0.3491, 0.6755, 0.6058], [0.5666, 0.3886, 0.8688, 0.3893, 0.3700, 0.4776, 0.5625, 0.5985], [0.5145, 0.3731, 0.8393, 0.5072, 0.3919, 0.4774, 0.6830, 0.5975], [0.5068, 0.3861, 0.8645, 0.4706, 0.4755, 0.5139, 0.5026, 0.5105], [0.5235, 0.3674, 0.8710, 0.5243, 0.4125, 0.4725, 0.5630, 0.5267], [0.5493, 0.3891, 0.8318, 0.2677, 0.4237, 0.2318, 0.5512, 0.5450], [0.5084, 0.3757, 0.6835, 0.2245, 0.3632, 0.2362, 0.4987, 0.5296]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6353, 0.4128, 0.9137, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991], [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6200, 0.4112, 0.8863, 0.4100, 0.3638, 0.4917, 0.6087, 0.6050], [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6102, 0.4005, 0.8687, 0.5100, 0.4812, 0.5400, 0.5404, 0.5064], [0.6334, 0.4037, 0.8637, 0.5733, 0.4013, 0.4717, 0.6369, 0.4938], [0.6199, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391], [0.6133, 0.4065, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358]]], device='cuda:0') loss_train_step before backward: tensor(0.0045, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0045, device='cuda:0', grad_fn=) loss_train: 0.09363721316913143 step: 56 running loss: 0.0016720930923059183 Train Steps: 56/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6117, 0.4019, 0.8538, 0.4067, 0.3513, 0.3583, 0.5663, 0.5133], [0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927], [0.6296, 0.4045, 0.9138, 0.4100, 0.4232, 0.4242, 0.7422, 0.5297], [0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5629, 0.4024, 0.8491, 0.5074, 0.3920, 0.4756, 0.7119, 0.6109], [0.5554, 0.3866, 0.8785, 0.4322, 0.3812, 0.4299, 0.5717, 0.5626], [0.5874, 0.3855, 0.8805, 0.4502, 0.4399, 0.5208, 0.6653, 0.5731], [0.5419, 0.3820, 0.8656, 0.5302, 0.4593, 0.4908, 0.5758, 0.5832], [0.6008, 0.4064, 0.8500, 0.3611, 0.3444, 0.3462, 0.5353, 0.5433], [0.5880, 0.4068, 0.8797, 0.5273, 0.3888, 0.4454, 0.5944, 0.5271], [0.5795, 0.4050, 0.9005, 0.3755, 0.4074, 0.4259, 0.6626, 0.5615], [0.6131, 0.4276, 0.8416, 0.2677, 0.4248, 0.2244, 0.5816, 0.5584]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6229, 0.4066, 0.8512, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6116, 0.4019, 0.8537, 0.4067, 0.3512, 0.3583, 0.5663, 0.5133], [0.6321, 0.4048, 0.8737, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927], [0.6296, 0.4045, 0.9137, 0.4100, 0.4232, 0.4242, 0.7422, 0.5297], [0.6199, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.09452643216354772 step: 57 running loss: 0.0016583584590096092 Train Steps: 57/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609], [0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5613, 0.5900], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810], [0.6120, 0.4014, 0.6863, 0.2817, 0.3700, 0.2783, 0.5513, 0.5667], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283], [ nan, nan, 0.7192, 0.2346, 0.4037, 0.2050, 0.5138, 0.5650]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6112, 0.4222, 0.8758, 0.4470, 0.4294, 0.5529, 0.6551, 0.5872], [0.6808, 0.4543, 0.8444, 0.5078, 0.3648, 0.3772, 0.6150, 0.5839], [0.6741, 0.4665, 0.8564, 0.2615, 0.5054, 0.1738, 0.6894, 0.5344], [0.6551, 0.4307, 0.8836, 0.5045, 0.4541, 0.5239, 0.6172, 0.5367], [0.6273, 0.4249, 0.8685, 0.2997, 0.4325, 0.2790, 0.6880, 0.5231], [0.5907, 0.3875, 0.7082, 0.2678, 0.3756, 0.2957, 0.5998, 0.5691], [0.7871, 0.5148, 0.9175, 0.3656, 0.4089, 0.2759, 0.7013, 0.5536], [0.1344, 0.0876, 0.7245, 0.2278, 0.4096, 0.2233, 0.5401, 0.5740]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6064, 0.4019, 0.8650, 0.4517, 0.4038, 0.5367, 0.5703, 0.5609], [0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5612, 0.5900], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810], [0.6120, 0.4013, 0.6862, 0.2817, 0.3700, 0.2783, 0.5512, 0.5667], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6187, 0.5283], [0.0000, 0.0000, 0.7192, 0.2346, 0.4038, 0.2050, 0.5138, 0.5650]]], device='cuda:0') loss_train_step before backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train: 0.09650458296528086 step: 58 running loss: 0.0016638721200910494 Train Steps: 58/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767], [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5837, 0.5583], [0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833], [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633], [ nan, nan, 0.6793, 0.2110, 0.4012, 0.2167, 0.5112, 0.5583], [0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7382, 0.4823, 0.8761, 0.5222, 0.4304, 0.4852, 0.6209, 0.5718], [0.7312, 0.4900, 0.8813, 0.4541, 0.4890, 0.5090, 0.6586, 0.5728], [0.6612, 0.4320, 0.7609, 0.2048, 0.4153, 0.2033, 0.6580, 0.5155], [0.6932, 0.4531, 0.8719, 0.5102, 0.4333, 0.4629, 0.6154, 0.5461], [0.7277, 0.4609, 0.8940, 0.5155, 0.4000, 0.4746, 0.6561, 0.5016], [0.7316, 0.4616, 0.9022, 0.5313, 0.4126, 0.5640, 0.7744, 0.5645], [0.1607, 0.0987, 0.6774, 0.2018, 0.4002, 0.1998, 0.5467, 0.5342], [0.7042, 0.4623, 0.8553, 0.5422, 0.4580, 0.5011, 0.6179, 0.5370]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6148, 0.4076, 0.8666, 0.4820, 0.4137, 0.5067, 0.5250, 0.5767], [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5838, 0.5583], [0.6202, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6037, 0.4833], [0.6293, 0.4024, 0.8750, 0.5000, 0.4013, 0.5733, 0.7121, 0.5633], [0.0000, 0.0000, 0.6793, 0.2110, 0.4013, 0.2167, 0.5113, 0.5583], [0.6075, 0.4000, 0.8512, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280]]], device='cuda:0') loss_train_step before backward: tensor(0.0030, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0030, device='cuda:0', grad_fn=) loss_train: 0.09950864681741223 step: 59 running loss: 0.0016865872341934275 Train Steps: 59/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167], [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124], [0.6182, 0.3930, 0.8841, 0.3892, 0.3556, 0.4967, 0.6222, 0.5279], [0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467], [0.6193, 0.4108, 0.7438, 0.2700, 0.3650, 0.3683, 0.6238, 0.5717], [0.6259, 0.4133, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804], [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6482, 0.3921, 0.7578, 0.2343, 0.4396, 0.1740, 0.6194, 0.5175], [0.7013, 0.4344, 0.8072, 0.2890, 0.4050, 0.2726, 0.6344, 0.5235], [0.6638, 0.3961, 0.8857, 0.4162, 0.3702, 0.4941, 0.6593, 0.5240], [0.6293, 0.3778, 0.8640, 0.6180, 0.4028, 0.4876, 0.6442, 0.5462], [0.6033, 0.3709, 0.7571, 0.2885, 0.3631, 0.3846, 0.6689, 0.5528], [0.7017, 0.4492, 0.8180, 0.2445, 0.4934, 0.1715, 0.6521, 0.4938], [0.6669, 0.3932, 0.9262, 0.4911, 0.3702, 0.4925, 0.6464, 0.5024], [0.7239, 0.4449, 0.8679, 0.4768, 0.3647, 0.3830, 0.6458, 0.5466]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6204, 0.4007, 0.7837, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167], [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124], [0.6182, 0.3930, 0.8841, 0.3892, 0.3556, 0.4967, 0.6222, 0.5279], [0.6273, 0.4143, 0.8750, 0.5700, 0.3988, 0.4717, 0.6012, 0.5467], [0.6193, 0.4108, 0.7437, 0.2700, 0.3650, 0.3683, 0.6237, 0.5717], [0.6259, 0.4132, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6263, 0.4038, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804], [0.6252, 0.4162, 0.8662, 0.4467, 0.3625, 0.3567, 0.6037, 0.5533]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.10043607087573037 step: 60 running loss: 0.0016739345145955061 Train Steps: 60/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6109, 0.4015, 0.7668, 0.3639, 0.3513, 0.3667, 0.5200, 0.5641], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.6218, 0.4137, 0.7263, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748], [0.6226, 0.4103, 0.8575, 0.3450, 0.4388, 0.2067, 0.5787, 0.5383], [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012], [0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6287, 0.3835, 0.7721, 0.3537, 0.3589, 0.3807, 0.5404, 0.5136], [0.7034, 0.4113, 0.8991, 0.4467, 0.3619, 0.3990, 0.6077, 0.5131], [0.6951, 0.4118, 0.7277, 0.2470, 0.4020, 0.2793, 0.6083, 0.5252], [0.7054, 0.4191, 0.8589, 0.4918, 0.3990, 0.4911, 0.5657, 0.5131], [0.7142, 0.4282, 0.9088, 0.4964, 0.4427, 0.3011, 0.7445, 0.5255], [0.7862, 0.4827, 0.8517, 0.3759, 0.4657, 0.2416, 0.5966, 0.5169], [0.7011, 0.4088, 0.8817, 0.5840, 0.3721, 0.4897, 0.6431, 0.4846], [0.5986, 0.3631, 0.7848, 0.2676, 0.3712, 0.3852, 0.6174, 0.5040]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6109, 0.4015, 0.7668, 0.3639, 0.3512, 0.3667, 0.5200, 0.5641], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.6218, 0.4137, 0.7262, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6329, 0.4196, 0.9237, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748], [0.6226, 0.4103, 0.8575, 0.3450, 0.4387, 0.2067, 0.5788, 0.5383], [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012], [0.6175, 0.4091, 0.7862, 0.2800, 0.3638, 0.3583, 0.6187, 0.5433]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.10195993207162246 step: 61 running loss: 0.0016714742962561059 Train Steps: 61/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.6216, 0.4167, 0.8588, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6247, 0.3661, 0.7157, 0.2869, 0.4271, 0.2154, 0.5333, 0.5766], [0.6447, 0.3784, 0.8774, 0.2905, 0.5059, 0.2369, 0.7035, 0.5209], [0.7440, 0.4345, 0.8946, 0.6275, 0.4013, 0.5598, 0.6141, 0.5292], [0.7049, 0.4133, 0.8685, 0.5509, 0.4432, 0.5565, 0.5157, 0.5046], [0.6179, 0.3676, 0.7443, 0.2568, 0.4057, 0.2316, 0.5449, 0.5404], [0.6160, 0.3691, 0.7046, 0.2429, 0.4079, 0.1970, 0.5292, 0.5258], [0.6514, 0.3870, 0.9177, 0.3847, 0.3822, 0.4629, 0.6947, 0.5112], [0.6073, 0.3739, 0.7232, 0.2653, 0.4281, 0.1929, 0.5387, 0.5596]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.6216, 0.4167, 0.8587, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6193, 0.4050, 0.7312, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6154, 0.4112, 0.7038, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [0.6218, 0.4185, 0.7337, 0.2650, 0.4625, 0.1950, 0.5688, 0.5800]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.10285078029846773 step: 62 running loss: 0.0016588835532010922 Train Steps: 62/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494], [ nan, nan, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6145, 0.4007, 0.8775, 0.4533, 0.4562, 0.5533, 0.6088, 0.5533], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810], [0.6126, 0.4039, 0.8237, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6361, 0.4076, 0.8862, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7079, 0.4305, 0.9018, 0.3640, 0.4346, 0.3790, 0.6754, 0.5370], [0.1807, 0.0795, 0.7081, 0.2036, 0.4204, 0.1630, 0.4939, 0.5363], [0.7387, 0.4639, 0.8380, 0.2717, 0.4993, 0.1799, 0.5849, 0.5192], [0.7147, 0.4226, 0.8808, 0.4658, 0.4636, 0.5462, 0.5793, 0.5490], [0.7203, 0.4444, 0.8488, 0.3098, 0.4027, 0.2778, 0.5885, 0.4940], [0.6518, 0.3961, 0.8113, 0.4107, 0.3574, 0.3735, 0.5625, 0.5967], [0.6976, 0.4263, 0.8730, 0.4764, 0.3820, 0.5138, 0.5499, 0.5146], [0.7670, 0.4823, 0.8740, 0.5626, 0.3858, 0.4831, 0.6404, 0.5375]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494], [0.0000, 0.0000, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6145, 0.4007, 0.8775, 0.4533, 0.4563, 0.5533, 0.6087, 0.5533], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810], [0.6126, 0.4038, 0.8238, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6361, 0.4076, 0.8863, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297]]], device='cuda:0') loss_train_step before backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train: 0.10497850127285346 step: 63 running loss: 0.00166632541702942 Train Steps: 63/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800], [0.6273, 0.4100, 0.7137, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633], [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301], [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528], [0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6207, 0.4110, 0.8738, 0.5000, 0.4800, 0.5633, 0.6300, 0.5433], [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378], [0.6131, 0.4037, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5489, 0.3741, 0.7715, 0.3019, 0.3528, 0.2394, 0.5202, 0.5679], [0.5608, 0.3765, 0.7200, 0.2122, 0.3870, 0.2277, 0.5736, 0.5654], [0.6949, 0.4254, 0.8806, 0.4887, 0.4158, 0.4699, 0.6067, 0.5323], [0.6474, 0.4089, 0.9201, 0.3635, 0.4119, 0.3545, 0.7003, 0.5663], [0.6503, 0.4275, 0.8552, 0.5129, 0.4399, 0.4933, 0.5123, 0.5289], [0.7275, 0.4710, 0.8964, 0.4896, 0.4696, 0.5465, 0.5877, 0.5373], [0.6598, 0.4386, 0.8856, 0.5192, 0.3787, 0.4187, 0.4934, 0.5376], [0.5645, 0.3624, 0.7051, 0.2746, 0.3734, 0.2543, 0.5114, 0.5627]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5462, 0.5800], [0.6273, 0.4099, 0.7138, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633], [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301], [0.6332, 0.4165, 0.9100, 0.3350, 0.4187, 0.3683, 0.7438, 0.5528], [0.6075, 0.4000, 0.8512, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6207, 0.4110, 0.8737, 0.5000, 0.4800, 0.5633, 0.6300, 0.5433], [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378], [0.6131, 0.4036, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.10589164070552215 step: 64 running loss: 0.0016545568860237836 Train Steps: 64/90 Loss: 0.0017 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683], [0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748], [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461], [0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019], [0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843], [0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6221, 0.4283, 0.8628, 0.4962, 0.3900, 0.3418, 0.5310, 0.5983], [0.7085, 0.4884, 0.8724, 0.4659, 0.4139, 0.3262, 0.6839, 0.5850], [0.5233, 0.3546, 0.8350, 0.3252, 0.3570, 0.3759, 0.5416, 0.5449], [0.5835, 0.3883, 0.8672, 0.3821, 0.3898, 0.5025, 0.5429, 0.5312], [0.6488, 0.4269, 0.8925, 0.4275, 0.3925, 0.4656, 0.6009, 0.5248], [0.5921, 0.4094, 0.8436, 0.4683, 0.3680, 0.3727, 0.5096, 0.5457], [0.6816, 0.4483, 0.8707, 0.4532, 0.4092, 0.5143, 0.6279, 0.5406], [0.6375, 0.4225, 0.8894, 0.4271, 0.3852, 0.3125, 0.6002, 0.5315]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5412, 0.5683], [0.6336, 0.4191, 0.8938, 0.5167, 0.3938, 0.3517, 0.7343, 0.5748], [0.6058, 0.3978, 0.8288, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461], [0.6161, 0.4099, 0.8737, 0.4383, 0.3787, 0.5483, 0.5605, 0.5019], [0.6264, 0.4049, 0.8988, 0.4633, 0.3812, 0.4983, 0.6326, 0.4843], [0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.10696358111454174 step: 65 running loss: 0.0016455935556083344 Train Steps: 65/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609], [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611], [0.6102, 0.4020, 0.8638, 0.3717, 0.3625, 0.5017, 0.6038, 0.5500], [0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913], [ nan, nan, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748], [0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5393, 0.3889, 0.7657, 0.1986, 0.4130, 0.1993, 0.5852, 0.5261], [0.5829, 0.4071, 0.8496, 0.4321, 0.3864, 0.5271, 0.5797, 0.5473], [0.6193, 0.4327, 0.8669, 0.4604, 0.4150, 0.4514, 0.5412, 0.5538], [0.5629, 0.3962, 0.8623, 0.3543, 0.3623, 0.4686, 0.6162, 0.5318], [0.4795, 0.3447, 0.7762, 0.2536, 0.4611, 0.1480, 0.5585, 0.5314], [0.1735, 0.1281, 0.8874, 0.2998, 0.4933, 0.2157, 0.6412, 0.5723], [0.6627, 0.4651, 0.8768, 0.4995, 0.3844, 0.3360, 0.7009, 0.5691], [0.6295, 0.4397, 0.8468, 0.3628, 0.3846, 0.2779, 0.5188, 0.5498]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6202, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6064, 0.4019, 0.8650, 0.4517, 0.4038, 0.5367, 0.5703, 0.5609], [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611], [0.6102, 0.4020, 0.8637, 0.3717, 0.3625, 0.5017, 0.6037, 0.5500], [0.6286, 0.4078, 0.8062, 0.2267, 0.4787, 0.1533, 0.5953, 0.4913], [0.0000, 0.0000, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6336, 0.4191, 0.8938, 0.5167, 0.3938, 0.3517, 0.7343, 0.5748], [0.6186, 0.4080, 0.8625, 0.3483, 0.3787, 0.2650, 0.5320, 0.5272]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.10856805794173852 step: 66 running loss: 0.001644970574874826 Train Steps: 66/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320], [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268], [0.6203, 0.4021, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [0.6160, 0.4093, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5448, 0.3894, 0.8145, 0.2338, 0.4142, 0.2702, 0.6812, 0.5341], [0.5694, 0.3942, 0.8913, 0.4585, 0.3639, 0.3936, 0.5592, 0.5375], [0.5310, 0.3859, 0.8618, 0.2606, 0.4940, 0.1833, 0.6343, 0.5224], [0.5874, 0.4226, 0.8470, 0.5582, 0.4386, 0.4280, 0.5736, 0.5882], [0.5255, 0.3685, 0.8450, 0.4086, 0.3649, 0.4420, 0.5704, 0.5609], [0.5708, 0.3805, 0.8753, 0.4588, 0.4241, 0.4992, 0.6315, 0.5281], [0.5485, 0.3756, 0.8864, 0.5135, 0.3552, 0.4017, 0.6095, 0.5225], [0.5485, 0.3837, 0.8391, 0.4335, 0.3586, 0.4426, 0.5607, 0.5755]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320], [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6248, 0.4185, 0.8500, 0.5767, 0.4462, 0.4550, 0.5612, 0.5917], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268], [0.6203, 0.4020, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [0.6160, 0.4092, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.10938900749897584 step: 67 running loss: 0.0016326717537160573 Train Steps: 67/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750], [0.6100, 0.4071, 0.7601, 0.3444, 0.3400, 0.4117, 0.5625, 0.5617], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235], [0.6223, 0.4028, 0.8988, 0.4200, 0.3763, 0.5733, 0.6375, 0.5167], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6268, 0.4029, 0.8500, 0.2683, 0.3937, 0.3500, 0.6860, 0.5297], [0.6131, 0.4037, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680], [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4256, 0.3133, 0.7312, 0.2664, 0.4504, 0.1735, 0.5968, 0.5748], [0.5070, 0.3664, 0.8311, 0.3491, 0.3637, 0.4180, 0.6030, 0.5581], [0.5938, 0.4139, 0.9050, 0.5002, 0.4267, 0.5329, 0.6676, 0.5504], [0.5597, 0.3864, 0.9127, 0.4380, 0.4046, 0.5862, 0.6737, 0.5258], [0.4858, 0.3554, 0.8112, 0.3820, 0.3644, 0.4135, 0.5584, 0.5594], [0.6025, 0.4247, 0.8645, 0.2894, 0.3939, 0.3295, 0.7073, 0.5438], [0.5373, 0.3748, 0.7314, 0.3082, 0.4009, 0.2635, 0.5578, 0.5699], [0.5954, 0.4201, 0.8981, 0.4821, 0.4482, 0.4694, 0.5724, 0.5615]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5512, 0.5750], [0.6100, 0.4071, 0.7601, 0.3444, 0.3400, 0.4117, 0.5625, 0.5617], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235], [0.6223, 0.4028, 0.8988, 0.4200, 0.3762, 0.5733, 0.6375, 0.5167], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6268, 0.4029, 0.8500, 0.2683, 0.3938, 0.3500, 0.6860, 0.5297], [0.6131, 0.4036, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680], [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.11124868033220991 step: 68 running loss: 0.00163601000488544 Train Steps: 68/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197], [0.6069, 0.3975, 0.8625, 0.5083, 0.4388, 0.5483, 0.5650, 0.4967], [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633], [0.6317, 0.4038, 0.8287, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986], [0.6218, 0.4137, 0.7263, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783], [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633], [0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5563, 0.3870, 0.8563, 0.4827, 0.4403, 0.5231, 0.5433, 0.5044], [0.5492, 0.3759, 0.8491, 0.4923, 0.4246, 0.5409, 0.5634, 0.5205], [0.5895, 0.4069, 0.8284, 0.2998, 0.4732, 0.1896, 0.5892, 0.5576], [0.5430, 0.3776, 0.8512, 0.5306, 0.4005, 0.5172, 0.5993, 0.5526], [0.5404, 0.3731, 0.8525, 0.5451, 0.3619, 0.4570, 0.6332, 0.5041], [0.5671, 0.3874, 0.7579, 0.2432, 0.3995, 0.2573, 0.6219, 0.5672], [0.5508, 0.3591, 0.8674, 0.4709, 0.3768, 0.5651, 0.6735, 0.5446], [0.5503, 0.3572, 0.9081, 0.4160, 0.3851, 0.5056, 0.7052, 0.5347]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197], [0.6069, 0.3975, 0.8625, 0.5083, 0.4387, 0.5483, 0.5650, 0.4967], [0.6229, 0.4107, 0.8138, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6192, 0.4128, 0.8512, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633], [0.6317, 0.4038, 0.8288, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986], [0.6218, 0.4137, 0.7262, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783], [0.6293, 0.4024, 0.8750, 0.5000, 0.4013, 0.5733, 0.7121, 0.5633], [0.6250, 0.3993, 0.9137, 0.4333, 0.3762, 0.5217, 0.6995, 0.5320]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.11220953316660598 step: 69 running loss: 0.0016262251183566084 Train Steps: 69/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817], [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5787, 0.5550], [0.6060, 0.3924, 0.8450, 0.5717, 0.4200, 0.5217, 0.5253, 0.4752], [ nan, nan, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117], [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5894, 0.3943, 0.8446, 0.4361, 0.4505, 0.5642, 0.6108, 0.5490], [0.5916, 0.3702, 0.8096, 0.5563, 0.4293, 0.5186, 0.5392, 0.4968], [0.6256, 0.4082, 0.8094, 0.2928, 0.3440, 0.3792, 0.5929, 0.5366], [0.5960, 0.3767, 0.8206, 0.5812, 0.4219, 0.5179, 0.5555, 0.4948], [0.1566, 0.1218, 0.7175, 0.2339, 0.4316, 0.1995, 0.5217, 0.5481], [0.6159, 0.4071, 0.7354, 0.2169, 0.4185, 0.2362, 0.6050, 0.5164], [0.6355, 0.4033, 0.8359, 0.3240, 0.3578, 0.3834, 0.6605, 0.5133], [0.6122, 0.3977, 0.8758, 0.4785, 0.3735, 0.3831, 0.6451, 0.5342]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817], [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5788, 0.5550], [0.6060, 0.3924, 0.8450, 0.5717, 0.4200, 0.5217, 0.5253, 0.4752], [0.0000, 0.0000, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583], [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5788, 0.5117], [0.6255, 0.4017, 0.8687, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.11311025434406474 step: 70 running loss: 0.001615860776343782 Train Steps: 70/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750], [0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433], [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667], [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183], [0.6275, 0.4048, 0.8488, 0.2883, 0.4463, 0.2033, 0.6321, 0.5155], [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162], [0.6097, 0.4000, 0.7325, 0.2667, 0.3450, 0.3517, 0.5284, 0.5045], [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5502, 0.3574, 0.6851, 0.2700, 0.4356, 0.2077, 0.5738, 0.5523], [0.5865, 0.3874, 0.8746, 0.4779, 0.4575, 0.5594, 0.6084, 0.5240], [0.6299, 0.3938, 0.8675, 0.3658, 0.3588, 0.4424, 0.6396, 0.5507], [0.5341, 0.3422, 0.8595, 0.4866, 0.4463, 0.5918, 0.5941, 0.5159], [0.6748, 0.4351, 0.8300, 0.2889, 0.4593, 0.2303, 0.6556, 0.5105], [0.5803, 0.3658, 0.8792, 0.4309, 0.3655, 0.5083, 0.6469, 0.5170], [0.5794, 0.3641, 0.7371, 0.3003, 0.3485, 0.3589, 0.5397, 0.5130], [0.6032, 0.3811, 0.8296, 0.4438, 0.3641, 0.4955, 0.5187, 0.5050]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5512, 0.5750], [0.6250, 0.4054, 0.8770, 0.4723, 0.4663, 0.5367, 0.6162, 0.5433], [0.6257, 0.4167, 0.8775, 0.3433, 0.3562, 0.4133, 0.6200, 0.5667], [0.6251, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6087, 0.5183], [0.6275, 0.4048, 0.8487, 0.2883, 0.4462, 0.2033, 0.6321, 0.5155], [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162], [0.6097, 0.4000, 0.7325, 0.2667, 0.3450, 0.3517, 0.5284, 0.5045], [0.6084, 0.4005, 0.8400, 0.4317, 0.3762, 0.4750, 0.5476, 0.5058]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.11381898814579472 step: 71 running loss: 0.0016030843400816157 Train Steps: 71/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378], [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6276, 0.3894, 0.8747, 0.4654, 0.3921, 0.3863, 0.6660, 0.5179], [0.6014, 0.3731, 0.8574, 0.4411, 0.4104, 0.5811, 0.5857, 0.5430], [0.6147, 0.3785, 0.7877, 0.2300, 0.4743, 0.1988, 0.6161, 0.4771], [0.6622, 0.4007, 0.8515, 0.5078, 0.3759, 0.5819, 0.6995, 0.5133], [0.5559, 0.3539, 0.8504, 0.4778, 0.4623, 0.5141, 0.5119, 0.5238], [0.6134, 0.3951, 0.7303, 0.3397, 0.5047, 0.2150, 0.5570, 0.5806], [0.5881, 0.3722, 0.7941, 0.3264, 0.3500, 0.3278, 0.4857, 0.5187], [0.5848, 0.3471, 0.8362, 0.4906, 0.3879, 0.4857, 0.5017, 0.4967]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6307, 0.4029, 0.8988, 0.4817, 0.3938, 0.3500, 0.7311, 0.5378], [0.6189, 0.4049, 0.8888, 0.4417, 0.4212, 0.5200, 0.5987, 0.5633], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6143, 0.3982, 0.8650, 0.4883, 0.3913, 0.4317, 0.5315, 0.5350]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.11468475608853623 step: 72 running loss: 0.0015928438345630032 Train Steps: 72/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540], [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500], [0.6228, 0.4119, 0.7938, 0.2233, 0.4674, 0.1773, 0.6188, 0.5433], [0.6274, 0.4003, 0.8638, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661], [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097], [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6321, 0.3942, 0.8170, 0.4639, 0.4226, 0.4967, 0.5175, 0.5260], [0.6072, 0.3800, 0.8428, 0.2650, 0.4421, 0.2778, 0.6690, 0.5455], [0.6211, 0.3606, 0.8914, 0.4243, 0.4077, 0.3264, 0.6256, 0.5129], [0.6232, 0.3805, 0.8581, 0.4738, 0.3662, 0.4002, 0.5599, 0.5180], [0.6466, 0.4008, 0.7879, 0.2387, 0.4705, 0.1779, 0.5898, 0.5244], [0.6400, 0.3735, 0.8262, 0.5660, 0.3768, 0.4563, 0.6083, 0.4759], [0.5742, 0.3790, 0.8567, 0.4308, 0.4156, 0.5919, 0.5513, 0.5092], [0.6213, 0.3897, 0.8506, 0.4837, 0.4374, 0.4786, 0.5193, 0.5460]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6098, 0.3991, 0.8637, 0.4717, 0.4263, 0.4967, 0.5213, 0.5650], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540], [0.6262, 0.4163, 0.8850, 0.5183, 0.3762, 0.4150, 0.6025, 0.5500], [0.6228, 0.4119, 0.7937, 0.2233, 0.4674, 0.1773, 0.6187, 0.5433], [0.6274, 0.4003, 0.8637, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661], [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097], [0.6140, 0.4070, 0.8700, 0.5000, 0.4613, 0.4900, 0.5260, 0.5852]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.11528142628958449 step: 73 running loss: 0.001579197620405267 Train Steps: 73/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317], [0.6224, 0.4061, 0.8988, 0.4300, 0.3838, 0.4750, 0.6112, 0.5483], [0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [ nan, nan, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927], [0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467], [0.6357, 0.4097, 0.9038, 0.3883, 0.4213, 0.2950, 0.6686, 0.5390]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7195, 0.4545, 0.8844, 0.5364, 0.3846, 0.4108, 0.5699, 0.5352], [0.6740, 0.4419, 0.9011, 0.4222, 0.3800, 0.5256, 0.5829, 0.5424], [0.6853, 0.4666, 0.7156, 0.2920, 0.4794, 0.2020, 0.5704, 0.5991], [0.2321, 0.1452, 0.7913, 0.2523, 0.4388, 0.2067, 0.5140, 0.5255], [0.6632, 0.4370, 0.8896, 0.4825, 0.3672, 0.4591, 0.6066, 0.5275], [0.6851, 0.4356, 0.8823, 0.5456, 0.3809, 0.4654, 0.6006, 0.5075], [0.6783, 0.4285, 0.8513, 0.5570, 0.3972, 0.4829, 0.5697, 0.5386], [0.7579, 0.4764, 0.8849, 0.3669, 0.4426, 0.3150, 0.6563, 0.5561]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317], [0.6224, 0.4061, 0.8988, 0.4300, 0.3837, 0.4750, 0.6112, 0.5483], [0.6270, 0.4266, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [0.0000, 0.0000, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6321, 0.4048, 0.8737, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927], [0.6273, 0.4143, 0.8750, 0.5700, 0.3988, 0.4717, 0.6012, 0.5467], [0.6357, 0.4097, 0.9038, 0.3883, 0.4212, 0.2950, 0.6686, 0.5390]]], device='cuda:0') loss_train_step before backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0022, device='cuda:0', grad_fn=) loss_train: 0.11752213741419837 step: 74 running loss: 0.0015881369920837617 Train Steps: 74/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6102, 0.4020, 0.8638, 0.3717, 0.3625, 0.5017, 0.6038, 0.5500], [0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050], [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917], [0.6152, 0.4131, 0.6863, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133], [0.6317, 0.4038, 0.8287, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986], [0.6273, 0.4100, 0.7137, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633], [0.6115, 0.4081, 0.6725, 0.2433, 0.4088, 0.1933, 0.5167, 0.5544]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6591, 0.4333, 0.8948, 0.4122, 0.3788, 0.4922, 0.6208, 0.5571], [0.6427, 0.4237, 0.9069, 0.5538, 0.4507, 0.5241, 0.5406, 0.5316], [0.6362, 0.4339, 0.8928, 0.5974, 0.4273, 0.4166, 0.5866, 0.6091], [0.6376, 0.4100, 0.7270, 0.2893, 0.3867, 0.3177, 0.5819, 0.5739], [0.6525, 0.4261, 0.8762, 0.2520, 0.5085, 0.1691, 0.6501, 0.5437], [0.6357, 0.4092, 0.8949, 0.6106, 0.3867, 0.4622, 0.6421, 0.5256], [0.6220, 0.4065, 0.7542, 0.2496, 0.4231, 0.2386, 0.6143, 0.5739], [0.6334, 0.4142, 0.7204, 0.2827, 0.4284, 0.2035, 0.5362, 0.5713]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6102, 0.4020, 0.8637, 0.3717, 0.3625, 0.5017, 0.6037, 0.5500], [0.6097, 0.3988, 0.8650, 0.5250, 0.4212, 0.5200, 0.5675, 0.5050], [0.6212, 0.4159, 0.8675, 0.5783, 0.4087, 0.4317, 0.5612, 0.5917], [0.6152, 0.4131, 0.6862, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [0.6266, 0.4065, 0.8313, 0.2467, 0.4787, 0.1733, 0.6313, 0.5133], [0.6317, 0.4038, 0.8288, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986], [0.6273, 0.4099, 0.7138, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633], [0.6115, 0.4081, 0.6725, 0.2433, 0.4087, 0.1933, 0.5167, 0.5544]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.11819850071333349 step: 75 running loss: 0.001575980009511113 Train Steps: 75/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131], [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400], [0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433], [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320], [0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6188, 0.5400], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6172, 0.3995, 0.8901, 0.5058, 0.3956, 0.4915, 0.6131, 0.5523], [0.6882, 0.4489, 0.8780, 0.4083, 0.3668, 0.3666, 0.5890, 0.5351], [0.5328, 0.3744, 0.8810, 0.5341, 0.4609, 0.4898, 0.5960, 0.5529], [0.7211, 0.4788, 0.8036, 0.3136, 0.3870, 0.3325, 0.6272, 0.5799], [0.6880, 0.4580, 0.8173, 0.2626, 0.4443, 0.2419, 0.6816, 0.5460], [0.6753, 0.4548, 0.8753, 0.3632, 0.3737, 0.2981, 0.6539, 0.5604], [0.6129, 0.3984, 0.9226, 0.4563, 0.3824, 0.3320, 0.6079, 0.5717], [0.6587, 0.4387, 0.9108, 0.5261, 0.4108, 0.4428, 0.5994, 0.5486]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6199, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131], [0.6201, 0.3970, 0.8413, 0.4950, 0.4412, 0.5183, 0.6087, 0.5400], [0.6175, 0.4091, 0.7862, 0.2800, 0.3638, 0.3583, 0.6187, 0.5433], [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320], [0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6187, 0.5400], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.11945557687431574 step: 76 running loss: 0.0015717839062409965 Train Steps: 76/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290], [0.6197, 0.4118, 0.8688, 0.5517, 0.4037, 0.5233, 0.5875, 0.5600], [0.6111, 0.4033, 0.8300, 0.3267, 0.3588, 0.3333, 0.5444, 0.5637], [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050], [0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050], [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116], [0.6265, 0.4088, 0.8025, 0.1850, 0.4163, 0.2500, 0.6290, 0.4947], [0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6638, 0.4265, 0.8378, 0.3199, 0.3502, 0.3498, 0.6437, 0.5390], [0.6458, 0.4382, 0.8623, 0.5751, 0.4099, 0.4719, 0.6228, 0.5623], [0.6353, 0.4273, 0.8509, 0.3646, 0.3605, 0.2750, 0.5698, 0.5699], [0.6264, 0.4253, 0.8758, 0.5270, 0.4639, 0.4688, 0.6128, 0.5361], [0.6012, 0.4146, 0.8752, 0.4587, 0.3832, 0.4310, 0.6454, 0.6015], [0.6165, 0.4263, 0.8658, 0.4376, 0.4099, 0.5349, 0.6132, 0.5368], [0.6423, 0.4167, 0.8071, 0.2262, 0.4394, 0.1830, 0.6501, 0.5213], [0.6808, 0.4402, 0.8347, 0.3076, 0.4046, 0.2507, 0.6324, 0.5565]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290], [0.6197, 0.4118, 0.8687, 0.5517, 0.4038, 0.5233, 0.5875, 0.5600], [0.6111, 0.4033, 0.8300, 0.3267, 0.3587, 0.3333, 0.5444, 0.5637], [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050], [0.6200, 0.4112, 0.8863, 0.4100, 0.3638, 0.4917, 0.6087, 0.6050], [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116], [0.6265, 0.4088, 0.8025, 0.1850, 0.4162, 0.2500, 0.6290, 0.4947], [0.6200, 0.4098, 0.8238, 0.2917, 0.4013, 0.2967, 0.6000, 0.5683]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.12034609029069543 step: 77 running loss: 0.001562936237541499 Train Steps: 77/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960], [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142], [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672], [ nan, nan, 0.8300, 0.3150, 0.3588, 0.3383, 0.5208, 0.5194], [0.6161, 0.4055, 0.8675, 0.3867, 0.3713, 0.4033, 0.5195, 0.5162], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882], [0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6829, 0.4321, 0.8792, 0.2930, 0.4530, 0.1758, 0.6499, 0.5164], [0.6334, 0.4151, 0.8681, 0.4462, 0.3797, 0.4307, 0.5416, 0.5134], [0.7628, 0.4811, 0.8505, 0.5250, 0.3577, 0.4349, 0.6527, 0.5248], [0.2527, 0.1632, 0.8017, 0.3036, 0.3434, 0.2813, 0.5369, 0.5235], [0.5974, 0.3909, 0.8661, 0.3871, 0.3583, 0.3448, 0.5662, 0.5360], [0.6522, 0.4458, 0.8638, 0.3950, 0.3395, 0.3446, 0.5653, 0.5265], [0.6673, 0.4463, 0.8625, 0.4778, 0.4617, 0.4634, 0.6032, 0.5708], [0.6972, 0.4597, 0.8791, 0.4854, 0.4432, 0.5452, 0.6703, 0.5676]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6259, 0.4156, 0.8813, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960], [0.6185, 0.4067, 0.8838, 0.4450, 0.4038, 0.4733, 0.5213, 0.5142], [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672], [0.0000, 0.0000, 0.8300, 0.3150, 0.3587, 0.3383, 0.5208, 0.5194], [0.6161, 0.4055, 0.8675, 0.3867, 0.3713, 0.4033, 0.5195, 0.5162], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882], [0.6112, 0.4029, 0.8637, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517]]], device='cuda:0') loss_train_step before backward: tensor(0.0027, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0027, device='cuda:0', grad_fn=) loss_train: 0.12301964685320854 step: 78 running loss: 0.0015771749596565198 Train Steps: 78/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947], [0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064], [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204], [0.6124, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402], [0.6132, 0.4037, 0.6963, 0.2217, 0.4100, 0.1950, 0.5395, 0.5175], [0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320], [ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6981, 0.4461, 0.7920, 0.1761, 0.4232, 0.2350, 0.6595, 0.5030], [0.6089, 0.4098, 0.8837, 0.4997, 0.4948, 0.5058, 0.5205, 0.5163], [0.6776, 0.4383, 0.8625, 0.5327, 0.4152, 0.5352, 0.6680, 0.5216], [0.6112, 0.3980, 0.8476, 0.4931, 0.3843, 0.4628, 0.5391, 0.5513], [0.5809, 0.3944, 0.7159, 0.2180, 0.3969, 0.1966, 0.5366, 0.4958], [0.7036, 0.4519, 0.9253, 0.4453, 0.3807, 0.5351, 0.7349, 0.5451], [0.0905, 0.0750, 0.6754, 0.2421, 0.3957, 0.2056, 0.5225, 0.5518], [0.6353, 0.4183, 0.8821, 0.5299, 0.3436, 0.3670, 0.5775, 0.5352]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947], [0.6128, 0.4022, 0.8737, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064], [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204], [0.6123, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402], [0.6132, 0.4037, 0.6963, 0.2217, 0.4100, 0.1950, 0.5395, 0.5175], [0.6250, 0.3993, 0.9137, 0.4333, 0.3762, 0.5217, 0.6995, 0.5320], [0.0000, 0.0000, 0.6688, 0.2513, 0.4112, 0.2117, 0.5193, 0.5933], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.12372523982776329 step: 79 running loss: 0.0015661422763008012 Train Steps: 79/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6250, 0.4106, 0.8700, 0.3717, 0.3588, 0.4967, 0.6038, 0.5167], [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5963, 0.6217], [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6288, 0.4135, 0.8620, 0.3507, 0.3612, 0.5392, 0.5979, 0.4968], [0.5228, 0.3384, 0.7171, 0.2279, 0.4291, 0.2401, 0.5491, 0.5494], [0.5521, 0.3427, 0.9261, 0.4161, 0.4091, 0.3444, 0.6386, 0.4978], [0.5937, 0.3695, 0.8879, 0.5150, 0.3578, 0.4873, 0.5942, 0.4688], [0.6119, 0.3992, 0.8160, 0.5159, 0.3865, 0.5166, 0.6645, 0.5292], [0.5653, 0.3847, 0.8176, 0.5684, 0.3951, 0.4822, 0.5524, 0.5696], [0.5591, 0.3564, 0.7779, 0.2521, 0.3498, 0.4319, 0.5708, 0.5020], [0.5886, 0.3754, 0.8601, 0.2555, 0.4770, 0.2343, 0.6406, 0.5053]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6250, 0.4105, 0.8700, 0.3717, 0.3587, 0.4967, 0.6037, 0.5167], [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5962, 0.6217], [0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.6346, 0.4086, 0.7937, 0.5500, 0.3963, 0.4867, 0.7343, 0.5702], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6197, 0.4051, 0.7812, 0.2650, 0.3512, 0.4050, 0.6112, 0.5500], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.12507811159593984 step: 80 running loss: 0.001563476394949248 Train Steps: 80/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6277, 0.4103, 0.8087, 0.5717, 0.4188, 0.4750, 0.5663, 0.6083], [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5737, 0.5350], [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800], [0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283], [0.6198, 0.4101, 0.8838, 0.5283, 0.3763, 0.5267, 0.5913, 0.5567], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6061, 0.3938, 0.8173, 0.5448, 0.4088, 0.4903, 0.6078, 0.5864], [0.6013, 0.3893, 0.8713, 0.4460, 0.3765, 0.3447, 0.5802, 0.5218], [0.5587, 0.3764, 0.7229, 0.2238, 0.4488, 0.2153, 0.5614, 0.5439], [0.5299, 0.3311, 0.8490, 0.4794, 0.4807, 0.5188, 0.5202, 0.4970], [0.5694, 0.3743, 0.8593, 0.5075, 0.3752, 0.5695, 0.5847, 0.5236], [0.5815, 0.3632, 0.8861, 0.4526, 0.3648, 0.5317, 0.6349, 0.4731], [0.5626, 0.3566, 0.8745, 0.3104, 0.3800, 0.3518, 0.5987, 0.5117], [0.5221, 0.3276, 0.8707, 0.3038, 0.3811, 0.3485, 0.5562, 0.4916]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6277, 0.4103, 0.8087, 0.5717, 0.4187, 0.4750, 0.5663, 0.6083], [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5738, 0.5350], [0.6218, 0.4185, 0.7337, 0.2650, 0.4625, 0.1950, 0.5688, 0.5800], [0.6084, 0.4008, 0.8587, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283], [0.6198, 0.4101, 0.8838, 0.5283, 0.3762, 0.5267, 0.5913, 0.5567], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.12628521776059642 step: 81 running loss: 0.001559076762476499 Train Steps: 81/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319], [0.6083, 0.3957, 0.8638, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980], [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467], [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450], [0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136], [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5418, 0.3460, 0.7083, 0.3052, 0.3531, 0.3603, 0.5018, 0.5154], [0.5441, 0.3577, 0.8451, 0.4836, 0.4316, 0.5414, 0.5323, 0.4998], [0.5531, 0.3617, 0.8243, 0.3293, 0.3425, 0.4509, 0.5539, 0.5333], [0.5925, 0.3804, 0.8683, 0.4660, 0.3918, 0.4476, 0.5878, 0.5417], [0.6002, 0.4011, 0.8016, 0.3649, 0.3546, 0.5386, 0.5675, 0.5146], [0.5849, 0.4086, 0.8510, 0.4163, 0.3650, 0.3590, 0.5596, 0.5523], [0.5821, 0.3788, 0.8625, 0.5083, 0.4160, 0.5501, 0.5990, 0.5181], [0.4306, 0.2809, 0.8359, 0.2389, 0.5374, 0.2232, 0.7439, 0.5276]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6109, 0.4041, 0.6975, 0.3167, 0.3512, 0.3383, 0.5153, 0.5319], [0.6083, 0.3957, 0.8637, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980], [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467], [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450], [0.6132, 0.4118, 0.8200, 0.3633, 0.3562, 0.5400, 0.5787, 0.5136], [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.12774768978124484 step: 82 running loss: 0.0015578986558688395 Train Steps: 82/90 Loss: 0.0016 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235], [0.6197, 0.3986, 0.8800, 0.4617, 0.4188, 0.4783, 0.5687, 0.5550], [0.6222, 0.3957, 0.8838, 0.5017, 0.3937, 0.4600, 0.5900, 0.5017], [ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633], [0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983], [0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6675, 0.4397, 0.8381, 0.5552, 0.3738, 0.5097, 0.6472, 0.5416], [ 0.5951, 0.3972, 0.8532, 0.4696, 0.4206, 0.5650, 0.6118, 0.5556], [ 0.6322, 0.4160, 0.8588, 0.4564, 0.4137, 0.5005, 0.5463, 0.5500], [ 0.6032, 0.3870, 0.8658, 0.4831, 0.3925, 0.4840, 0.5734, 0.5137], [-0.0529, -0.0354, 0.7400, 0.2637, 0.3871, 0.2772, 0.4593, 0.5618], [ 0.6284, 0.4051, 0.8570, 0.5077, 0.4325, 0.5167, 0.5824, 0.5051], [ 0.6477, 0.4342, 0.8797, 0.3638, 0.3833, 0.2962, 0.6136, 0.5325], [ 0.6533, 0.4331, 0.8344, 0.5550, 0.4074, 0.4584, 0.5385, 0.5828]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235], [0.6197, 0.3986, 0.8800, 0.4617, 0.4187, 0.4783, 0.5688, 0.5550], [0.6222, 0.3957, 0.8838, 0.5017, 0.3938, 0.4600, 0.5900, 0.5017], [0.0000, 0.0000, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633], [0.6219, 0.3934, 0.8687, 0.5267, 0.4313, 0.4967, 0.5987, 0.4983], [0.6277, 0.4117, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.1281739636324346 step: 83 running loss: 0.0015442646220775254 Train Steps: 83/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862], [0.6164, 0.4119, 0.7913, 0.2650, 0.3538, 0.3500, 0.5614, 0.5038], [0.6127, 0.4118, 0.8650, 0.5083, 0.4088, 0.5367, 0.5300, 0.5456], [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5837, 0.5167], [0.6277, 0.4083, 0.8350, 0.2717, 0.4562, 0.1800, 0.5918, 0.4878], [0.6199, 0.4065, 0.7598, 0.2385, 0.4317, 0.1981, 0.5933, 0.5221], [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6597, 0.4264, 0.8902, 0.4278, 0.3658, 0.4644, 0.6411, 0.5222], [0.5860, 0.3856, 0.7773, 0.2711, 0.3504, 0.3750, 0.5729, 0.5375], [0.5656, 0.3925, 0.8694, 0.5230, 0.4260, 0.5606, 0.5162, 0.5660], [0.6103, 0.3844, 0.8472, 0.4252, 0.3782, 0.3238, 0.5537, 0.5528], [0.5067, 0.3293, 0.8092, 0.2659, 0.4708, 0.2124, 0.5685, 0.5118], [0.5839, 0.3781, 0.7251, 0.2463, 0.4241, 0.2149, 0.5706, 0.5286], [0.5335, 0.3611, 0.8560, 0.5066, 0.3550, 0.4511, 0.5697, 0.6224], [0.5556, 0.3549, 0.8647, 0.5021, 0.4329, 0.5528, 0.6810, 0.5749]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862], [0.6164, 0.4119, 0.7912, 0.2650, 0.3537, 0.3500, 0.5614, 0.5038], [0.6127, 0.4118, 0.8650, 0.5083, 0.4087, 0.5367, 0.5300, 0.5456], [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5838, 0.5167], [0.6277, 0.4083, 0.8350, 0.2717, 0.4563, 0.1800, 0.5918, 0.4878], [0.6199, 0.4065, 0.7598, 0.2385, 0.4317, 0.1981, 0.5933, 0.5221], [0.6186, 0.4060, 0.8750, 0.5050, 0.3537, 0.4367, 0.5813, 0.6083], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.12917014537379146 step: 84 running loss: 0.0015377398258784698 Train Steps: 84/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433], [0.6059, 0.4002, 0.7562, 0.2767, 0.3538, 0.3033, 0.5529, 0.5455], [0.6145, 0.4007, 0.8775, 0.4533, 0.4562, 0.5533, 0.6088, 0.5533], [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116], [0.6250, 0.4013, 0.8525, 0.5417, 0.4037, 0.5117, 0.6325, 0.5017], [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5787, 0.5550], [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133], [0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6153, 0.4149, 0.8889, 0.4847, 0.4558, 0.5046, 0.6244, 0.5498], [0.5586, 0.3738, 0.7486, 0.2882, 0.3541, 0.2953, 0.5606, 0.5527], [0.6257, 0.4160, 0.8654, 0.4678, 0.4573, 0.5368, 0.5980, 0.5739], [0.5703, 0.3918, 0.7460, 0.1940, 0.3798, 0.2339, 0.5797, 0.5188], [0.5688, 0.3682, 0.8581, 0.5712, 0.4034, 0.4855, 0.6434, 0.5377], [0.6166, 0.4107, 0.8232, 0.3031, 0.3427, 0.3449, 0.5720, 0.5591], [0.6142, 0.3968, 0.8740, 0.5402, 0.4438, 0.4831, 0.6002, 0.5364], [0.5238, 0.3544, 0.8481, 0.3092, 0.4216, 0.2158, 0.5670, 0.5483]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6250, 0.4054, 0.8770, 0.4723, 0.4663, 0.5367, 0.6162, 0.5433], [0.6059, 0.4002, 0.7563, 0.2767, 0.3537, 0.3033, 0.5529, 0.5455], [0.6145, 0.4007, 0.8775, 0.4533, 0.4563, 0.5533, 0.6087, 0.5533], [0.6165, 0.4106, 0.7575, 0.1733, 0.3837, 0.2650, 0.5680, 0.5116], [0.6250, 0.4013, 0.8525, 0.5417, 0.4038, 0.5117, 0.6325, 0.5017], [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5788, 0.5550], [0.6202, 0.4053, 0.8637, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133], [0.6199, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.12972077063750476 step: 85 running loss: 0.001526126713382409 Train Steps: 85/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150], [ nan, nan, 0.8463, 0.2550, 0.5850, 0.2133, 0.7129, 0.6072], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799], [0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693], [0.6161, 0.4055, 0.8675, 0.3867, 0.3713, 0.4033, 0.5195, 0.5162], [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717], [0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6324, 0.4024, 0.7867, 0.2706, 0.4302, 0.1842, 0.5873, 0.5167], [0.1120, 0.0905, 0.8614, 0.2757, 0.5323, 0.2532, 0.7227, 0.5899], [0.6636, 0.4510, 0.8866, 0.5082, 0.3449, 0.4067, 0.6281, 0.5082], [0.6678, 0.4606, 0.8678, 0.3985, 0.3912, 0.2996, 0.5728, 0.5700], [0.6000, 0.4066, 0.8471, 0.5015, 0.3561, 0.4322, 0.5428, 0.5055], [0.6050, 0.4145, 0.8732, 0.3977, 0.3691, 0.3673, 0.5318, 0.5119], [0.5998, 0.4105, 0.6665, 0.2674, 0.3806, 0.2248, 0.5654, 0.5679], [0.6570, 0.4326, 0.8951, 0.4556, 0.3951, 0.5170, 0.6160, 0.5746]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150], [0.0000, 0.0000, 0.8462, 0.2550, 0.5850, 0.2133, 0.7129, 0.6072], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6250, 0.4236, 0.8637, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799], [0.6038, 0.3946, 0.8413, 0.4883, 0.3562, 0.4550, 0.5266, 0.4693], [0.6161, 0.4055, 0.8675, 0.3867, 0.3713, 0.4033, 0.5195, 0.5162], [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717], [0.6199, 0.4102, 0.8950, 0.4417, 0.4013, 0.5367, 0.6112, 0.5967]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.13042904192116112 step: 86 running loss: 0.0015166167665251293 Train Steps: 86/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6199, 0.4093, 0.7913, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700], [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528], [0.6216, 0.4099, 0.7225, 0.2033, 0.4188, 0.2217, 0.5975, 0.5283], [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378], [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431], [0.6200, 0.3993, 0.8519, 0.4923, 0.3962, 0.4717, 0.6013, 0.5433], [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690], [0.6060, 0.3924, 0.8450, 0.5717, 0.4200, 0.5217, 0.5253, 0.4752]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5883, 0.3915, 0.8001, 0.2407, 0.4200, 0.2433, 0.5929, 0.5700], [0.6419, 0.4123, 0.9253, 0.3650, 0.4035, 0.3305, 0.7279, 0.5396], [0.6671, 0.4461, 0.7152, 0.1981, 0.4120, 0.1980, 0.5596, 0.5310], [0.6116, 0.4015, 0.9169, 0.4768, 0.3858, 0.3125, 0.7088, 0.5173], [0.6014, 0.3990, 0.8853, 0.4906, 0.4388, 0.4627, 0.5007, 0.5111], [0.5797, 0.3709, 0.8777, 0.5019, 0.3913, 0.4391, 0.5890, 0.5408], [0.6075, 0.3936, 0.8572, 0.5368, 0.3973, 0.5263, 0.7019, 0.5504], [0.6125, 0.3837, 0.8629, 0.5984, 0.4356, 0.4845, 0.5144, 0.4990]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6198, 0.4093, 0.7912, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700], [0.6332, 0.4165, 0.9100, 0.3350, 0.4187, 0.3683, 0.7438, 0.5528], [0.6216, 0.4099, 0.7225, 0.2033, 0.4187, 0.2217, 0.5975, 0.5283], [0.6307, 0.4029, 0.8988, 0.4817, 0.3938, 0.3500, 0.7311, 0.5378], [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431], [0.6200, 0.3993, 0.8519, 0.4923, 0.3963, 0.4717, 0.6012, 0.5433], [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690], [0.6060, 0.3924, 0.8450, 0.5717, 0.4200, 0.5217, 0.5253, 0.4752]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.13084795890608802 step: 87 running loss: 0.0015039995276561842 Train Steps: 87/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6218, 0.4098, 0.7238, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350], [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050], [0.6284, 0.4093, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.6064, 0.3953, 0.8738, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416], [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133], [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496], [0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6102, 0.4038, 0.7431, 0.1998, 0.4382, 0.2402, 0.6259, 0.5483], [0.6186, 0.3994, 0.7977, 0.3072, 0.3726, 0.3181, 0.5868, 0.5125], [0.5738, 0.3796, 0.9057, 0.4642, 0.3647, 0.3801, 0.6272, 0.5086], [0.6229, 0.4183, 0.8687, 0.2433, 0.5416, 0.2083, 0.7472, 0.5412], [0.5756, 0.3719, 0.8908, 0.4649, 0.3554, 0.4701, 0.5623, 0.5266], [0.6130, 0.3902, 0.8879, 0.5309, 0.4536, 0.5069, 0.5984, 0.5160], [0.6295, 0.4091, 0.8896, 0.2798, 0.4066, 0.2617, 0.6012, 0.5375], [0.6980, 0.4612, 0.8706, 0.5554, 0.3980, 0.3224, 0.5773, 0.5992]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6218, 0.4098, 0.7237, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350], [0.6135, 0.3994, 0.7912, 0.3050, 0.3625, 0.3050, 0.5838, 0.5050], [0.6284, 0.4092, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.6064, 0.3952, 0.8737, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416], [0.6202, 0.4053, 0.8637, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133], [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496], [0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.1311794207431376 step: 88 running loss: 0.0014906752357174728 Train Steps: 88/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5887, 0.5017], [0.6228, 0.4119, 0.7938, 0.2233, 0.4674, 0.1773, 0.6188, 0.5433], [0.6207, 0.4110, 0.8738, 0.5000, 0.4800, 0.5633, 0.6300, 0.5433], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6030, 0.3969, 0.7988, 0.3917, 0.3450, 0.3667, 0.5266, 0.4700]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6583, 0.4411, 0.8900, 0.4753, 0.3661, 0.4107, 0.6070, 0.5610], [0.7032, 0.4607, 0.8342, 0.5524, 0.4124, 0.4293, 0.6039, 0.6208], [0.6121, 0.3993, 0.7666, 0.2818, 0.3541, 0.3070, 0.5600, 0.5471], [0.6439, 0.4244, 0.8729, 0.4387, 0.3880, 0.4663, 0.5833, 0.5265], [0.7073, 0.4632, 0.8083, 0.1969, 0.4801, 0.1462, 0.6399, 0.5277], [0.6096, 0.3859, 0.9005, 0.4553, 0.4819, 0.5500, 0.6491, 0.5414], [0.6181, 0.4030, 0.9020, 0.4781, 0.4361, 0.4963, 0.6299, 0.5102], [0.6141, 0.4031, 0.8295, 0.3727, 0.3551, 0.3516, 0.5576, 0.4941]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5888, 0.5017], [0.6228, 0.4119, 0.7937, 0.2233, 0.4674, 0.1773, 0.6187, 0.5433], [0.6207, 0.4110, 0.8737, 0.5000, 0.4800, 0.5633, 0.6300, 0.5433], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6030, 0.3969, 0.7987, 0.3917, 0.3450, 0.3667, 0.5266, 0.4700]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.13189973769476637 step: 89 running loss: 0.0014820195246602963 Train Steps: 89/90 Loss: 0.0015 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667], [0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773], [0.6357, 0.4118, 0.8400, 0.2500, 0.5413, 0.1633, 0.6725, 0.5586], [0.6115, 0.3998, 0.7063, 0.2383, 0.4037, 0.1950, 0.5320, 0.4993], [0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249], [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6113, 0.4123, 0.9156, 0.4842, 0.3787, 0.4339, 0.6803, 0.5313], [0.5968, 0.3982, 0.8830, 0.4667, 0.3824, 0.5372, 0.5866, 0.4975], [0.6615, 0.4335, 0.8602, 0.3215, 0.3607, 0.3666, 0.5859, 0.5735], [0.6616, 0.4508, 0.6859, 0.2386, 0.4228, 0.2136, 0.5463, 0.5839], [0.6522, 0.4216, 0.8659, 0.2377, 0.5468, 0.1720, 0.6670, 0.5417], [0.6198, 0.4215, 0.7188, 0.2078, 0.4245, 0.2021, 0.5389, 0.5014], [0.5990, 0.3871, 0.9110, 0.5404, 0.3889, 0.4972, 0.6665, 0.5114], [0.5953, 0.4015, 0.6792, 0.2568, 0.4020, 0.2465, 0.5699, 0.5747]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667], [0.6147, 0.4026, 0.6600, 0.2467, 0.4087, 0.2150, 0.5489, 0.5773], [0.6357, 0.4118, 0.8400, 0.2500, 0.5412, 0.1633, 0.6725, 0.5586], [0.6115, 0.3998, 0.7063, 0.2383, 0.4038, 0.1950, 0.5320, 0.4993], [0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249], [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.13218482348020189 step: 90 running loss: 0.001468720260891132 Valid Steps: 10/10 Loss: nan 15 -------------------------------------------------- Epoch: 7 Train Loss: 0.0015 Valid Loss: nan -------------------------------------------------- size of train loader is: 90 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167], [0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6250, 0.4146, 0.8838, 0.3933, 0.3588, 0.4283, 0.6162, 0.5367], [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6952, 0.4387, 0.8725, 0.4735, 0.3666, 0.3972, 0.6239, 0.5173], [0.6620, 0.4335, 0.8837, 0.4076, 0.3828, 0.3460, 0.6416, 0.5286], [0.6551, 0.4304, 0.8760, 0.4881, 0.4687, 0.5299, 0.6150, 0.5283], [0.6975, 0.4636, 0.8789, 0.4989, 0.3623, 0.4215, 0.6250, 0.5132], [0.6783, 0.4559, 0.8491, 0.3904, 0.3644, 0.3979, 0.6092, 0.5457], [0.6325, 0.4093, 0.7707, 0.2715, 0.3652, 0.4031, 0.6094, 0.5560], [0.6950, 0.4435, 0.8780, 0.4944, 0.4598, 0.6014, 0.6045, 0.5163], [0.6692, 0.4511, 0.8783, 0.4058, 0.3701, 0.3708, 0.5801, 0.5369]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6263, 0.5167], [0.6250, 0.4054, 0.8770, 0.4723, 0.4663, 0.5367, 0.6162, 0.5433], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6250, 0.4146, 0.8838, 0.3933, 0.3587, 0.4283, 0.6162, 0.5367], [0.6197, 0.4051, 0.7812, 0.2650, 0.3512, 0.4050, 0.6112, 0.5500], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.000634737079963088 step: 1 running loss: 0.000634737079963088 Train Steps: 1/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393], [0.6115, 0.3998, 0.7063, 0.2383, 0.4037, 0.1950, 0.5320, 0.4993], [0.6261, 0.4066, 0.8325, 0.2150, 0.4763, 0.2667, 0.7002, 0.5633], [0.6153, 0.4119, 0.8463, 0.3833, 0.3600, 0.3200, 0.5106, 0.5563], [0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050], [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097], [0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071], [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6766, 0.4289, 0.8286, 0.2968, 0.4127, 0.2863, 0.6748, 0.5418], [0.6179, 0.4140, 0.6912, 0.2371, 0.4145, 0.2128, 0.5311, 0.5148], [0.6247, 0.4153, 0.8053, 0.2420, 0.4797, 0.2715, 0.7081, 0.5556], [0.5787, 0.3854, 0.8454, 0.4129, 0.3846, 0.3143, 0.5029, 0.5668], [0.6452, 0.4174, 0.8559, 0.4356, 0.3796, 0.5023, 0.6076, 0.5970], [0.6454, 0.4280, 0.8659, 0.4704, 0.4320, 0.6108, 0.5659, 0.5241], [0.5738, 0.3689, 0.7364, 0.2182, 0.3825, 0.2897, 0.5679, 0.5108], [0.6694, 0.4415, 0.8888, 0.4933, 0.3867, 0.3707, 0.6565, 0.5217]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6286, 0.4086, 0.8408, 0.2801, 0.4162, 0.2800, 0.6725, 0.5393], [0.6115, 0.3998, 0.7063, 0.2383, 0.4038, 0.1950, 0.5320, 0.4993], [0.6261, 0.4066, 0.8325, 0.2150, 0.4762, 0.2667, 0.7002, 0.5633], [0.6153, 0.4119, 0.8462, 0.3833, 0.3600, 0.3200, 0.5106, 0.5563], [0.6200, 0.4112, 0.8863, 0.4100, 0.3638, 0.4917, 0.6087, 0.6050], [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097], [0.6163, 0.4114, 0.7650, 0.2017, 0.3762, 0.2867, 0.5631, 0.5071], [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.0009739936213009059 step: 2 running loss: 0.00048699681065045297 Train Steps: 2/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6307, 0.4029, 0.8650, 0.5200, 0.3763, 0.4017, 0.7311, 0.5366], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467], [0.6131, 0.4037, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680], [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6063, 0.6083], [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309], [0.6118, 0.4052, 0.8463, 0.3917, 0.3538, 0.3450, 0.5053, 0.5593], [0.6224, 0.4061, 0.8988, 0.4300, 0.3838, 0.4750, 0.6112, 0.5483]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6684, 0.4218, 0.8629, 0.5071, 0.3801, 0.4074, 0.7064, 0.5017], [0.6440, 0.3897, 0.8907, 0.4888, 0.3580, 0.5210, 0.6386, 0.4876], [0.6618, 0.4176, 0.8623, 0.4999, 0.4258, 0.5417, 0.5987, 0.5448], [0.6504, 0.4116, 0.6684, 0.2818, 0.3705, 0.3098, 0.5226, 0.5757], [0.6074, 0.4032, 0.8490, 0.3834, 0.3605, 0.4385, 0.6010, 0.6027], [0.7068, 0.4541, 0.8974, 0.4816, 0.3890, 0.3849, 0.6512, 0.5143], [0.6663, 0.4288, 0.8405, 0.3848, 0.3601, 0.3499, 0.4727, 0.5486], [0.6665, 0.4204, 0.8850, 0.4391, 0.3886, 0.5139, 0.6090, 0.5330]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6307, 0.4029, 0.8650, 0.5200, 0.3762, 0.4017, 0.7311, 0.5366], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6199, 0.4059, 0.8700, 0.4900, 0.4162, 0.5000, 0.6162, 0.5467], [0.6131, 0.4036, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680], [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6062, 0.6083], [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309], [0.6118, 0.4052, 0.8462, 0.3917, 0.3537, 0.3450, 0.5053, 0.5593], [0.6224, 0.4061, 0.8988, 0.4300, 0.3837, 0.4750, 0.6112, 0.5483]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.0014395180041901767 step: 3 running loss: 0.0004798393347300589 Train Steps: 3/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892], [0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6202, 0.4079, 0.8025, 0.2500, 0.3763, 0.3217, 0.6125, 0.5533], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683], [0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391], [0.6100, 0.4071, 0.7601, 0.3444, 0.3400, 0.4117, 0.5625, 0.5617], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6081, 0.3873, 0.8497, 0.4549, 0.3521, 0.3878, 0.5540, 0.5651], [0.6429, 0.4091, 0.7206, 0.2405, 0.4152, 0.2421, 0.5812, 0.5704], [0.6230, 0.3810, 0.7868, 0.2496, 0.3900, 0.3198, 0.6284, 0.5419], [0.6585, 0.4105, 0.8837, 0.5345, 0.3681, 0.3895, 0.5702, 0.5555], [0.6274, 0.3987, 0.8372, 0.3034, 0.4211, 0.2373, 0.6061, 0.5390], [0.6069, 0.3815, 0.7686, 0.3443, 0.3403, 0.4284, 0.5813, 0.5239], [0.6252, 0.3910, 0.8705, 0.4495, 0.4389, 0.6050, 0.6370, 0.5267], [0.6166, 0.3865, 0.8532, 0.4427, 0.3593, 0.3599, 0.5404, 0.5638]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892], [0.6193, 0.4050, 0.7312, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6202, 0.4079, 0.8025, 0.2500, 0.3762, 0.3217, 0.6125, 0.5533], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5412, 0.5683], [0.6199, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391], [0.6100, 0.4071, 0.7601, 0.3444, 0.3400, 0.4117, 0.5625, 0.5617], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6182, 0.4058, 0.8737, 0.4350, 0.3562, 0.3400, 0.5290, 0.5822]]], device='cuda:0') loss_train_step before backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train: 0.001678855056525208 step: 4 running loss: 0.000419713764131302 Train Steps: 4/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167], [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6535, 0.4268, 0.9218, 0.4248, 0.3633, 0.3752, 0.6456, 0.5334], [0.6299, 0.4239, 0.8814, 0.4930, 0.4228, 0.5131, 0.5455, 0.5724], [0.6410, 0.4221, 0.8750, 0.2759, 0.4460, 0.2838, 0.7228, 0.5797], [0.6569, 0.4039, 0.9135, 0.5521, 0.3506, 0.4886, 0.6536, 0.4932], [0.0340, 0.0284, 0.6861, 0.2670, 0.4044, 0.2420, 0.5503, 0.6324], [0.7165, 0.4682, 0.7702, 0.1931, 0.4608, 0.1776, 0.6070, 0.5090], [0.6661, 0.4247, 0.8615, 0.4843, 0.3716, 0.5002, 0.5485, 0.5507], [0.6522, 0.4341, 0.6852, 0.3236, 0.3517, 0.3098, 0.5355, 0.5977]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6263, 0.5167], [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.0000, 0.0000, 0.6688, 0.2513, 0.4112, 0.2117, 0.5193, 0.5933], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.002325661844224669 step: 5 running loss: 0.0004651323688449338 Train Steps: 5/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5737, 0.5700], [0.6261, 0.3987, 0.8688, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575], [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483], [0.6177, 0.4086, 0.8738, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6606, 0.4245, 0.8836, 0.5583, 0.3539, 0.4847, 0.5659, 0.5619], [0.6123, 0.3990, 0.8744, 0.5123, 0.4089, 0.5486, 0.6709, 0.5649], [0.5972, 0.3926, 0.8888, 0.3162, 0.3896, 0.4101, 0.6946, 0.5617], [0.5883, 0.3712, 0.8911, 0.5428, 0.3359, 0.4267, 0.6049, 0.5045], [0.6754, 0.4337, 0.7266, 0.2542, 0.3625, 0.2963, 0.5969, 0.5932], [0.5790, 0.3739, 0.8706, 0.4205, 0.3542, 0.4721, 0.5582, 0.5324], [0.5863, 0.3972, 0.8765, 0.4084, 0.3567, 0.5534, 0.6111, 0.5487], [0.6434, 0.4218, 0.7339, 0.3618, 0.4775, 0.1961, 0.5330, 0.6443]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5738, 0.5700], [0.6261, 0.3987, 0.8687, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575], [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483], [0.6177, 0.4085, 0.8737, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.0027636906161205843 step: 6 running loss: 0.00046061510268676403 Train Steps: 6/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6208, 0.4082, 0.8538, 0.3067, 0.3588, 0.3717, 0.6112, 0.5517], [0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650], [ nan, nan, 0.7725, 0.2611, 0.3675, 0.2733, 0.5413, 0.5167], [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501], [0.6336, 0.4154, 0.8900, 0.2767, 0.4988, 0.2867, 0.7422, 0.5540], [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6093, 0.4016, 0.8455, 0.3136, 0.3467, 0.3746, 0.5997, 0.5630], [0.6374, 0.4256, 0.8931, 0.3969, 0.3701, 0.2547, 0.6111, 0.5143], [0.6286, 0.4087, 0.8342, 0.4794, 0.3684, 0.4703, 0.5179, 0.5471], [0.6256, 0.4069, 0.8910, 0.4117, 0.3544, 0.4030, 0.5732, 0.5737], [0.0518, 0.0447, 0.7377, 0.2758, 0.3662, 0.2630, 0.5332, 0.5880], [0.6257, 0.4142, 0.7941, 0.2575, 0.4349, 0.1931, 0.5780, 0.5478], [0.6132, 0.4034, 0.8644, 0.3036, 0.4870, 0.2848, 0.7201, 0.5546], [0.6095, 0.4068, 0.8269, 0.3836, 0.3205, 0.4037, 0.5424, 0.5314]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6208, 0.4082, 0.8537, 0.3067, 0.3587, 0.3717, 0.6112, 0.5517], [0.6277, 0.4117, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6203, 0.4096, 0.8863, 0.4267, 0.3537, 0.4117, 0.6025, 0.5650], [0.0000, 0.0000, 0.7725, 0.2611, 0.3675, 0.2733, 0.5412, 0.5167], [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501], [0.6336, 0.4154, 0.8900, 0.2767, 0.4988, 0.2867, 0.7422, 0.5540], [0.6058, 0.3978, 0.8288, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.003107938144239597 step: 7 running loss: 0.0004439911634627996 Train Steps: 7/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6095, 0.3970, 0.8688, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822], [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552], [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795], [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167], [0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6563, 0.4177, 0.8839, 0.4649, 0.4715, 0.4626, 0.5183, 0.5332], [ 0.6144, 0.4091, 0.6869, 0.3082, 0.3530, 0.2762, 0.5134, 0.5890], [ 0.6505, 0.4218, 0.7742, 0.1857, 0.4556, 0.1476, 0.5805, 0.5013], [ 0.6151, 0.4132, 0.8729, 0.4727, 0.4237, 0.4703, 0.5191, 0.5730], [-0.0467, -0.0208, 0.8933, 0.2386, 0.5126, 0.2212, 0.7333, 0.5870], [ 0.6474, 0.4061, 0.8690, 0.5207, 0.3831, 0.5279, 0.6993, 0.5735], [ 0.6017, 0.3957, 0.8910, 0.5184, 0.4018, 0.5002, 0.5945, 0.5251], [ 0.6256, 0.4011, 0.7802, 0.1927, 0.3695, 0.2676, 0.5585, 0.5046]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6095, 0.3970, 0.8687, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822], [0.0000, 0.0000, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552], [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795], [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167], [0.6163, 0.4114, 0.7650, 0.2017, 0.3762, 0.2867, 0.5631, 0.5071]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.003420340406592004 step: 8 running loss: 0.0004275425508240005 Train Steps: 8/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823], [0.6081, 0.3950, 0.8538, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954], [0.6250, 0.4013, 0.8525, 0.5417, 0.4037, 0.5117, 0.6325, 0.5017], [0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550], [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633], [ nan, nan, 0.8525, 0.2217, 0.5413, 0.2367, 0.7367, 0.5482], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5593, 0.3640, 0.7191, 0.2089, 0.4127, 0.1832, 0.5501, 0.5045], [0.5864, 0.3865, 0.8691, 0.4170, 0.3943, 0.4823, 0.5247, 0.5135], [0.6240, 0.3902, 0.8571, 0.5376, 0.4267, 0.5107, 0.6087, 0.5094], [0.6303, 0.4107, 0.8826, 0.4569, 0.3914, 0.4682, 0.5766, 0.5644], [0.6528, 0.4362, 0.8672, 0.4647, 0.3807, 0.3725, 0.5555, 0.5838], [0.0477, 0.0448, 0.8623, 0.2045, 0.5571, 0.2246, 0.7361, 0.5878], [0.5963, 0.4004, 0.9031, 0.4261, 0.3855, 0.3173, 0.5781, 0.5401], [0.6139, 0.3800, 0.8932, 0.5163, 0.3783, 0.4553, 0.6279, 0.4962]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6055, 0.4015, 0.7425, 0.2033, 0.4112, 0.1883, 0.5217, 0.4823], [0.6081, 0.3950, 0.8537, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954], [0.6250, 0.4013, 0.8525, 0.5417, 0.4038, 0.5117, 0.6325, 0.5017], [0.6197, 0.4091, 0.8800, 0.4783, 0.3537, 0.4767, 0.5950, 0.5550], [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5638, 0.5633], [0.0000, 0.0000, 0.8525, 0.2217, 0.5412, 0.2367, 0.7367, 0.5482], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.0038243775634327903 step: 9 running loss: 0.00042493084038142115 Train Steps: 9/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367], [0.6200, 0.3993, 0.8519, 0.4923, 0.3962, 0.4717, 0.6013, 0.5433], [0.6193, 0.4108, 0.7438, 0.2700, 0.3650, 0.3683, 0.6238, 0.5717], [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320], [0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183], [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5691, 0.3796, 0.8062, 0.2591, 0.4247, 0.2604, 0.6077, 0.5386], [0.5695, 0.3792, 0.8748, 0.5051, 0.4170, 0.4710, 0.6073, 0.5523], [0.5765, 0.3793, 0.7552, 0.2764, 0.3750, 0.3704, 0.6181, 0.5690], [0.5474, 0.3674, 0.8702, 0.3460, 0.3741, 0.4950, 0.6105, 0.5452], [0.5676, 0.3823, 0.8079, 0.3202, 0.3746, 0.3489, 0.5917, 0.5139], [0.5589, 0.3677, 0.8124, 0.2105, 0.4556, 0.2762, 0.6631, 0.5270], [0.5715, 0.3594, 0.7832, 0.2142, 0.4573, 0.1787, 0.5668, 0.5262], [0.3931, 0.2683, 0.7582, 0.2373, 0.4466, 0.1760, 0.5537, 0.5485]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6204, 0.4110, 0.7912, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367], [0.6200, 0.3993, 0.8519, 0.4923, 0.3963, 0.4717, 0.6012, 0.5433], [0.6193, 0.4108, 0.7437, 0.2700, 0.3650, 0.3683, 0.6237, 0.5717], [0.6168, 0.4029, 0.8523, 0.3417, 0.3587, 0.5000, 0.6125, 0.5400], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320], [0.6213, 0.4001, 0.7713, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183], [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240]]], device='cuda:0') loss_train_step before backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train: 0.005475856611155905 step: 10 running loss: 0.0005475856611155905 Train Steps: 10/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268], [0.6183, 0.4076, 0.8838, 0.4517, 0.3813, 0.4483, 0.5775, 0.5633], [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083], [0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767], [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431], [0.6138, 0.4101, 0.8800, 0.5083, 0.4637, 0.5950, 0.5587, 0.5077], [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5778, 0.3631, 0.8630, 0.4630, 0.4427, 0.4802, 0.6320, 0.5042], [0.5571, 0.3702, 0.8847, 0.4294, 0.3884, 0.4266, 0.5899, 0.5506], [0.5534, 0.3684, 0.8602, 0.4760, 0.3661, 0.4216, 0.5846, 0.5699], [0.5960, 0.3879, 0.7611, 0.2382, 0.4183, 0.2803, 0.6296, 0.5309], [0.5524, 0.3774, 0.8657, 0.4767, 0.4540, 0.4675, 0.5284, 0.4991], [0.5111, 0.3414, 0.8815, 0.5135, 0.4777, 0.5534, 0.5801, 0.5005], [0.6022, 0.3933, 0.8279, 0.2484, 0.3888, 0.2621, 0.6464, 0.4693], [0.5504, 0.3655, 0.7145, 0.2024, 0.3990, 0.2294, 0.5790, 0.5200]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268], [0.6183, 0.4076, 0.8838, 0.4517, 0.3812, 0.4483, 0.5775, 0.5633], [0.6186, 0.4060, 0.8750, 0.5050, 0.3537, 0.4367, 0.5813, 0.6083], [0.6230, 0.4152, 0.7588, 0.2283, 0.4013, 0.2883, 0.6200, 0.5767], [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431], [0.6138, 0.4101, 0.8800, 0.5083, 0.4638, 0.5950, 0.5587, 0.5077], [0.6257, 0.4034, 0.8288, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901], [0.6187, 0.4103, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.006415960393496789 step: 11 running loss: 0.0005832691266815262 Train Steps: 11/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901], [0.6261, 0.3987, 0.8688, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309], [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667], [0.6115, 0.4081, 0.6725, 0.2433, 0.4088, 0.1933, 0.5167, 0.5544], [0.6095, 0.3970, 0.8688, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6084, 0.3981, 0.8588, 0.5233, 0.4600, 0.5367, 0.5680, 0.5006]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6073, 0.3913, 0.7963, 0.2558, 0.4119, 0.2415, 0.6013, 0.5137], [0.5290, 0.3648, 0.7915, 0.2882, 0.3676, 0.2543, 0.5319, 0.4917], [0.6104, 0.4107, 0.8586, 0.4751, 0.4234, 0.5301, 0.6855, 0.5266], [0.0803, 0.0688, 0.7548, 0.2152, 0.3793, 0.2732, 0.5980, 0.5617], [0.5532, 0.3812, 0.6788, 0.2455, 0.4015, 0.1991, 0.5203, 0.5415], [0.5990, 0.3987, 0.8612, 0.4508, 0.4700, 0.4684, 0.5258, 0.5003], [0.5285, 0.3691, 0.8373, 0.2425, 0.4806, 0.1649, 0.6189, 0.4771], [0.6092, 0.4158, 0.8430, 0.5042, 0.4554, 0.5251, 0.5594, 0.4881]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901], [0.6261, 0.3987, 0.8687, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309], [0.0000, 0.0000, 0.7525, 0.2291, 0.3837, 0.3017, 0.6050, 0.5667], [0.6115, 0.4081, 0.6725, 0.2433, 0.4087, 0.1933, 0.5167, 0.5544], [0.6095, 0.3970, 0.8687, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6084, 0.3981, 0.8587, 0.5233, 0.4600, 0.5367, 0.5680, 0.5006]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.007117415210814215 step: 12 running loss: 0.000593117934234518 Train Steps: 12/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220], [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517], [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267], [0.6081, 0.3950, 0.8538, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5904, 0.3898, 0.8998, 0.4425, 0.3572, 0.3174, 0.6593, 0.4781], [0.5662, 0.3711, 0.6896, 0.1980, 0.4226, 0.1710, 0.5716, 0.4987], [0.5754, 0.3873, 0.8808, 0.4244, 0.4162, 0.5290, 0.6554, 0.5356], [0.5711, 0.3879, 0.8730, 0.4059, 0.4015, 0.5412, 0.6156, 0.5187], [0.6003, 0.4078, 0.8135, 0.5540, 0.3982, 0.4318, 0.6006, 0.4987], [0.5563, 0.3824, 0.8493, 0.4039, 0.3722, 0.4691, 0.5495, 0.4847], [0.5480, 0.3759, 0.8244, 0.4778, 0.4327, 0.5286, 0.5453, 0.5120], [0.5987, 0.4079, 0.8454, 0.3477, 0.3766, 0.3348, 0.6164, 0.5470]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220], [0.6241, 0.4142, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.6122, 0.4006, 0.8850, 0.4217, 0.4087, 0.5517, 0.6062, 0.5517], [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267], [0.6081, 0.3950, 0.8537, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5987, 0.5667]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.007744449147139676 step: 13 running loss: 0.0005957268574722827 Train Steps: 13/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183], [0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927], [0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6222, 0.3957, 0.8838, 0.5017, 0.3937, 0.4600, 0.5900, 0.5017], [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131], [ nan, nan, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819], [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6580, 0.4510, 0.7884, 0.2578, 0.3902, 0.2608, 0.5653, 0.5235], [0.6374, 0.4277, 0.8697, 0.5432, 0.3760, 0.4533, 0.6301, 0.4782], [0.6094, 0.4081, 0.8409, 0.5083, 0.4566, 0.5553, 0.5343, 0.5236], [0.6468, 0.4245, 0.8740, 0.4897, 0.3949, 0.4702, 0.5960, 0.4987], [0.6265, 0.4372, 0.7443, 0.2968, 0.3676, 0.2812, 0.5386, 0.5554], [0.6135, 0.4098, 0.8615, 0.3593, 0.3455, 0.3919, 0.5960, 0.4974], [0.1953, 0.1307, 0.6816, 0.2437, 0.3851, 0.2490, 0.5604, 0.5534], [0.5964, 0.4031, 0.7323, 0.1757, 0.4227, 0.2616, 0.6200, 0.5441]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6204, 0.4049, 0.7975, 0.2700, 0.3938, 0.2567, 0.5700, 0.5183], [0.6321, 0.4048, 0.8737, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927], [0.6075, 0.4000, 0.8512, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6222, 0.3957, 0.8838, 0.5017, 0.3938, 0.4600, 0.5900, 0.5017], [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5462, 0.5800], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131], [0.0000, 0.0000, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819], [0.6200, 0.4071, 0.7337, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.008856130370986648 step: 14 running loss: 0.0006325807407847606 Train Steps: 14/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882], [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583], [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5724, 0.3819, 0.8246, 0.3438, 0.4118, 0.2549, 0.5484, 0.5102], [0.6460, 0.4393, 0.8570, 0.5261, 0.3701, 0.4076, 0.5309, 0.5025], [0.7027, 0.4749, 0.8744, 0.4550, 0.4139, 0.5604, 0.6192, 0.5783], [0.6758, 0.4445, 0.8976, 0.4109, 0.3588, 0.3803, 0.5950, 0.5482], [0.1091, 0.0646, 0.6863, 0.2242, 0.4313, 0.2017, 0.5191, 0.5573], [0.6893, 0.4546, 0.7051, 0.2249, 0.3795, 0.2830, 0.6013, 0.5383], [0.6455, 0.4368, 0.8594, 0.4563, 0.4097, 0.5432, 0.5876, 0.5352], [0.6468, 0.4288, 0.8710, 0.4869, 0.4198, 0.5645, 0.6359, 0.5339]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6143, 0.4040, 0.8238, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882], [0.6189, 0.4049, 0.8888, 0.4417, 0.4212, 0.5200, 0.5987, 0.5633], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.0000, 0.0000, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6087, 0.5583], [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5838, 0.5500], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.00981652743939776 step: 15 running loss: 0.0006544351626265173 Train Steps: 15/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6092, 0.4001, 0.8638, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064], [0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123], [0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333], [0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5996, 0.3768, 0.8086, 0.2620, 0.4501, 0.1922, 0.6225, 0.5224], [0.6341, 0.4035, 0.8394, 0.4824, 0.4247, 0.5578, 0.5244, 0.5363], [0.6212, 0.4082, 0.8375, 0.3579, 0.3429, 0.4226, 0.5650, 0.5385], [0.6247, 0.4267, 0.8590, 0.4681, 0.3597, 0.4251, 0.5458, 0.5396], [0.6561, 0.4138, 0.8484, 0.5444, 0.4495, 0.5152, 0.5933, 0.5391], [0.6217, 0.4007, 0.8524, 0.4722, 0.3707, 0.5247, 0.6042, 0.5369], [0.6051, 0.4039, 0.8702, 0.4627, 0.3706, 0.4843, 0.6178, 0.5528], [0.6225, 0.3987, 0.8799, 0.4781, 0.3568, 0.3649, 0.5072, 0.5542]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6092, 0.4001, 0.8637, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064], [0.6213, 0.4131, 0.8438, 0.3550, 0.3512, 0.4400, 0.5716, 0.5123], [0.6108, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6202, 0.4053, 0.8637, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6137, 0.5333], [0.6127, 0.4084, 0.8737, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.010077989150886424 step: 16 running loss: 0.0006298743219304015 Train Steps: 16/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133], [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6250, 0.4103, 0.8950, 0.4400, 0.3912, 0.5650, 0.6050, 0.5133], [0.6132, 0.4037, 0.6963, 0.2217, 0.4100, 0.1950, 0.5395, 0.5175], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843], [0.6083, 0.3957, 0.8638, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980], [0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5917, 0.3692, 0.8392, 0.2501, 0.4780, 0.1923, 0.6008, 0.5347], [0.6336, 0.4016, 0.8444, 0.5631, 0.3798, 0.4912, 0.6907, 0.6007], [0.6221, 0.3998, 0.8935, 0.4706, 0.3912, 0.5854, 0.5955, 0.5473], [0.6236, 0.4051, 0.7214, 0.2565, 0.3991, 0.2110, 0.5162, 0.5318], [0.6419, 0.4099, 0.8480, 0.5898, 0.3739, 0.4971, 0.6184, 0.5603], [0.6154, 0.3999, 0.9063, 0.4833, 0.3787, 0.5135, 0.6094, 0.5174], [0.6243, 0.3887, 0.8631, 0.5044, 0.4348, 0.5294, 0.5217, 0.5199], [0.6385, 0.3931, 0.7728, 0.2902, 0.4140, 0.2608, 0.5593, 0.5696]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6266, 0.4065, 0.8313, 0.2467, 0.4787, 0.1733, 0.6313, 0.5133], [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6250, 0.4103, 0.8950, 0.4400, 0.3913, 0.5650, 0.6050, 0.5133], [0.6132, 0.4037, 0.6963, 0.2217, 0.4100, 0.1950, 0.5395, 0.5175], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6264, 0.4049, 0.8988, 0.4633, 0.3812, 0.4983, 0.6326, 0.4843], [0.6083, 0.3957, 0.8637, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980], [0.6229, 0.4066, 0.7613, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.01048219391668681 step: 17 running loss: 0.0006165996421580477 Train Steps: 17/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550], [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869], [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6092, 0.4001, 0.8638, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064], [0.6246, 0.4028, 0.8738, 0.4867, 0.4088, 0.5667, 0.6362, 0.5200], [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[6.8750e-01, 4.4056e-01, 9.0004e-01, 4.9645e-01, 3.8054e-01, 4.6319e-01, 5.7499e-01, 5.7014e-01], [7.5926e-01, 4.6508e-01, 8.8464e-01, 3.6296e-01, 4.5571e-01, 1.9916e-01, 5.7849e-01, 5.0027e-01], [6.7550e-01, 4.1488e-01, 8.6852e-01, 5.8610e-01, 4.1434e-01, 4.5401e-01, 5.9690e-01, 5.4086e-01], [6.8842e-01, 4.4305e-01, 8.9183e-01, 4.6133e-01, 3.9197e-01, 4.4541e-01, 5.1547e-01, 5.5174e-01], [6.6559e-01, 4.2687e-01, 8.7362e-01, 5.4431e-01, 4.1048e-01, 5.4079e-01, 7.0992e-01, 5.4931e-01], [6.8891e-01, 4.2708e-01, 8.7033e-01, 5.0001e-01, 4.3708e-01, 5.4782e-01, 5.0905e-01, 5.1298e-01], [6.8797e-01, 4.2947e-01, 8.8706e-01, 5.1611e-01, 4.3128e-01, 5.7506e-01, 6.0441e-01, 5.3036e-01], [5.4723e-02, 5.4600e-04, 7.2110e-01, 2.2487e-01, 4.1556e-01, 2.5140e-01, 5.0913e-01, 5.2042e-01]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6197, 0.4091, 0.8800, 0.4783, 0.3537, 0.4767, 0.5950, 0.5550], [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869], [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.6339, 0.4123, 0.8637, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6092, 0.4001, 0.8637, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064], [0.6246, 0.4028, 0.8737, 0.4867, 0.4087, 0.5667, 0.6363, 0.5200], [0.0000, 0.0000, 0.6900, 0.1917, 0.3938, 0.2367, 0.5240, 0.5246]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.011588190231123008 step: 18 running loss: 0.0006437883461735004 Train Steps: 18/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633], [0.6350, 0.4043, 0.8738, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950], [0.6197, 0.3986, 0.8800, 0.4617, 0.4188, 0.4783, 0.5687, 0.5550], [0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167], [0.6022, 0.3994, 0.8025, 0.3350, 0.3350, 0.4400, 0.5565, 0.5025], [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6147, 0.3887, 0.8863, 0.4724, 0.4163, 0.5164, 0.5955, 0.5731], [0.6220, 0.3836, 0.8847, 0.6055, 0.3916, 0.4640, 0.6227, 0.5119], [0.6361, 0.4026, 0.8757, 0.4867, 0.4067, 0.4620, 0.5409, 0.5462], [0.6163, 0.3732, 0.7845, 0.2437, 0.4469, 0.1474, 0.5833, 0.5044], [0.6080, 0.3835, 0.8076, 0.3627, 0.3498, 0.4062, 0.5678, 0.5156], [0.5851, 0.3613, 0.8399, 0.2569, 0.4622, 0.2293, 0.7007, 0.5473], [0.6045, 0.3970, 0.8956, 0.5041, 0.4461, 0.5515, 0.5509, 0.5079], [0.5926, 0.3736, 0.8545, 0.4993, 0.4274, 0.4609, 0.5187, 0.5249]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6189, 0.4049, 0.8888, 0.4417, 0.4212, 0.5200, 0.5987, 0.5633], [0.6350, 0.4043, 0.8737, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950], [0.6197, 0.3986, 0.8800, 0.4617, 0.4187, 0.4783, 0.5688, 0.5550], [0.6204, 0.4007, 0.7837, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167], [0.6022, 0.3994, 0.8025, 0.3350, 0.3350, 0.4400, 0.5565, 0.5025], [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6155, 0.4124, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.011935614966205321 step: 19 running loss: 0.0006281902613792274 Train Steps: 19/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6183, 0.4076, 0.8838, 0.4517, 0.3813, 0.4483, 0.5775, 0.5633], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6200, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6263, 0.3925, 0.8732, 0.3921, 0.3849, 0.3956, 0.5455, 0.4974], [0.6533, 0.4087, 0.8859, 0.4419, 0.3958, 0.4445, 0.5874, 0.5733], [0.6267, 0.3943, 0.8719, 0.5444, 0.4120, 0.4272, 0.5390, 0.4761], [0.6193, 0.3927, 0.8507, 0.5603, 0.3943, 0.4842, 0.6691, 0.5243], [0.5988, 0.3827, 0.8665, 0.3997, 0.3696, 0.3102, 0.5047, 0.5236], [0.6087, 0.3742, 0.9097, 0.4067, 0.3816, 0.3709, 0.6709, 0.5170], [0.6347, 0.3979, 0.8637, 0.5529, 0.3940, 0.4270, 0.6110, 0.5263], [0.5986, 0.3606, 0.8703, 0.5145, 0.4212, 0.5126, 0.6077, 0.4776]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6082, 0.4024, 0.8737, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6183, 0.4076, 0.8838, 0.4517, 0.3812, 0.4483, 0.5775, 0.5633], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6133, 0.4094, 0.8495, 0.4028, 0.3587, 0.3200, 0.5003, 0.5407], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6199, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913]]], device='cuda:0') loss_train_step before backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train: 0.012161795937572606 step: 20 running loss: 0.0006080897968786303 Train Steps: 20/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6259, 0.4133, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309], [0.6205, 0.4081, 0.8950, 0.4017, 0.3788, 0.4700, 0.5963, 0.5667], [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6362, 0.5000], [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267], [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116], [0.6346, 0.4165, 0.9138, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471], [0.6126, 0.4067, 0.8638, 0.5383, 0.4188, 0.4850, 0.5016, 0.5392]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5987, 0.3954, 0.8173, 0.2192, 0.4885, 0.1446, 0.6238, 0.4868], [0.5582, 0.3632, 0.8971, 0.4730, 0.3848, 0.3590, 0.6402, 0.5108], [0.6092, 0.3868, 0.8781, 0.3871, 0.3858, 0.4487, 0.6047, 0.5587], [0.5910, 0.3785, 0.8639, 0.5020, 0.3823, 0.4878, 0.6334, 0.5027], [0.5871, 0.3863, 0.8189, 0.5730, 0.4102, 0.4278, 0.5889, 0.5085], [0.5511, 0.3794, 0.8619, 0.4024, 0.4057, 0.5856, 0.5989, 0.5047], [0.6610, 0.4281, 0.9137, 0.3863, 0.4046, 0.4023, 0.7138, 0.5366], [0.6162, 0.4093, 0.8579, 0.5548, 0.4164, 0.4813, 0.5180, 0.5221]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6259, 0.4132, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309], [0.6205, 0.4081, 0.8950, 0.4017, 0.3787, 0.4700, 0.5962, 0.5667], [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6363, 0.5000], [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267], [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116], [0.6346, 0.4165, 0.9137, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471], [0.6126, 0.4067, 0.8637, 0.5383, 0.4187, 0.4850, 0.5016, 0.5392]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.012637686988455243 step: 21 running loss: 0.0006017946184978687 Train Steps: 21/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250], [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864], [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700], [0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6649, 0.4329, 0.9265, 0.4418, 0.3756, 0.4570, 0.6535, 0.5047], [0.6515, 0.4245, 0.8946, 0.5157, 0.3987, 0.5116, 0.7217, 0.5745], [0.6301, 0.4162, 0.8319, 0.2472, 0.4605, 0.1993, 0.6361, 0.4867], [0.6229, 0.3967, 0.9096, 0.5021, 0.4142, 0.5424, 0.6717, 0.4770], [0.6070, 0.4140, 0.7183, 0.2585, 0.4197, 0.2502, 0.5961, 0.5439], [0.1898, 0.1257, 0.6954, 0.2147, 0.4328, 0.2136, 0.5403, 0.5564], [0.6513, 0.4347, 0.8825, 0.5707, 0.3661, 0.5091, 0.6079, 0.5719], [0.6652, 0.4476, 0.8582, 0.4722, 0.3887, 0.4969, 0.5615, 0.5470]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6263, 0.4065, 0.9038, 0.4317, 0.3587, 0.4550, 0.6325, 0.5250], [0.6314, 0.4107, 0.8750, 0.5100, 0.3787, 0.4900, 0.7121, 0.5864], [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.0000, 0.0000, 0.6413, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700], [0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.013873206989956088 step: 22 running loss: 0.0006306003177252768 Train Steps: 22/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6124, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402], [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750], [0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496], [ nan, nan, 0.7335, 0.2569, 0.3788, 0.2667, 0.5066, 0.5578], [0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320], [0.6179, 0.4082, 0.6688, 0.2667, 0.3588, 0.3317, 0.5750, 0.5783]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6789, 0.4531, 0.8366, 0.5151, 0.3868, 0.4881, 0.5628, 0.5444], [ 0.5932, 0.3999, 0.6746, 0.2478, 0.4322, 0.2141, 0.5409, 0.5745], [ 0.6355, 0.4195, 0.8298, 0.2498, 0.4403, 0.2101, 0.6484, 0.4799], [ 0.6366, 0.4174, 0.8163, 0.2554, 0.3985, 0.2939, 0.6574, 0.5077], [ 0.6709, 0.4485, 0.8697, 0.2799, 0.4090, 0.2833, 0.6203, 0.5327], [-0.0907, -0.0473, 0.7337, 0.2581, 0.3986, 0.2763, 0.5389, 0.5449], [ 0.6507, 0.4282, 0.9220, 0.4580, 0.3847, 0.5605, 0.7420, 0.5411], [ 0.6149, 0.4125, 0.6879, 0.2764, 0.3574, 0.3549, 0.5735, 0.5754]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6123, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402], [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5512, 0.5750], [0.6277, 0.4056, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496], [0.0000, 0.0000, 0.7335, 0.2569, 0.3787, 0.2667, 0.5066, 0.5578], [0.6250, 0.3993, 0.9137, 0.4333, 0.3762, 0.5217, 0.6995, 0.5320], [0.6179, 0.4082, 0.6687, 0.2667, 0.3587, 0.3317, 0.5750, 0.5783]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.014488388042082079 step: 23 running loss: 0.0006299299148731338 Train Steps: 23/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350], [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116], [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823], [0.6263, 0.4233, 0.7924, 0.4626, 0.3788, 0.2883, 0.5573, 0.6047], [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118], [0.6289, 0.4032, 0.8419, 0.5446, 0.4075, 0.5017, 0.6312, 0.5117]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6425, 0.4334, 0.7857, 0.2444, 0.4534, 0.2241, 0.6116, 0.5370], [0.5337, 0.3651, 0.7608, 0.1682, 0.3684, 0.2772, 0.5912, 0.5040], [0.6076, 0.4188, 0.9171, 0.4704, 0.3646, 0.5334, 0.6596, 0.5171], [0.5587, 0.3836, 0.8954, 0.4398, 0.3733, 0.4680, 0.5639, 0.5459], [0.4811, 0.3287, 0.7080, 0.1976, 0.3847, 0.2021, 0.5528, 0.4906], [0.5850, 0.4175, 0.8096, 0.4730, 0.3751, 0.3225, 0.5790, 0.6054], [0.5410, 0.3818, 0.7291, 0.3556, 0.4151, 0.2339, 0.5605, 0.6092], [0.6152, 0.3929, 0.8487, 0.5331, 0.3923, 0.5295, 0.6696, 0.5108]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350], [0.6165, 0.4106, 0.7575, 0.1733, 0.3837, 0.2650, 0.5680, 0.5116], [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.6055, 0.4015, 0.7425, 0.2033, 0.4112, 0.1883, 0.5217, 0.4823], [0.6263, 0.4232, 0.7924, 0.4626, 0.3787, 0.2883, 0.5573, 0.6047], [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118], [0.6289, 0.4031, 0.8419, 0.5446, 0.4075, 0.5017, 0.6313, 0.5117]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.015468385492567904 step: 24 running loss: 0.0006445160621903293 Train Steps: 24/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6284, 0.4093, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167], [0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6307, 0.4029, 0.8650, 0.5200, 0.3763, 0.4017, 0.7311, 0.5366], [0.6197, 0.3986, 0.8800, 0.4617, 0.4188, 0.4783, 0.5687, 0.5550], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5988, 0.5283]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6276, 0.4176, 0.8725, 0.4386, 0.3636, 0.3857, 0.6137, 0.5281], [0.5841, 0.3924, 0.8455, 0.4871, 0.3654, 0.3801, 0.5033, 0.5044], [0.5597, 0.3609, 0.8477, 0.4645, 0.4077, 0.4724, 0.5645, 0.5286], [0.5953, 0.3937, 0.8515, 0.4216, 0.3513, 0.3099, 0.5330, 0.5782], [0.6220, 0.4098, 0.8538, 0.4849, 0.3751, 0.3998, 0.7113, 0.5320], [0.6160, 0.4092, 0.8647, 0.4288, 0.4002, 0.4723, 0.5555, 0.5552], [0.6098, 0.3939, 0.8777, 0.4970, 0.3560, 0.4419, 0.6439, 0.5046], [0.5940, 0.3962, 0.8593, 0.5128, 0.3781, 0.4755, 0.5813, 0.5490]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6284, 0.4092, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167], [0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6307, 0.4029, 0.8650, 0.5200, 0.3762, 0.4017, 0.7311, 0.5366], [0.6197, 0.3986, 0.8800, 0.4617, 0.4187, 0.4783, 0.5688, 0.5550], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5987, 0.5283]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.015839603831409477 step: 25 running loss: 0.000633584153256379 Train Steps: 25/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.6083, 0.3957, 0.8638, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980], [0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413], [0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967], [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862], [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5226, 0.3504, 0.8105, 0.2154, 0.5130, 0.2438, 0.6988, 0.5625], [0.5402, 0.3682, 0.8372, 0.4816, 0.4074, 0.4962, 0.5079, 0.5018], [0.5685, 0.3850, 0.8234, 0.5167, 0.4233, 0.5160, 0.4992, 0.5361], [0.5714, 0.3812, 0.8884, 0.3620, 0.4291, 0.2457, 0.6756, 0.5418], [0.6046, 0.4120, 0.8562, 0.4381, 0.3657, 0.2545, 0.5891, 0.5079], [0.5323, 0.3624, 0.7989, 0.2456, 0.4330, 0.1772, 0.5702, 0.5084], [0.5645, 0.3777, 0.8732, 0.3930, 0.3376, 0.4368, 0.5953, 0.5110], [0.5504, 0.3856, 0.8446, 0.3207, 0.3229, 0.3875, 0.5939, 0.5799]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6300, 0.4133, 0.8537, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.6083, 0.3957, 0.8637, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980], [0.6075, 0.4000, 0.8512, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6271, 0.4040, 0.9137, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413], [0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967], [0.6275, 0.4111, 0.8462, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862], [0.6257, 0.4167, 0.8775, 0.3433, 0.3562, 0.4133, 0.6200, 0.5667]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.01718238960893359 step: 26 running loss: 0.0006608611388051381 Train Steps: 26/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954], [ nan, nan, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650], [0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656], [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823], [0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285], [0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960], [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6554, 0.4223, 0.8275, 0.4110, 0.3327, 0.3789, 0.5315, 0.5074], [-0.0383, -0.0347, 0.6959, 0.2131, 0.4342, 0.2363, 0.5394, 0.5475], [ 0.7011, 0.4454, 0.8374, 0.4684, 0.3629, 0.4386, 0.5320, 0.5673], [ 0.6363, 0.4132, 0.7260, 0.2594, 0.4300, 0.2263, 0.5920, 0.6204], [ 0.6163, 0.4085, 0.7112, 0.2127, 0.3985, 0.1573, 0.5243, 0.5062], [ 0.5961, 0.3852, 0.9198, 0.5011, 0.4065, 0.5625, 0.6961, 0.5370], [ 0.7267, 0.4630, 0.8801, 0.3093, 0.4681, 0.1737, 0.6029, 0.5127], [ 0.5984, 0.3974, 0.7103, 0.1963, 0.3715, 0.2508, 0.5763, 0.5615]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6076, 0.3952, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954], [0.0000, 0.0000, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650], [0.6091, 0.3997, 0.8314, 0.4334, 0.3787, 0.4550, 0.5213, 0.5656], [0.6193, 0.4078, 0.7287, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6055, 0.4015, 0.7425, 0.2033, 0.4112, 0.1883, 0.5217, 0.4823], [0.6254, 0.3993, 0.8988, 0.4767, 0.3988, 0.5517, 0.6955, 0.5285], [0.6259, 0.4156, 0.8813, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960], [0.6114, 0.4018, 0.7212, 0.1967, 0.3762, 0.2700, 0.5875, 0.5533]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.01784341588790994 step: 27 running loss: 0.0006608672551077756 Train Steps: 27/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913], [ nan, nan, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650], [0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283], [0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164], [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350], [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5963, 0.6217], [ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.7125, 0.4531, 0.7847, 0.2330, 0.4667, 0.1434, 0.5637, 0.4987], [ 0.0361, 0.0144, 0.6946, 0.1977, 0.4312, 0.2426, 0.5346, 0.5346], [ 0.7027, 0.4471, 0.8598, 0.5141, 0.4767, 0.4845, 0.5079, 0.5195], [ 0.6661, 0.4406, 0.8856, 0.5504, 0.3870, 0.3181, 0.5491, 0.5983], [ 0.7124, 0.4633, 0.7861, 0.2478, 0.4556, 0.1856, 0.5815, 0.5314], [ 0.7486, 0.4859, 0.7194, 0.2394, 0.4215, 0.1986, 0.5841, 0.6084], [-0.0305, -0.0297, 0.6597, 0.2384, 0.4110, 0.2072, 0.5175, 0.5639], [ 0.6786, 0.4400, 0.8910, 0.3322, 0.4522, 0.3354, 0.7192, 0.5160]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6286, 0.4078, 0.8062, 0.2267, 0.4787, 0.1533, 0.5953, 0.4913], [0.0000, 0.0000, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650], [0.6084, 0.4008, 0.8587, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283], [0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164], [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350], [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5962, 0.6217], [0.0000, 0.0000, 0.6688, 0.2513, 0.4112, 0.2117, 0.5193, 0.5933], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.018996318729477935 step: 28 running loss: 0.0006784399546242119 Train Steps: 28/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507], [0.6277, 0.4083, 0.8350, 0.2717, 0.4562, 0.1800, 0.5918, 0.4878], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633], [0.6346, 0.4165, 0.9138, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471], [0.6259, 0.4133, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6224, 0.4061, 0.8988, 0.4300, 0.3838, 0.4750, 0.6112, 0.5483]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5747, 0.3813, 0.6918, 0.2215, 0.3995, 0.2139, 0.5478, 0.5785], [0.6207, 0.4144, 0.8046, 0.2775, 0.4792, 0.1880, 0.5633, 0.5005], [0.5985, 0.3829, 0.8660, 0.3888, 0.3766, 0.3070, 0.5694, 0.5266], [0.5662, 0.3804, 0.7525, 0.2375, 0.3794, 0.3052, 0.6041, 0.5262], [0.5694, 0.3678, 0.8466, 0.5783, 0.4286, 0.4979, 0.5485, 0.5595], [0.6115, 0.4068, 0.9104, 0.4118, 0.4048, 0.4170, 0.7017, 0.5510], [0.6342, 0.4301, 0.8145, 0.2441, 0.5075, 0.1358, 0.6096, 0.5103], [0.5751, 0.3817, 0.8953, 0.4413, 0.3908, 0.4685, 0.5948, 0.5521]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6072, 0.4029, 0.7038, 0.2150, 0.3913, 0.2267, 0.5516, 0.5507], [0.6277, 0.4083, 0.8350, 0.2717, 0.4563, 0.1800, 0.5918, 0.4878], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6192, 0.4128, 0.8512, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633], [0.6346, 0.4165, 0.9137, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471], [0.6259, 0.4132, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6224, 0.4061, 0.8988, 0.4300, 0.3837, 0.4750, 0.6112, 0.5483]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.019415606089751236 step: 29 running loss: 0.000669503658267284 Train Steps: 29/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6117, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [ nan, nan, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737], [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750], [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621], [0.6193, 0.4108, 0.7438, 0.2700, 0.3650, 0.3683, 0.6238, 0.5717], [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6362, 0.5000], [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6897, 0.4506, 0.6948, 0.2281, 0.4008, 0.2410, 0.5203, 0.5133], [0.7451, 0.4733, 0.8854, 0.5929, 0.4019, 0.4944, 0.6341, 0.5262], [0.0708, 0.0346, 0.8714, 0.2453, 0.5403, 0.2397, 0.7223, 0.5470], [0.6919, 0.4506, 0.7064, 0.2530, 0.4442, 0.1918, 0.5331, 0.5808], [0.0642, 0.0430, 0.8695, 0.2792, 0.5340, 0.2003, 0.6744, 0.5507], [0.6989, 0.4623, 0.7750, 0.2946, 0.3784, 0.3741, 0.6267, 0.5688], [0.7815, 0.4923, 0.9183, 0.5498, 0.4017, 0.5253, 0.6147, 0.5207], [0.7112, 0.4725, 0.7207, 0.2791, 0.4066, 0.2466, 0.5596, 0.5634]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6116, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.0000, 0.0000, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737], [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5512, 0.5750], [0.0000, 0.0000, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621], [0.6193, 0.4108, 0.7437, 0.2700, 0.3650, 0.3683, 0.6237, 0.5717], [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6363, 0.5000], [0.6162, 0.4134, 0.6700, 0.2467, 0.3963, 0.2533, 0.5738, 0.5467]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.021278868938679807 step: 30 running loss: 0.0007092956312893269 Train Steps: 30/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297], [0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700], [0.6081, 0.3950, 0.8538, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954], [0.6136, 0.4117, 0.8700, 0.5167, 0.4188, 0.5083, 0.5147, 0.5495], [0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6238, 0.5350], [0.6284, 0.4093, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167], [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461], [0.6284, 0.4127, 0.8538, 0.5867, 0.4363, 0.5083, 0.6038, 0.5433]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5884, 0.3731, 0.8906, 0.5048, 0.3881, 0.4043, 0.6788, 0.5305], [0.6297, 0.4199, 0.7994, 0.3490, 0.3894, 0.2737, 0.6291, 0.5945], [0.6037, 0.3834, 0.8661, 0.4282, 0.3997, 0.4526, 0.5736, 0.4915], [0.5989, 0.3808, 0.8840, 0.5084, 0.4319, 0.4836, 0.5536, 0.5220], [0.6533, 0.4188, 0.8890, 0.3762, 0.3981, 0.5358, 0.6787, 0.5290], [0.6001, 0.3861, 0.8929, 0.4558, 0.3773, 0.3699, 0.6534, 0.5160], [0.6126, 0.4025, 0.8269, 0.3528, 0.3623, 0.3757, 0.5925, 0.5358], [0.6032, 0.3817, 0.8537, 0.5689, 0.4503, 0.4667, 0.6179, 0.5194]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297], [0.6212, 0.4171, 0.7875, 0.3633, 0.3812, 0.2933, 0.5675, 0.5700], [0.6081, 0.3950, 0.8537, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954], [0.6136, 0.4117, 0.8700, 0.5167, 0.4187, 0.5083, 0.5147, 0.5495], [0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6237, 0.5350], [0.6284, 0.4092, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167], [0.6058, 0.3978, 0.8288, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461], [0.6284, 0.4127, 0.8537, 0.5867, 0.4363, 0.5083, 0.6037, 0.5433]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.021944493215414695 step: 31 running loss: 0.000707886877916603 Train Steps: 31/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6268, 0.4061, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300], [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344], [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795], [0.6199, 0.4071, 0.7600, 0.2117, 0.4037, 0.2767, 0.6138, 0.5550], [0.6153, 0.4117, 0.8688, 0.5167, 0.4895, 0.5647, 0.5524, 0.5136], [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377], [0.6161, 0.4055, 0.8675, 0.3867, 0.3713, 0.4033, 0.5195, 0.5162], [0.6261, 0.4066, 0.8325, 0.2150, 0.4763, 0.2667, 0.7002, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6376, 0.4051, 0.8445, 0.2461, 0.4565, 0.1849, 0.6426, 0.5119], [0.5506, 0.3638, 0.8830, 0.4665, 0.3697, 0.3586, 0.5518, 0.5292], [0.5575, 0.3506, 0.8518, 0.5144, 0.3972, 0.5224, 0.7341, 0.5738], [0.5898, 0.3928, 0.7805, 0.2062, 0.4126, 0.2712, 0.6223, 0.5467], [0.5837, 0.3926, 0.8823, 0.5018, 0.4803, 0.5185, 0.5774, 0.5102], [0.5228, 0.3175, 0.8610, 0.5440, 0.4067, 0.4438, 0.6141, 0.5258], [0.5061, 0.3396, 0.8566, 0.3850, 0.3814, 0.3897, 0.5234, 0.5217], [0.6307, 0.4236, 0.8495, 0.2266, 0.4794, 0.2508, 0.7256, 0.5409]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6268, 0.4060, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300], [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344], [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795], [0.6199, 0.4071, 0.7600, 0.2117, 0.4038, 0.2767, 0.6137, 0.5550], [0.6154, 0.4117, 0.8687, 0.5167, 0.4895, 0.5647, 0.5524, 0.5136], [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377], [0.6161, 0.4055, 0.8675, 0.3867, 0.3713, 0.4033, 0.5195, 0.5162], [0.6261, 0.4066, 0.8325, 0.2150, 0.4762, 0.2667, 0.7002, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.022897559145349078 step: 32 running loss: 0.0007155487232921587 Train Steps: 32/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312], [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767], [0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100], [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6576, 0.4233, 0.7940, 0.2923, 0.3964, 0.2534, 0.6247, 0.5410], [0.5724, 0.3634, 0.8504, 0.4078, 0.3608, 0.4618, 0.6095, 0.5409], [0.6195, 0.4325, 0.8775, 0.4804, 0.4494, 0.5989, 0.6058, 0.5275], [0.5582, 0.3695, 0.8496, 0.5030, 0.4034, 0.4863, 0.5666, 0.5961], [0.5600, 0.3584, 0.8529, 0.3570, 0.3881, 0.2920, 0.6447, 0.4917], [0.6144, 0.3968, 0.8115, 0.2539, 0.4409, 0.2646, 0.6695, 0.5365], [0.5736, 0.3816, 0.8761, 0.4166, 0.3680, 0.4029, 0.6566, 0.5264], [0.5960, 0.3976, 0.8047, 0.3775, 0.3493, 0.4274, 0.5664, 0.5595]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312], [0.6155, 0.4124, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6163, 0.4001, 0.8788, 0.5033, 0.4013, 0.4633, 0.5337, 0.5767], [0.6277, 0.4036, 0.8687, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100], [0.6058, 0.3978, 0.8288, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.023489262952352874 step: 33 running loss: 0.0007117958470409961 Train Steps: 33/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633], [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166], [ nan, nan, 0.8488, 0.2300, 0.5563, 0.2100, 0.7390, 0.5679], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250], [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083], [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377], [0.6265, 0.4088, 0.8025, 0.1850, 0.4163, 0.2500, 0.6290, 0.4947]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6593, 0.4331, 0.8618, 0.5456, 0.3639, 0.4010, 0.5979, 0.4803], [0.6536, 0.4266, 0.8628, 0.4762, 0.3945, 0.5987, 0.6864, 0.5626], [0.6165, 0.3938, 0.8499, 0.5151, 0.4208, 0.5274, 0.6455, 0.5208], [0.1122, 0.0747, 0.8574, 0.2338, 0.5162, 0.2554, 0.7275, 0.5477], [0.6545, 0.4259, 0.8812, 0.4674, 0.4134, 0.5498, 0.5881, 0.5204], [0.6227, 0.4213, 0.8596, 0.4772, 0.3509, 0.4564, 0.5783, 0.6005], [0.6466, 0.4088, 0.8482, 0.5273, 0.3951, 0.4719, 0.5904, 0.5376], [0.6413, 0.4213, 0.8073, 0.1971, 0.4224, 0.2603, 0.6283, 0.5116]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6272, 0.4071, 0.8737, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6293, 0.4024, 0.8750, 0.5000, 0.4013, 0.5733, 0.7121, 0.5633], [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166], [0.0000, 0.0000, 0.8487, 0.2300, 0.5562, 0.2100, 0.7390, 0.5679], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250], [0.6186, 0.4060, 0.8750, 0.5050, 0.3537, 0.4367, 0.5813, 0.6083], [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377], [0.6265, 0.4088, 0.8025, 0.1850, 0.4162, 0.2500, 0.6290, 0.4947]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.02405052950780373 step: 34 running loss: 0.0007073685149354039 Train Steps: 34/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6081, 0.3950, 0.8538, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954], [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426], [0.6115, 0.4005, 0.8838, 0.3867, 0.3763, 0.4700, 0.5800, 0.5550], [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183], [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233], [0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5737, 0.5700], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6548, 0.4210, 0.8627, 0.4514, 0.3951, 0.4915, 0.5448, 0.4980], [0.1056, 0.0795, 0.7924, 0.3010, 0.4140, 0.2797, 0.5885, 0.5713], [0.6397, 0.4316, 0.8833, 0.4366, 0.3750, 0.3403, 0.6105, 0.5304], [0.6451, 0.4234, 0.8708, 0.4025, 0.3790, 0.4824, 0.5783, 0.5597], [0.6710, 0.4289, 0.8743, 0.4705, 0.4077, 0.5930, 0.6590, 0.5049], [0.6303, 0.4196, 0.8991, 0.4671, 0.3868, 0.5141, 0.6538, 0.5146], [0.6531, 0.4207, 0.8667, 0.5391, 0.3833, 0.5041, 0.6007, 0.5643], [0.6968, 0.4398, 0.8925, 0.4197, 0.3636, 0.4451, 0.6342, 0.5185]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6081, 0.3950, 0.8537, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954], [0.0000, 0.0000, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426], [0.6115, 0.4005, 0.8838, 0.3867, 0.3762, 0.4700, 0.5800, 0.5550], [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183], [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233], [0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5738, 0.5700], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.0247619836445665 step: 35 running loss: 0.0007074852469876143 Train Steps: 35/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6199, 0.4071, 0.7600, 0.2117, 0.4037, 0.2767, 0.6138, 0.5550], [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133], [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220], [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333], [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5583, 0.3769, 0.7909, 0.2271, 0.3903, 0.3063, 0.6082, 0.5704], [0.5810, 0.4007, 0.9030, 0.4943, 0.4276, 0.6022, 0.5849, 0.5148], [0.5632, 0.3702, 0.8655, 0.2920, 0.4494, 0.2302, 0.6230, 0.5132], [0.6146, 0.4037, 0.8404, 0.2668, 0.4663, 0.2104, 0.6326, 0.5392], [0.5761, 0.3819, 0.8072, 0.4148, 0.3366, 0.4056, 0.5089, 0.5536], [0.5267, 0.3407, 0.7181, 0.2295, 0.4272, 0.1918, 0.5570, 0.5498], [0.5548, 0.3801, 0.8961, 0.4657, 0.3552, 0.5163, 0.6202, 0.5391], [0.5590, 0.3516, 0.8754, 0.5194, 0.3642, 0.5391, 0.5913, 0.5152]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6199, 0.4071, 0.7600, 0.2117, 0.4038, 0.2767, 0.6137, 0.5550], [0.6155, 0.4124, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6275, 0.4111, 0.8462, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6266, 0.4065, 0.8313, 0.2467, 0.4787, 0.1733, 0.6313, 0.5133], [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220], [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6137, 0.5333], [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.025636436403146945 step: 36 running loss: 0.0007121232334207485 Train Steps: 36/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983], [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700], [0.6201, 0.4017, 0.8871, 0.4621, 0.3517, 0.4675, 0.5999, 0.5106], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750], [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833], [0.6339, 0.4102, 0.8588, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089], [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5802, 0.3709, 0.8561, 0.5348, 0.3591, 0.4219, 0.5316, 0.5887], [0.5363, 0.3499, 0.8974, 0.4331, 0.4099, 0.2639, 0.5611, 0.5330], [0.5620, 0.3608, 0.8742, 0.4556, 0.3872, 0.5100, 0.5308, 0.5577], [0.5790, 0.3515, 0.8831, 0.4553, 0.3560, 0.4888, 0.5747, 0.5238], [0.5446, 0.3546, 0.7188, 0.2502, 0.3910, 0.3107, 0.6053, 0.5817], [0.6499, 0.4162, 0.8837, 0.5015, 0.3725, 0.5004, 0.5665, 0.4842], [0.6564, 0.4173, 0.8841, 0.3146, 0.4628, 0.2244, 0.6416, 0.4986], [0.6285, 0.4154, 0.8605, 0.3613, 0.3461, 0.3816, 0.5735, 0.5304]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6260, 0.4214, 0.8537, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983], [0.6226, 0.4098, 0.8913, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6111, 0.3995, 0.8788, 0.4567, 0.3812, 0.4833, 0.5450, 0.5700], [0.6201, 0.4017, 0.8871, 0.4621, 0.3517, 0.4675, 0.5999, 0.5106], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750], [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6037, 0.4833], [0.6339, 0.4102, 0.8587, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089], [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.0263910611247411 step: 37 running loss: 0.0007132719222903 Train Steps: 37/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717], [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799], [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567], [0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980], [0.6136, 0.4085, 0.6688, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5990, 0.3769, 0.9140, 0.4488, 0.3714, 0.4375, 0.6351, 0.5066], [0.6271, 0.3925, 0.8840, 0.5031, 0.4382, 0.4743, 0.5681, 0.5371], [0.6133, 0.3925, 0.8837, 0.5092, 0.4141, 0.4641, 0.5364, 0.5583], [0.5439, 0.3584, 0.8692, 0.3327, 0.3847, 0.2954, 0.5554, 0.5394], [0.5175, 0.3475, 0.8748, 0.4092, 0.3998, 0.3059, 0.5479, 0.5673], [0.5929, 0.3816, 0.7363, 0.2116, 0.3944, 0.2486, 0.5592, 0.5466], [0.5964, 0.3888, 0.7736, 0.3105, 0.3548, 0.2785, 0.5229, 0.4887], [0.5833, 0.3863, 0.6853, 0.2263, 0.3996, 0.2373, 0.5364, 0.5512]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6263, 0.4065, 0.9038, 0.4317, 0.3587, 0.4550, 0.6325, 0.5250], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5412, 0.5717], [0.6163, 0.4001, 0.8788, 0.5033, 0.4013, 0.4633, 0.5337, 0.5767], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6250, 0.4236, 0.8637, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799], [0.6108, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567], [0.6079, 0.3964, 0.7420, 0.2958, 0.3562, 0.2917, 0.5351, 0.4980], [0.6136, 0.4085, 0.6687, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.027080467421910726 step: 38 running loss: 0.0007126438795239665 Train Steps: 38/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947], [0.6289, 0.4081, 0.8720, 0.3487, 0.3900, 0.3183, 0.6703, 0.5376], [0.6289, 0.4032, 0.8419, 0.5446, 0.4075, 0.5017, 0.6312, 0.5117], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567], [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6115, 0.3998, 0.7063, 0.2383, 0.4037, 0.1950, 0.5320, 0.4993]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5684, 0.3738, 0.7840, 0.1812, 0.4079, 0.1931, 0.6037, 0.5029], [0.5567, 0.3666, 0.8722, 0.3481, 0.3623, 0.2860, 0.6309, 0.5371], [0.5954, 0.3750, 0.8283, 0.5332, 0.3828, 0.4742, 0.5910, 0.4946], [0.5968, 0.3983, 0.8844, 0.4523, 0.4209, 0.5508, 0.5773, 0.5281], [0.5925, 0.3815, 0.8645, 0.4634, 0.4638, 0.4607, 0.5284, 0.5370], [0.6031, 0.3976, 0.8330, 0.2334, 0.4379, 0.1981, 0.6606, 0.5579], [0.5907, 0.3900, 0.8811, 0.4777, 0.4345, 0.5356, 0.5625, 0.5505], [0.5687, 0.3794, 0.7224, 0.2112, 0.3896, 0.1612, 0.4978, 0.4984]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947], [0.6289, 0.4081, 0.8720, 0.3487, 0.3900, 0.3183, 0.6703, 0.5376], [0.6289, 0.4031, 0.8419, 0.5446, 0.4075, 0.5017, 0.6313, 0.5117], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6112, 0.4029, 0.8637, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567], [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517], [0.6115, 0.3998, 0.7063, 0.2383, 0.4038, 0.1950, 0.5320, 0.4993]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.027903765891096555 step: 39 running loss: 0.0007154811766947835 Train Steps: 39/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6207, 0.4110, 0.8738, 0.5000, 0.4800, 0.5633, 0.6300, 0.5433], [0.6125, 0.3974, 0.7725, 0.2517, 0.3538, 0.3317, 0.5887, 0.5500], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748], [0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164], [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183], [0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272], [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6086, 0.3948, 0.8793, 0.4601, 0.4524, 0.5521, 0.5810, 0.5304], [0.6241, 0.4068, 0.7623, 0.2233, 0.3301, 0.3224, 0.5753, 0.5376], [0.5699, 0.3948, 0.7303, 0.2490, 0.4237, 0.1971, 0.5398, 0.5759], [0.5910, 0.3813, 0.9273, 0.4244, 0.4138, 0.2584, 0.7003, 0.5499], [0.6257, 0.4145, 0.8526, 0.4932, 0.3787, 0.3170, 0.5312, 0.5960], [0.6313, 0.4051, 0.8683, 0.4292, 0.3831, 0.5610, 0.6172, 0.4978], [0.6149, 0.4161, 0.8498, 0.3317, 0.3739, 0.2650, 0.5100, 0.5272], [0.6453, 0.4217, 0.8295, 0.5290, 0.3750, 0.4164, 0.5556, 0.5208]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6207, 0.4110, 0.8737, 0.5000, 0.4800, 0.5633, 0.6300, 0.5433], [0.6125, 0.3974, 0.7725, 0.2517, 0.3537, 0.3317, 0.5888, 0.5500], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6329, 0.4196, 0.9237, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748], [0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164], [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183], [0.6186, 0.4080, 0.8625, 0.3483, 0.3787, 0.2650, 0.5320, 0.5272], [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.028414304062607698 step: 40 running loss: 0.0007103576015651924 Train Steps: 40/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050], [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207], [0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656], [ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767], [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6467, 0.4431, 0.8648, 0.3886, 0.3671, 0.4597, 0.6357, 0.5992], [0.6807, 0.4511, 0.8120, 0.2432, 0.4405, 0.1834, 0.6839, 0.5110], [0.6613, 0.4373, 0.8320, 0.4391, 0.3748, 0.4331, 0.5437, 0.5581], [0.2648, 0.1831, 0.7805, 0.2907, 0.4217, 0.2114, 0.5688, 0.5838], [0.7139, 0.4862, 0.8317, 0.5578, 0.3945, 0.4067, 0.6070, 0.5315], [0.6573, 0.4524, 0.8424, 0.4128, 0.3647, 0.4494, 0.5517, 0.5118], [0.6732, 0.4647, 0.8023, 0.3019, 0.3642, 0.2834, 0.5257, 0.5534], [0.6547, 0.4513, 0.8251, 0.3458, 0.3673, 0.4919, 0.6024, 0.5175]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6200, 0.4112, 0.8863, 0.4100, 0.3638, 0.4917, 0.6087, 0.6050], [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207], [0.6091, 0.3997, 0.8314, 0.4334, 0.3787, 0.4550, 0.5213, 0.5656], [0.0000, 0.0000, 0.7613, 0.3250, 0.4038, 0.2533, 0.5437, 0.5767], [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [0.6084, 0.4005, 0.8400, 0.4317, 0.3762, 0.4750, 0.5476, 0.5058], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6132, 0.4118, 0.8200, 0.3633, 0.3562, 0.5400, 0.5787, 0.5136]]], device='cuda:0') loss_train_step before backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0026, device='cuda:0', grad_fn=) loss_train: 0.03098702679562848 step: 41 running loss: 0.0007557811413567922 Train Steps: 41/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6307, 0.4029, 0.8650, 0.5200, 0.3763, 0.4017, 0.7311, 0.5366], [0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167], [ nan, nan, 0.8363, 0.3317, 0.3563, 0.3367, 0.5329, 0.5142], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350], [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7133, 0.4949, 0.8636, 0.4180, 0.4097, 0.5950, 0.5953, 0.5434], [0.7292, 0.4689, 0.8951, 0.4901, 0.3736, 0.5000, 0.6631, 0.5112], [0.6799, 0.4530, 0.8630, 0.5163, 0.3925, 0.4040, 0.7449, 0.5381], [0.6545, 0.4295, 0.7728, 0.1951, 0.4598, 0.1320, 0.6202, 0.5347], [0.1823, 0.1531, 0.8082, 0.3131, 0.3363, 0.3036, 0.5638, 0.5512], [0.6235, 0.4221, 0.7693, 0.3914, 0.3424, 0.4040, 0.5297, 0.5574], [0.7180, 0.4684, 0.8529, 0.5158, 0.4348, 0.4772, 0.5545, 0.5509], [0.6849, 0.4614, 0.8755, 0.4868, 0.4155, 0.4720, 0.5656, 0.5568]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116], [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6307, 0.4029, 0.8650, 0.5200, 0.3762, 0.4017, 0.7311, 0.5366], [0.6204, 0.4007, 0.7837, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167], [0.0000, 0.0000, 0.8363, 0.3317, 0.3562, 0.3367, 0.5329, 0.5142], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6126, 0.3954, 0.8537, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350], [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261]]], device='cuda:0') loss_train_step before backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0021, device='cuda:0', grad_fn=) loss_train: 0.033130172334495 step: 42 running loss: 0.0007888136270117858 Train Steps: 42/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6040, 0.4002, 0.7338, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285], [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6275, 0.4157, 0.8337, 0.5800, 0.3763, 0.4200, 0.5547, 0.6125], [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6053, 0.3980, 0.7228, 0.2082, 0.3935, 0.2359, 0.5442, 0.5130], [0.5986, 0.3876, 0.7927, 0.2864, 0.3970, 0.2688, 0.5976, 0.5458], [0.5661, 0.3729, 0.7679, 0.2413, 0.4758, 0.1661, 0.5989, 0.5626], [0.6387, 0.4324, 0.9024, 0.4804, 0.4105, 0.5948, 0.7053, 0.5499], [0.6400, 0.4180, 0.8542, 0.5340, 0.4257, 0.5459, 0.5904, 0.5136], [0.6048, 0.4021, 0.7705, 0.2558, 0.4410, 0.1949, 0.5803, 0.5480], [0.6067, 0.4071, 0.8136, 0.5844, 0.3786, 0.4786, 0.5582, 0.6415], [0.5370, 0.3652, 0.8080, 0.2506, 0.4418, 0.2339, 0.6032, 0.5683]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6040, 0.4002, 0.7337, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778], [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6245, 0.4100, 0.7763, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6254, 0.3993, 0.8988, 0.4767, 0.3988, 0.5517, 0.6955, 0.5285], [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6275, 0.4157, 0.8338, 0.5800, 0.3762, 0.4200, 0.5547, 0.6125], [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.033696947430144064 step: 43 running loss: 0.0007836499402359084 Train Steps: 43/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804], [0.6268, 0.4029, 0.8500, 0.2683, 0.3937, 0.3500, 0.6860, 0.5297], [0.6336, 0.4086, 0.8900, 0.3950, 0.3900, 0.2950, 0.6504, 0.5066], [0.6234, 0.4179, 0.7825, 0.3450, 0.3813, 0.2867, 0.5675, 0.5617], [ nan, nan, 0.8850, 0.3000, 0.5363, 0.2250, 0.7343, 0.5771], [0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609], [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6206, 0.4001, 0.8900, 0.3933, 0.3588, 0.3567, 0.5837, 0.5083]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6887, 0.4348, 0.8442, 0.4089, 0.3687, 0.3939, 0.6161, 0.4755], [ 0.6412, 0.4167, 0.8052, 0.2971, 0.3953, 0.3711, 0.6636, 0.5303], [ 0.7081, 0.4484, 0.8750, 0.4183, 0.4020, 0.3306, 0.6459, 0.5059], [ 0.5995, 0.4066, 0.7609, 0.3685, 0.3983, 0.3026, 0.5671, 0.5638], [-0.0731, -0.0399, 0.8660, 0.2882, 0.5323, 0.2049, 0.7043, 0.6052], [ 0.6184, 0.4114, 0.8384, 0.4840, 0.4189, 0.5765, 0.5598, 0.5455], [ 0.7184, 0.4611, 0.8971, 0.4411, 0.4320, 0.4057, 0.7127, 0.5755], [ 0.6672, 0.4277, 0.8704, 0.4425, 0.3659, 0.3813, 0.5603, 0.5099]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804], [0.6268, 0.4029, 0.8500, 0.2683, 0.3938, 0.3500, 0.6860, 0.5297], [0.6336, 0.4086, 0.8900, 0.3950, 0.3900, 0.2950, 0.6504, 0.5066], [0.6234, 0.4179, 0.7825, 0.3450, 0.3812, 0.2867, 0.5675, 0.5617], [0.0000, 0.0000, 0.8850, 0.3000, 0.5362, 0.2250, 0.7343, 0.5771], [0.6064, 0.4019, 0.8650, 0.4517, 0.4038, 0.5367, 0.5703, 0.5609], [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6206, 0.4001, 0.8900, 0.3933, 0.3587, 0.3567, 0.5838, 0.5083]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.0345825739641441 step: 44 running loss: 0.0007859675900941842 Train Steps: 44/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343], [0.6277, 0.4103, 0.8087, 0.5717, 0.4188, 0.4750, 0.5663, 0.6083], [ nan, nan, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220], [0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650], [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5963, 0.6217], [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6366, 0.4079, 0.8650, 0.5747, 0.3955, 0.5086, 0.6662, 0.5088], [0.6423, 0.4086, 0.8256, 0.5419, 0.3860, 0.4891, 0.7217, 0.5036], [0.6544, 0.4121, 0.8499, 0.5703, 0.4315, 0.4819, 0.5847, 0.5896], [0.1003, 0.0558, 0.7747, 0.2473, 0.3955, 0.2725, 0.5176, 0.5151], [0.5841, 0.3731, 0.9133, 0.4263, 0.3761, 0.4109, 0.6111, 0.5613], [0.5823, 0.3797, 0.7245, 0.2530, 0.4314, 0.2382, 0.6005, 0.5880], [0.6369, 0.4191, 0.8270, 0.3852, 0.4887, 0.2643, 0.5677, 0.5828], [0.6239, 0.3974, 0.7727, 0.2497, 0.3892, 0.2870, 0.6044, 0.5587]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6311, 0.3998, 0.7975, 0.5767, 0.3837, 0.4850, 0.7327, 0.5343], [0.6277, 0.4103, 0.8087, 0.5717, 0.4187, 0.4750, 0.5663, 0.6083], [0.0000, 0.0000, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220], [0.6203, 0.4096, 0.8863, 0.4267, 0.3537, 0.4117, 0.6025, 0.5650], [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5962, 0.6217], [0.6299, 0.4303, 0.7962, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.03513337757613044 step: 45 running loss: 0.0007807417239140098 Train Steps: 45/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6179, 0.4040, 0.7412, 0.1850, 0.3825, 0.2783, 0.5837, 0.5600], [ nan, nan, 0.7335, 0.2569, 0.3788, 0.2667, 0.5066, 0.5578], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.6277, 0.4083, 0.8350, 0.2717, 0.4562, 0.1800, 0.5918, 0.4878], [0.6201, 0.4102, 0.7288, 0.2417, 0.4150, 0.2383, 0.6100, 0.5500], [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767], [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609], [0.6193, 0.4108, 0.7438, 0.2700, 0.3650, 0.3683, 0.6238, 0.5717]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6316, 0.4101, 0.7578, 0.2223, 0.4062, 0.2772, 0.5956, 0.5518], [0.0738, 0.0471, 0.7439, 0.2771, 0.4110, 0.2470, 0.5182, 0.5831], [0.6456, 0.3919, 0.9053, 0.5674, 0.3980, 0.4672, 0.6566, 0.4862], [0.5411, 0.3353, 0.8373, 0.2930, 0.4931, 0.2127, 0.6068, 0.4882], [0.6293, 0.3939, 0.7605, 0.2568, 0.4241, 0.2575, 0.6009, 0.5605], [0.6253, 0.3865, 0.9118, 0.4759, 0.4028, 0.3663, 0.5960, 0.5789], [0.6018, 0.3726, 0.8972, 0.3387, 0.4325, 0.2984, 0.7147, 0.5485], [0.5627, 0.3518, 0.7641, 0.2984, 0.3925, 0.3745, 0.6367, 0.5628]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6179, 0.4040, 0.7412, 0.1850, 0.3825, 0.2783, 0.5838, 0.5600], [0.0000, 0.0000, 0.7335, 0.2569, 0.3787, 0.2667, 0.5066, 0.5578], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.6277, 0.4083, 0.8350, 0.2717, 0.4563, 0.1800, 0.5918, 0.4878], [0.6201, 0.4102, 0.7287, 0.2417, 0.4150, 0.2383, 0.6100, 0.5500], [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767], [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609], [0.6193, 0.4108, 0.7437, 0.2700, 0.3650, 0.3683, 0.6237, 0.5717]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.03593642565829214 step: 46 running loss: 0.0007812266447454813 Train Steps: 46/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350], [0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550], [0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5747, 0.3567, 0.8971, 0.5689, 0.3822, 0.3893, 0.5378, 0.5800], [0.6002, 0.3645, 0.8841, 0.5117, 0.4197, 0.4937, 0.5724, 0.5320], [0.5817, 0.3587, 0.7647, 0.2957, 0.4275, 0.2383, 0.5902, 0.5493], [0.5815, 0.3735, 0.8794, 0.4650, 0.3519, 0.4093, 0.5114, 0.5625], [0.5597, 0.3534, 0.7042, 0.2082, 0.3968, 0.2579, 0.5914, 0.5546], [0.5851, 0.3438, 0.8778, 0.2913, 0.5018, 0.2339, 0.7276, 0.5327], [0.5661, 0.3607, 0.9040, 0.5104, 0.3726, 0.4870, 0.6096, 0.5701], [0.5549, 0.3462, 0.6927, 0.2035, 0.3760, 0.2776, 0.5572, 0.5165]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5412, 0.5683], [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390], [0.6229, 0.4066, 0.7613, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350], [0.6151, 0.4125, 0.8737, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483], [0.6187, 0.4103, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.6197, 0.4091, 0.8800, 0.4783, 0.3537, 0.4767, 0.5950, 0.5550], [0.6043, 0.4022, 0.6888, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.03665601489774417 step: 47 running loss: 0.0007799152105903015 Train Steps: 47/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393], [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297], [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136], [0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6047, 0.3709, 0.8453, 0.2752, 0.4076, 0.2719, 0.6861, 0.5352], [ 0.6716, 0.4370, 0.8774, 0.5461, 0.3600, 0.4168, 0.6446, 0.5402], [-0.0229, -0.0220, 0.7496, 0.2566, 0.3989, 0.2077, 0.5066, 0.5667], [ 0.5969, 0.3846, 0.8858, 0.5054, 0.4045, 0.5154, 0.6771, 0.5271], [ 0.6160, 0.4008, 0.8757, 0.4793, 0.4367, 0.4673, 0.5788, 0.5784], [ 0.6028, 0.3947, 0.8308, 0.3613, 0.3697, 0.4870, 0.5969, 0.5300], [ 0.6021, 0.3922, 0.9019, 0.4375, 0.4062, 0.2941, 0.6574, 0.5230], [ 0.5390, 0.3624, 0.7708, 0.2638, 0.3830, 0.2645, 0.5742, 0.5576]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6286, 0.4086, 0.8408, 0.2801, 0.4162, 0.2800, 0.6725, 0.5393], [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297], [0.0000, 0.0000, 0.7515, 0.2708, 0.3988, 0.2267, 0.5163, 0.5567], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6132, 0.4118, 0.8200, 0.3633, 0.3562, 0.5400, 0.5787, 0.5136], [0.6364, 0.4165, 0.9087, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.03708239090337884 step: 48 running loss: 0.0007725498104870591 Train Steps: 48/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235], [0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319], [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5988, 0.5700], [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459], [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391], [0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5392, 0.3393, 0.8966, 0.4762, 0.4050, 0.5081, 0.6086, 0.5354], [0.5919, 0.4023, 0.7150, 0.2929, 0.3263, 0.3258, 0.4798, 0.5289], [0.4935, 0.3126, 0.7921, 0.2691, 0.3399, 0.3424, 0.5831, 0.5745], [0.6140, 0.4021, 0.8272, 0.5483, 0.3646, 0.4634, 0.6519, 0.5380], [0.5755, 0.3592, 0.9023, 0.4503, 0.3514, 0.4043, 0.6516, 0.5154], [0.5746, 0.3584, 0.8680, 0.2639, 0.4762, 0.2027, 0.6777, 0.5342], [0.5433, 0.3501, 0.8058, 0.2698, 0.3872, 0.2175, 0.5622, 0.5379], [0.6003, 0.4070, 0.7896, 0.2489, 0.3604, 0.2387, 0.5386, 0.5305]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235], [0.6109, 0.4041, 0.6975, 0.3167, 0.3512, 0.3383, 0.5153, 0.5319], [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5987, 0.5700], [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459], [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391], [0.6204, 0.4049, 0.7975, 0.2700, 0.3938, 0.2567, 0.5700, 0.5183]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.03830938627652358 step: 49 running loss: 0.0007818242097249711 Train Steps: 49/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967], [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567], [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [ nan, nan, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879], [0.6228, 0.4119, 0.7938, 0.2233, 0.4674, 0.1773, 0.6188, 0.5433]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6319, 0.4052, 0.9060, 0.4367, 0.3820, 0.5091, 0.6136, 0.5904], [0.5923, 0.3935, 0.7427, 0.2612, 0.3888, 0.2576, 0.5829, 0.5510], [0.6742, 0.4459, 0.9098, 0.4459, 0.3704, 0.4277, 0.7321, 0.5875], [0.6198, 0.4075, 0.8555, 0.2908, 0.3433, 0.2992, 0.5808, 0.5473], [0.0068, 0.0069, 0.7769, 0.2453, 0.4023, 0.1820, 0.5361, 0.5231], [0.5984, 0.4031, 0.7380, 0.3369, 0.4850, 0.1821, 0.5506, 0.6082], [0.6052, 0.4033, 0.8811, 0.4277, 0.3689, 0.5446, 0.6375, 0.4999], [0.5963, 0.3933, 0.7923, 0.1988, 0.4512, 0.1713, 0.6343, 0.5319]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6199, 0.4102, 0.8950, 0.4417, 0.4013, 0.5367, 0.6112, 0.5967], [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5688, 0.5567], [0.6346, 0.4144, 0.9087, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899], [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.0000, 0.0000, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879], [0.6228, 0.4119, 0.7937, 0.2233, 0.4674, 0.1773, 0.6187, 0.5433]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.03863833627838176 step: 50 running loss: 0.0007727667255676352 Train Steps: 50/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6173, 0.4114, 0.7325, 0.2500, 0.4213, 0.1917, 0.5338, 0.5700], [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833], [ nan, nan, 0.8850, 0.2817, 0.5112, 0.2183, 0.7184, 0.5436], [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6284, 0.4127, 0.8538, 0.5867, 0.4363, 0.5083, 0.6038, 0.5433], [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183], [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6040, 0.4282, 0.7308, 0.2059, 0.4077, 0.2071, 0.5314, 0.5556], [ 0.6273, 0.4338, 0.6940, 0.2815, 0.3589, 0.2807, 0.5497, 0.5789], [-0.0056, -0.0037, 0.8923, 0.2670, 0.4995, 0.2198, 0.7089, 0.5389], [ 0.6117, 0.4112, 0.8859, 0.3281, 0.3341, 0.4142, 0.6316, 0.5609], [ 0.6467, 0.4352, 0.7644, 0.1785, 0.4511, 0.1633, 0.5915, 0.4808], [ 0.6760, 0.4614, 0.8418, 0.5814, 0.4076, 0.4921, 0.5650, 0.5409], [ 0.6194, 0.4256, 0.7702, 0.2751, 0.3781, 0.3243, 0.6166, 0.6052], [ 0.5795, 0.4023, 0.8600, 0.3549, 0.3603, 0.3462, 0.6070, 0.5654]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6173, 0.4114, 0.7325, 0.2500, 0.4212, 0.1917, 0.5337, 0.5700], [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833], [0.0000, 0.0000, 0.8850, 0.2817, 0.5113, 0.2183, 0.7184, 0.5436], [0.6257, 0.4167, 0.8775, 0.3433, 0.3562, 0.4133, 0.6200, 0.5667], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6284, 0.4127, 0.8537, 0.5867, 0.4363, 0.5083, 0.6037, 0.5433], [0.6196, 0.4094, 0.7563, 0.2817, 0.3938, 0.3183, 0.6012, 0.6183], [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5987, 0.5667]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.03896989063650835 step: 51 running loss: 0.0007641155026766344 Train Steps: 51/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050], [0.6160, 0.4093, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808], [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320], [0.6069, 0.3975, 0.8625, 0.5083, 0.4388, 0.5483, 0.5650, 0.4967], [0.6256, 0.4199, 0.8638, 0.5800, 0.3987, 0.4383, 0.5600, 0.5950], [0.6230, 0.4113, 0.7213, 0.1983, 0.4325, 0.2367, 0.6262, 0.5400], [0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6073, 0.4136, 0.8816, 0.4450, 0.3556, 0.4199, 0.5715, 0.5052], [0.6147, 0.4262, 0.8284, 0.4442, 0.3397, 0.4541, 0.5100, 0.5697], [0.6346, 0.4331, 0.7867, 0.2099, 0.4025, 0.2692, 0.6605, 0.5185], [0.5712, 0.4087, 0.8530, 0.5211, 0.4311, 0.5627, 0.5652, 0.5148], [0.5833, 0.4068, 0.8546, 0.5685, 0.3776, 0.4327, 0.5370, 0.6009], [0.5960, 0.4159, 0.7190, 0.1794, 0.4325, 0.2432, 0.6347, 0.5452], [0.5953, 0.4253, 0.6915, 0.2885, 0.4620, 0.1969, 0.5367, 0.5910], [0.5961, 0.4056, 0.8904, 0.4810, 0.3547, 0.4024, 0.6478, 0.5261]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050], [0.6160, 0.4092, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808], [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320], [0.6069, 0.3975, 0.8625, 0.5083, 0.4387, 0.5483, 0.5650, 0.4967], [0.6256, 0.4199, 0.8637, 0.5800, 0.3988, 0.4383, 0.5600, 0.5950], [0.6230, 0.4113, 0.7212, 0.1983, 0.4325, 0.2367, 0.6263, 0.5400], [0.6270, 0.4266, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.03928698522213381 step: 52 running loss: 0.0007555189465794963 Train Steps: 52/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395], [0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019], [ nan, nan, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6855, 0.4642, 0.8893, 0.5334, 0.3714, 0.3921, 0.6544, 0.5189], [0.7078, 0.4839, 0.7207, 0.2155, 0.4106, 0.2417, 0.6140, 0.5604], [0.5593, 0.3895, 0.8413, 0.5402, 0.4644, 0.5556, 0.5571, 0.5589], [0.6221, 0.4285, 0.6530, 0.2349, 0.4115, 0.2204, 0.5106, 0.5732], [0.6814, 0.4617, 0.7495, 0.1913, 0.4757, 0.1683, 0.6202, 0.5073], [0.6239, 0.4233, 0.7985, 0.2182, 0.5043, 0.1667, 0.6382, 0.5138], [0.6505, 0.4486, 0.8450, 0.4286, 0.3794, 0.5564, 0.5866, 0.5351], [0.0618, 0.0508, 0.6994, 0.2725, 0.3897, 0.2711, 0.5051, 0.5950]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395], [0.6075, 0.4000, 0.8512, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6149, 0.4054, 0.6712, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834], [0.6161, 0.4099, 0.8737, 0.4383, 0.3787, 0.5483, 0.5605, 0.5019], [0.0000, 0.0000, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.040181065662181936 step: 53 running loss: 0.0007581333143807913 Train Steps: 53/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383], [0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901], [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667], [0.6216, 0.4167, 0.8588, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667], [0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6203, 0.4021, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767], [0.6152, 0.4131, 0.6863, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7365, 0.4962, 0.7534, 0.2470, 0.4813, 0.1644, 0.5918, 0.5210], [0.6629, 0.4288, 0.8927, 0.3460, 0.3785, 0.3965, 0.6518, 0.5165], [0.6685, 0.4532, 0.8460, 0.3589, 0.3941, 0.3674, 0.6022, 0.5691], [0.6363, 0.4391, 0.8478, 0.5446, 0.4133, 0.5225, 0.5803, 0.5728], [0.6585, 0.4363, 0.8502, 0.3563, 0.4291, 0.2718, 0.6160, 0.4798], [0.6618, 0.4396, 0.8653, 0.5108, 0.3825, 0.4054, 0.5755, 0.5395], [0.0434, 0.0198, 0.7778, 0.3281, 0.4289, 0.2592, 0.5454, 0.5704], [0.6542, 0.4441, 0.6695, 0.2451, 0.3776, 0.3397, 0.5739, 0.5523]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383], [0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901], [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5987, 0.5667], [0.6216, 0.4167, 0.8587, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667], [0.6277, 0.4036, 0.8687, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6203, 0.4020, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405], [0.0000, 0.0000, 0.7613, 0.3250, 0.4038, 0.2533, 0.5437, 0.5767], [0.6152, 0.4131, 0.6862, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.04089540946006309 step: 54 running loss: 0.0007573223974085758 Train Steps: 54/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901], [0.6246, 0.4028, 0.8738, 0.4867, 0.4088, 0.5667, 0.6362, 0.5200], [ nan, nan, 0.7192, 0.2346, 0.4037, 0.2050, 0.5138, 0.5650], [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500], [0.6250, 0.4110, 0.7238, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633], [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804], [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407], [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6796, 0.4268, 0.9074, 0.3819, 0.3924, 0.3890, 0.6588, 0.5231], [0.6585, 0.4226, 0.8758, 0.5104, 0.4502, 0.5876, 0.6470, 0.5290], [0.0220, 0.0207, 0.7197, 0.2960, 0.4424, 0.2298, 0.4976, 0.5429], [0.6749, 0.4446, 0.7745, 0.2926, 0.3850, 0.4026, 0.6171, 0.5571], [0.6448, 0.4231, 0.7123, 0.2585, 0.4646, 0.1773, 0.5542, 0.5607], [0.6679, 0.4259, 0.8739, 0.3943, 0.3975, 0.3687, 0.6295, 0.4941], [0.6708, 0.4429, 0.8537, 0.4344, 0.3786, 0.3161, 0.5023, 0.5432], [0.7364, 0.4848, 0.7535, 0.2071, 0.4139, 0.2637, 0.5635, 0.5158]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901], [0.6246, 0.4028, 0.8737, 0.4867, 0.4087, 0.5667, 0.6363, 0.5200], [0.0000, 0.0000, 0.7192, 0.2346, 0.4038, 0.2050, 0.5138, 0.5650], [0.6197, 0.4051, 0.7812, 0.2650, 0.3512, 0.4050, 0.6112, 0.5500], [0.6250, 0.4110, 0.7237, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633], [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804], [0.6133, 0.4094, 0.8495, 0.4028, 0.3587, 0.3200, 0.5003, 0.5407], [0.6165, 0.4106, 0.7575, 0.1733, 0.3837, 0.2650, 0.5680, 0.5116]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.04186867403041106 step: 55 running loss: 0.0007612486187347464 Train Steps: 55/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6277, 0.4083, 0.8350, 0.2717, 0.4562, 0.1800, 0.5918, 0.4878], [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567], [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5837, 0.5167], [0.6223, 0.4028, 0.8988, 0.4200, 0.3763, 0.5733, 0.6375, 0.5167], [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6891, 0.4279, 0.8262, 0.2814, 0.4698, 0.1748, 0.5825, 0.4751], [0.6379, 0.4060, 0.7332, 0.2904, 0.4166, 0.2478, 0.5579, 0.5494], [0.5869, 0.3768, 0.7651, 0.2341, 0.3812, 0.2749, 0.5775, 0.4873], [0.5375, 0.3461, 0.8667, 0.5664, 0.4084, 0.4206, 0.5235, 0.5708], [0.6034, 0.3690, 0.8506, 0.3974, 0.3891, 0.2989, 0.5728, 0.5234], [0.5709, 0.3539, 0.8911, 0.4125, 0.3997, 0.5595, 0.6092, 0.5050], [0.5394, 0.3326, 0.8761, 0.5040, 0.4639, 0.5729, 0.6081, 0.5213], [0.6028, 0.3882, 0.8502, 0.2723, 0.4557, 0.2620, 0.7067, 0.5437]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6277, 0.4083, 0.8350, 0.2717, 0.4563, 0.1800, 0.5918, 0.4878], [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5688, 0.5567], [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926], [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5838, 0.5167], [0.6223, 0.4028, 0.8988, 0.4200, 0.3762, 0.5733, 0.6375, 0.5167], [0.6185, 0.4098, 0.8838, 0.4900, 0.4538, 0.5800, 0.6288, 0.5400], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.04255643083888572 step: 56 running loss: 0.0007599362649801021 Train Steps: 56/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6224, 0.4061, 0.8988, 0.4300, 0.3838, 0.4750, 0.6112, 0.5483], [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783], [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667], [ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.6234, 0.4179, 0.7825, 0.3450, 0.3813, 0.2867, 0.5675, 0.5617], [0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064], [0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6451, 0.3948, 0.8907, 0.4315, 0.3767, 0.4683, 0.6095, 0.5213], [ 0.6609, 0.4179, 0.8393, 0.3767, 0.3521, 0.2983, 0.5139, 0.5342], [ 0.6633, 0.4080, 0.8775, 0.3444, 0.3452, 0.4230, 0.6186, 0.5333], [ 0.0151, -0.0232, 0.7885, 0.3428, 0.4042, 0.2410, 0.5492, 0.5479], [ 0.6191, 0.3628, 0.8780, 0.4186, 0.3605, 0.4454, 0.6142, 0.4847], [ 0.7265, 0.4772, 0.7809, 0.3525, 0.3794, 0.2705, 0.5712, 0.5360], [ 0.6252, 0.4042, 0.8688, 0.5054, 0.4890, 0.5129, 0.5313, 0.4892], [ 0.5614, 0.3362, 0.8642, 0.2500, 0.5491, 0.1969, 0.7115, 0.5232]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6224, 0.4061, 0.8988, 0.4300, 0.3837, 0.4750, 0.6112, 0.5483], [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783], [0.6257, 0.4167, 0.8775, 0.3433, 0.3562, 0.4133, 0.6200, 0.5667], [0.0000, 0.0000, 0.7613, 0.3250, 0.4038, 0.2533, 0.5437, 0.5767], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.6234, 0.4179, 0.7825, 0.3450, 0.3812, 0.2867, 0.5675, 0.5617], [0.6128, 0.4022, 0.8737, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064], [0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.04318221162247937 step: 57 running loss: 0.0007575826600434977 Train Steps: 57/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150], [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383], [0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050], [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.0021, -0.0080, 0.7031, 0.2469, 0.4050, 0.1991, 0.5280, 0.5715], [ 0.6362, 0.3893, 0.8048, 0.2570, 0.4196, 0.2289, 0.5890, 0.5236], [ 0.6615, 0.4203, 0.7796, 0.2559, 0.4539, 0.1697, 0.5960, 0.5247], [ 0.7031, 0.4374, 0.7803, 0.1934, 0.4075, 0.2273, 0.5867, 0.4858], [ 0.6264, 0.3888, 0.7986, 0.3021, 0.3568, 0.3290, 0.5865, 0.5070], [ 0.6252, 0.3895, 0.9177, 0.4737, 0.3849, 0.5662, 0.7386, 0.5326], [ 0.6240, 0.3859, 0.8420, 0.5800, 0.3921, 0.5099, 0.5730, 0.6145], [ 0.6473, 0.3830, 0.7823, 0.2062, 0.4194, 0.1886, 0.5721, 0.5217]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150], [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383], [0.6202, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6135, 0.3994, 0.7912, 0.3050, 0.3625, 0.3050, 0.5838, 0.5050], [0.6339, 0.4102, 0.9087, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6213, 0.4001, 0.7713, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.04352180591376964 step: 58 running loss: 0.0007503759640305111 Train Steps: 58/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033], [0.6136, 0.4117, 0.8700, 0.5167, 0.4188, 0.5083, 0.5147, 0.5495], [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050], [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700], [0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6153, 0.4119, 0.8463, 0.3833, 0.3600, 0.3200, 0.5106, 0.5563]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5926, 0.3791, 0.8699, 0.3925, 0.3340, 0.3080, 0.5218, 0.5129], [0.5920, 0.3798, 0.8317, 0.5436, 0.3708, 0.4394, 0.5764, 0.5986], [0.5888, 0.3719, 0.8841, 0.4800, 0.3814, 0.5299, 0.6021, 0.5108], [0.6247, 0.3948, 0.8895, 0.5144, 0.4037, 0.5002, 0.5442, 0.5280], [0.5756, 0.3660, 0.8945, 0.4966, 0.4299, 0.5158, 0.6156, 0.5064], [0.5044, 0.3224, 0.8838, 0.4386, 0.3716, 0.4736, 0.5628, 0.5452], [0.6710, 0.4162, 0.7985, 0.2600, 0.4690, 0.1384, 0.5956, 0.5194], [0.6026, 0.3820, 0.8637, 0.3745, 0.3446, 0.3149, 0.5279, 0.5321]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6133, 0.4094, 0.8495, 0.4028, 0.3587, 0.3200, 0.5003, 0.5407], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033], [0.6136, 0.4117, 0.8700, 0.5167, 0.4187, 0.5083, 0.5147, 0.5495], [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050], [0.6111, 0.3995, 0.8788, 0.4567, 0.3812, 0.4833, 0.5450, 0.5700], [0.6245, 0.4100, 0.7763, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6153, 0.4119, 0.8462, 0.3833, 0.3600, 0.3200, 0.5106, 0.5563]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.04425763736071531 step: 59 running loss: 0.0007501294467917849 Train Steps: 59/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6161, 0.4055, 0.8675, 0.3867, 0.3713, 0.4033, 0.5195, 0.5162], [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6362, 0.5000], [0.6087, 0.3951, 0.8387, 0.5833, 0.4188, 0.4933, 0.5146, 0.4830], [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800], [0.6167, 0.4048, 0.6831, 0.3639, 0.3763, 0.3017, 0.5700, 0.5883], [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896], [0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913], [ nan, nan, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6232, 0.4094, 0.9060, 0.3644, 0.3718, 0.3913, 0.5427, 0.5234], [0.5923, 0.3678, 0.8974, 0.4886, 0.3832, 0.5251, 0.6586, 0.5065], [0.5933, 0.3822, 0.8689, 0.5705, 0.4047, 0.4963, 0.5346, 0.5147], [0.6270, 0.4227, 0.7658, 0.2528, 0.4602, 0.1894, 0.5709, 0.5945], [0.5592, 0.3721, 0.7411, 0.3281, 0.3730, 0.3106, 0.5833, 0.5873], [0.6357, 0.4168, 0.8609, 0.3741, 0.3467, 0.3067, 0.5425, 0.5905], [0.6143, 0.3925, 0.8168, 0.2518, 0.4762, 0.1560, 0.5998, 0.5089], [0.0199, 0.0254, 0.7184, 0.2600, 0.3754, 0.2183, 0.5676, 0.5825]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6161, 0.4055, 0.8675, 0.3867, 0.3713, 0.4033, 0.5195, 0.5162], [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6363, 0.5000], [0.6087, 0.3951, 0.8388, 0.5833, 0.4187, 0.4933, 0.5146, 0.4830], [0.6218, 0.4185, 0.7337, 0.2650, 0.4625, 0.1950, 0.5688, 0.5800], [0.6167, 0.4048, 0.6831, 0.3639, 0.3762, 0.3017, 0.5700, 0.5883], [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896], [0.6286, 0.4078, 0.8062, 0.2267, 0.4787, 0.1533, 0.5953, 0.4913], [0.0000, 0.0000, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.0447188821999589 step: 60 running loss: 0.0007453147033326483 Train Steps: 60/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136], [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700], [0.6214, 0.4112, 0.7838, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083], [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167], [0.6127, 0.4118, 0.8650, 0.5083, 0.4088, 0.5367, 0.5300, 0.5456], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927], [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5633, 0.3762, 0.6976, 0.2267, 0.3762, 0.2533, 0.5589, 0.5271], [0.5295, 0.3590, 0.8761, 0.4640, 0.3860, 0.4623, 0.5525, 0.5737], [0.5826, 0.3891, 0.7838, 0.2237, 0.3603, 0.2857, 0.5810, 0.5268], [0.6006, 0.4065, 0.8024, 0.3331, 0.3506, 0.3795, 0.6129, 0.6237], [0.6266, 0.4274, 0.8681, 0.4975, 0.4155, 0.5203, 0.5503, 0.5702], [0.5805, 0.3821, 0.8804, 0.4771, 0.3678, 0.5073, 0.6005, 0.5190], [0.5910, 0.3802, 0.8822, 0.5528, 0.3864, 0.4164, 0.6233, 0.4992], [0.5723, 0.3710, 0.8342, 0.2592, 0.4869, 0.1630, 0.6223, 0.5278]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6043, 0.4022, 0.6888, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136], [0.6111, 0.3995, 0.8788, 0.4567, 0.3812, 0.4833, 0.5450, 0.5700], [0.6214, 0.4112, 0.7837, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083], [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6037, 0.6167], [0.6127, 0.4118, 0.8650, 0.5083, 0.4087, 0.5367, 0.5300, 0.5456], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6321, 0.4048, 0.8737, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927], [0.6266, 0.4065, 0.8313, 0.2467, 0.4787, 0.1733, 0.6313, 0.5133]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.04516495233110618 step: 61 running loss: 0.0007404090546082981 Train Steps: 61/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6176, 0.3911, 0.8738, 0.4217, 0.3488, 0.4033, 0.6025, 0.4817], [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869], [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6063, 0.5617], [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6147, 0.4107, 0.8137, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412], [0.6200, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367], [0.6193, 0.3930, 0.8949, 0.4437, 0.3852, 0.5435, 0.6263, 0.5263]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5905, 0.3962, 0.7859, 0.2966, 0.4024, 0.2382, 0.5886, 0.5458], [0.5662, 0.3733, 0.8546, 0.4214, 0.3753, 0.3908, 0.5882, 0.5191], [0.5349, 0.3653, 0.8111, 0.2279, 0.4760, 0.1968, 0.6072, 0.5061], [0.6284, 0.4320, 0.8794, 0.4852, 0.3632, 0.4742, 0.5870, 0.5755], [0.5406, 0.3842, 0.7860, 0.2888, 0.4766, 0.1790, 0.5583, 0.5668], [0.5116, 0.3594, 0.7696, 0.3321, 0.3682, 0.2809, 0.4983, 0.5764], [0.6265, 0.4087, 0.8644, 0.4585, 0.3857, 0.5276, 0.5999, 0.5688], [0.5494, 0.3629, 0.8488, 0.4420, 0.3872, 0.5560, 0.6158, 0.5347]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208], [0.6176, 0.3911, 0.8737, 0.4217, 0.3487, 0.4033, 0.6025, 0.4817], [0.6276, 0.4095, 0.8238, 0.2250, 0.4663, 0.1783, 0.6171, 0.4869], [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6062, 0.5617], [0.6229, 0.4107, 0.8138, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6147, 0.4107, 0.8138, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412], [0.6199, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367], [0.6193, 0.3930, 0.8949, 0.4437, 0.3852, 0.5435, 0.6263, 0.5263]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.04607175434648525 step: 62 running loss: 0.0007430928120400847 Train Steps: 62/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767], [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389], [0.6129, 0.4063, 0.8738, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246], [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433], [0.6361, 0.4076, 0.8862, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6707, 0.4670, 0.8677, 0.4491, 0.3788, 0.3431, 0.5712, 0.5661], [0.6315, 0.4356, 0.8234, 0.3544, 0.3526, 0.3590, 0.5785, 0.5125], [0.5871, 0.4064, 0.7431, 0.2339, 0.3850, 0.2712, 0.6034, 0.5493], [0.6304, 0.4422, 0.8382, 0.5349, 0.4207, 0.4679, 0.5292, 0.5819], [0.5833, 0.4111, 0.7829, 0.3156, 0.3633, 0.2995, 0.4954, 0.5502], [0.0056, 0.0311, 0.6748, 0.2038, 0.3992, 0.2276, 0.5228, 0.5308], [0.6268, 0.4436, 0.8411, 0.5354, 0.4784, 0.4864, 0.5279, 0.5470], [0.6584, 0.4570, 0.8667, 0.5405, 0.3740, 0.4484, 0.6698, 0.5254]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767], [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389], [0.6130, 0.4063, 0.8737, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.0000, 0.0000, 0.6900, 0.1917, 0.3938, 0.2367, 0.5240, 0.5246], [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433], [0.6361, 0.4076, 0.8863, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.04644800083769951 step: 63 running loss: 0.0007372698545666589 Train Steps: 63/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6228, 0.4119, 0.7938, 0.2233, 0.4674, 0.1773, 0.6188, 0.5433], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817], [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400], [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312], [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917], [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5616, 0.3987, 0.6612, 0.2353, 0.4208, 0.1947, 0.5107, 0.5745], [0.5573, 0.3817, 0.7776, 0.2782, 0.3594, 0.3186, 0.5920, 0.5305], [0.5528, 0.3848, 0.7682, 0.2134, 0.4712, 0.1687, 0.5880, 0.5412], [0.5727, 0.3796, 0.8123, 0.5452, 0.4452, 0.4979, 0.5041, 0.4830], [0.6293, 0.4308, 0.8392, 0.3437, 0.3631, 0.4830, 0.5959, 0.5178], [0.6417, 0.4228, 0.8615, 0.4075, 0.3550, 0.4194, 0.5687, 0.5197], [0.5581, 0.3868, 0.7555, 0.3847, 0.3587, 0.3180, 0.5133, 0.5807], [0.6269, 0.4197, 0.8552, 0.4713, 0.3764, 0.3860, 0.5693, 0.5327]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6147, 0.4026, 0.6600, 0.2467, 0.4087, 0.2150, 0.5489, 0.5773], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6228, 0.4119, 0.7937, 0.2233, 0.4674, 0.1773, 0.6187, 0.5433], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817], [0.6168, 0.4029, 0.8523, 0.3417, 0.3587, 0.5000, 0.6125, 0.5400], [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312], [0.6179, 0.4118, 0.7278, 0.4237, 0.3587, 0.3400, 0.5675, 0.5917], [0.6262, 0.4163, 0.8850, 0.5183, 0.3762, 0.4150, 0.6025, 0.5500]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.0470950579183409 step: 64 running loss: 0.0007358602799740765 Train Steps: 64/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058], [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841], [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896], [0.6202, 0.4079, 0.8025, 0.2500, 0.3763, 0.3217, 0.6125, 0.5533], [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702], [0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233], [0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6238, 0.5350]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6138, 0.4214, 0.8354, 0.4241, 0.3648, 0.4590, 0.5131, 0.4924], [0.6428, 0.4388, 0.8917, 0.4322, 0.4117, 0.3551, 0.6836, 0.5497], [0.5849, 0.4031, 0.7995, 0.3712, 0.3463, 0.3027, 0.4893, 0.5656], [0.6113, 0.4225, 0.7709, 0.2444, 0.3745, 0.3282, 0.5668, 0.5318], [0.6188, 0.4291, 0.8725, 0.4718, 0.3644, 0.4302, 0.6124, 0.5090], [0.6344, 0.4364, 0.7759, 0.5476, 0.3937, 0.4836, 0.6670, 0.5534], [0.6023, 0.3985, 0.8128, 0.5589, 0.4047, 0.4458, 0.5374, 0.5205], [0.6651, 0.4503, 0.8545, 0.4076, 0.3714, 0.5456, 0.5912, 0.5054]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6084, 0.4005, 0.8400, 0.4317, 0.3762, 0.4750, 0.5476, 0.5058], [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841], [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896], [0.6202, 0.4079, 0.8025, 0.2500, 0.3762, 0.3217, 0.6125, 0.5533], [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6346, 0.4086, 0.7937, 0.5500, 0.3963, 0.4867, 0.7343, 0.5702], [0.6222, 0.3937, 0.8350, 0.5617, 0.4137, 0.4600, 0.5800, 0.5233], [0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6237, 0.5350]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.04768331356171984 step: 65 running loss: 0.0007335894394110745 Train Steps: 65/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6147, 0.4112, 0.7988, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550], [0.6120, 0.4014, 0.6863, 0.2817, 0.3700, 0.2783, 0.5513, 0.5667], [0.6260, 0.4120, 0.8013, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895], [0.6272, 0.4045, 0.8538, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133], [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6108, 0.4151, 0.8757, 0.4770, 0.4467, 0.4963, 0.5417, 0.5156], [0.6673, 0.4383, 0.8964, 0.4764, 0.4516, 0.5921, 0.6215, 0.5118], [0.5872, 0.3994, 0.7957, 0.3206, 0.3719, 0.3037, 0.5088, 0.5784], [0.5984, 0.3938, 0.6967, 0.2951, 0.3714, 0.3026, 0.5459, 0.5644], [0.5771, 0.3856, 0.8076, 0.2430, 0.4911, 0.1712, 0.6498, 0.5274], [0.5973, 0.4015, 0.8380, 0.5838, 0.3782, 0.4484, 0.5942, 0.4922], [0.5895, 0.3874, 0.8228, 0.2661, 0.4062, 0.2829, 0.5728, 0.5299], [0.6101, 0.4115, 0.8977, 0.3835, 0.3710, 0.4011, 0.6480, 0.5377]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6147, 0.4112, 0.7987, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550], [0.6120, 0.4013, 0.6862, 0.2817, 0.3700, 0.2783, 0.5512, 0.5667], [0.6259, 0.4120, 0.8012, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895], [0.6271, 0.4045, 0.8537, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6205, 0.4015, 0.8350, 0.2717, 0.3988, 0.2550, 0.5788, 0.5133], [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.04797610554669518 step: 66 running loss: 0.0007269106901014421 Train Steps: 66/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334], [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517], [0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583], [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6038, 0.3741, 0.9009, 0.5030, 0.3543, 0.5153, 0.6121, 0.5218], [0.6826, 0.4435, 0.8553, 0.2553, 0.5435, 0.2630, 0.7351, 0.5675], [0.6452, 0.4126, 0.7562, 0.2424, 0.4441, 0.2027, 0.5805, 0.5247], [0.5573, 0.3575, 0.7552, 0.2212, 0.4434, 0.2005, 0.5919, 0.5020], [0.6784, 0.4411, 0.8808, 0.4649, 0.4005, 0.5945, 0.5945, 0.5313], [0.6025, 0.4080, 0.8856, 0.4498, 0.3573, 0.4614, 0.5600, 0.5423], [0.6007, 0.3935, 0.7706, 0.2596, 0.4448, 0.2150, 0.5805, 0.5448], [0.6657, 0.4195, 0.8535, 0.5146, 0.4174, 0.5708, 0.7015, 0.5428]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334], [0.6300, 0.4133, 0.8537, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.6204, 0.4007, 0.7837, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6122, 0.4006, 0.8850, 0.4217, 0.4087, 0.5517, 0.6062, 0.5517], [0.6212, 0.4033, 0.8938, 0.4167, 0.3812, 0.4267, 0.5612, 0.5583], [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.048750259753433056 step: 67 running loss: 0.0007276158172154187 Train Steps: 67/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450], [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6110, 0.4047, 0.8700, 0.4483, 0.3713, 0.3967, 0.5088, 0.5517], [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879], [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413], [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426], [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6498, 0.4061, 0.8830, 0.4685, 0.3808, 0.4473, 0.6150, 0.5376], [0.6717, 0.4231, 0.8865, 0.5102, 0.4039, 0.5551, 0.6107, 0.5116], [0.6743, 0.4255, 0.8788, 0.4651, 0.3708, 0.3492, 0.5595, 0.5616], [0.6536, 0.4194, 0.8888, 0.4669, 0.3753, 0.4371, 0.5224, 0.5394], [0.7025, 0.4403, 0.8933, 0.4476, 0.3991, 0.5874, 0.6223, 0.4826], [0.6191, 0.3752, 0.9168, 0.3248, 0.5109, 0.2552, 0.7435, 0.5421], [0.6457, 0.3986, 0.7897, 0.2415, 0.4592, 0.1989, 0.6085, 0.5358], [0.6341, 0.3853, 0.7629, 0.2232, 0.4380, 0.2254, 0.6620, 0.5276]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450], [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033], [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767], [0.6110, 0.4047, 0.8700, 0.4483, 0.3713, 0.3967, 0.5088, 0.5517], [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879], [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413], [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426], [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.049255486417678185 step: 68 running loss: 0.0007243453884952675 Train Steps: 68/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183], [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650], [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413], [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459], [0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6066, 0.3856, 0.8011, 0.2484, 0.4016, 0.2708, 0.5804, 0.5224], [0.6193, 0.3879, 0.8731, 0.3425, 0.3602, 0.5105, 0.6248, 0.5186], [0.6137, 0.3860, 0.7856, 0.2496, 0.3973, 0.2793, 0.5676, 0.5596], [0.6431, 0.3960, 0.9349, 0.3631, 0.4764, 0.2546, 0.7226, 0.5344], [0.6240, 0.3707, 0.9125, 0.4803, 0.3718, 0.4266, 0.6152, 0.5204], [0.6428, 0.3944, 0.7469, 0.2015, 0.4372, 0.2012, 0.6081, 0.5121], [0.6284, 0.3823, 0.8317, 0.5589, 0.4057, 0.4849, 0.6951, 0.5196], [0.6600, 0.4035, 0.9298, 0.4591, 0.4420, 0.2887, 0.7233, 0.5457]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6204, 0.4049, 0.7975, 0.2700, 0.3938, 0.2567, 0.5700, 0.5183], [0.6168, 0.4029, 0.8523, 0.3417, 0.3587, 0.5000, 0.6125, 0.5400], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650], [0.6271, 0.4040, 0.9137, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413], [0.6236, 0.3965, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459], [0.6329, 0.4196, 0.9237, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.04951172370056156 step: 69 running loss: 0.0007175612130516168 Train Steps: 69/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.6099, 0.4030, 0.8638, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388], [0.6198, 0.4101, 0.8838, 0.5283, 0.3763, 0.5267, 0.5913, 0.5567], [0.6034, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084], [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124], [0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297], [0.6125, 0.4076, 0.8488, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6369, 0.4063, 0.9166, 0.4760, 0.3783, 0.4141, 0.6935, 0.5250], [0.6562, 0.3967, 0.9160, 0.4163, 0.3726, 0.3441, 0.6424, 0.5450], [0.6723, 0.4198, 0.8920, 0.5082, 0.5132, 0.4629, 0.5744, 0.5426], [0.6825, 0.4162, 0.9109, 0.5298, 0.3909, 0.5015, 0.6304, 0.5498], [0.6299, 0.3838, 0.7454, 0.2568, 0.3605, 0.3079, 0.6139, 0.5104], [0.6160, 0.3820, 0.8326, 0.2532, 0.4049, 0.2346, 0.6360, 0.5246], [0.6479, 0.4071, 0.9284, 0.4265, 0.4301, 0.2949, 0.6847, 0.5333], [0.6430, 0.3917, 0.8725, 0.3664, 0.3732, 0.3443, 0.5314, 0.5447]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.6098, 0.4030, 0.8637, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388], [0.6198, 0.4101, 0.8838, 0.5283, 0.3762, 0.5267, 0.5913, 0.5567], [0.6033, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084], [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124], [0.6364, 0.4165, 0.9087, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297], [0.6125, 0.4076, 0.8487, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.05016214041097555 step: 70 running loss: 0.0007166020058710793 Train Steps: 70/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383], [0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290], [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012], [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050], [0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954], [0.6102, 0.4020, 0.8638, 0.3717, 0.3625, 0.5017, 0.6038, 0.5500], [0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.2846, 0.1741, 0.7245, 0.1989, 0.4228, 0.1664, 0.5666, 0.5444], [0.7737, 0.4870, 0.8862, 0.4056, 0.3876, 0.4681, 0.6411, 0.6069], [0.6626, 0.4021, 0.8599, 0.2900, 0.3500, 0.3739, 0.6576, 0.5374], [0.6829, 0.4202, 0.8933, 0.5592, 0.3822, 0.4282, 0.6746, 0.5033], [0.6728, 0.4183, 0.8102, 0.3008, 0.3831, 0.2867, 0.6244, 0.5226], [0.7139, 0.4413, 0.8702, 0.4020, 0.4185, 0.2396, 0.5509, 0.5229], [0.6474, 0.4059, 0.8866, 0.3840, 0.3860, 0.4751, 0.6363, 0.5533], [0.7377, 0.4589, 0.8799, 0.5809, 0.4689, 0.4086, 0.5932, 0.6058]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.7268, 0.2333, 0.4125, 0.1933, 0.5113, 0.5383], [0.6200, 0.4112, 0.8863, 0.4100, 0.3638, 0.4917, 0.6087, 0.6050], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290], [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012], [0.6135, 0.3994, 0.7912, 0.3050, 0.3625, 0.3050, 0.5838, 0.5050], [0.6137, 0.4038, 0.8562, 0.4050, 0.3812, 0.2550, 0.5106, 0.4954], [0.6102, 0.4020, 0.8637, 0.3717, 0.3625, 0.5017, 0.6037, 0.5500], [0.6248, 0.4185, 0.8500, 0.5767, 0.4462, 0.4550, 0.5612, 0.5917]]], device='cuda:0') loss_train_step before backward: tensor(0.0032, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0032, device='cuda:0', grad_fn=) loss_train: 0.053396602292195894 step: 71 running loss: 0.0007520648210168436 Train Steps: 71/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533], [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783], [0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748], [0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748], [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378], [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183], [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.1801, 0.1107, 0.7775, 0.2522, 0.3755, 0.2760, 0.5482, 0.5658], [0.6555, 0.4322, 0.8262, 0.3429, 0.3661, 0.2883, 0.5107, 0.5873], [0.6422, 0.4133, 0.8809, 0.3643, 0.3938, 0.2512, 0.6069, 0.5259], [0.6714, 0.4308, 0.9102, 0.4524, 0.4319, 0.2764, 0.7034, 0.5672], [0.6379, 0.4200, 0.8942, 0.5023, 0.4042, 0.3349, 0.7128, 0.5673], [0.6437, 0.4143, 0.8999, 0.4609, 0.3997, 0.3374, 0.7002, 0.5466], [0.6936, 0.4358, 0.8862, 0.4660, 0.4581, 0.5420, 0.6077, 0.5310], [0.6167, 0.4021, 0.8470, 0.3338, 0.3548, 0.4915, 0.5990, 0.5446]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533], [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783], [0.6277, 0.4117, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6329, 0.4196, 0.9237, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748], [0.6336, 0.4191, 0.8938, 0.5167, 0.3938, 0.3517, 0.7343, 0.5748], [0.6307, 0.4029, 0.8988, 0.4817, 0.3938, 0.3500, 0.7311, 0.5378], [0.6251, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6087, 0.5183], [0.6168, 0.4029, 0.8523, 0.3417, 0.3587, 0.5000, 0.6125, 0.5400]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.05442776322888676 step: 72 running loss: 0.0007559411559567605 Train Steps: 72/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272], [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633], [0.6304, 0.4024, 0.8925, 0.4800, 0.3937, 0.4817, 0.7485, 0.5297], [0.6157, 0.4102, 0.8513, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890], [0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517], [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050], [0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6175, 0.4065, 0.8519, 0.3611, 0.3844, 0.2618, 0.5318, 0.5447], [0.6762, 0.4368, 0.8453, 0.2408, 0.4535, 0.2396, 0.6948, 0.5665], [0.1374, 0.0842, 0.7511, 0.2508, 0.3693, 0.2625, 0.5106, 0.5620], [0.6211, 0.3957, 0.8926, 0.4968, 0.3888, 0.4682, 0.6999, 0.5258], [0.6164, 0.4151, 0.8564, 0.3892, 0.3551, 0.3569, 0.5156, 0.5850], [0.6030, 0.4068, 0.8657, 0.4823, 0.4596, 0.5002, 0.5685, 0.5478], [0.6145, 0.3968, 0.7989, 0.3191, 0.3619, 0.3048, 0.5876, 0.5153], [0.6064, 0.3935, 0.8939, 0.3244, 0.4601, 0.1780, 0.6137, 0.5080]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6186, 0.4080, 0.8625, 0.3483, 0.3787, 0.2650, 0.5320, 0.5272], [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.0000, 0.0000, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633], [0.6304, 0.4024, 0.8925, 0.4800, 0.3938, 0.4817, 0.7485, 0.5297], [0.6157, 0.4102, 0.8512, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890], [0.6175, 0.3957, 0.8700, 0.4817, 0.4663, 0.5133, 0.5800, 0.5517], [0.6135, 0.3994, 0.7912, 0.3050, 0.3625, 0.3050, 0.5838, 0.5050], [0.6259, 0.4156, 0.8813, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.05502122991310898 step: 73 running loss: 0.0007537154782617668 Train Steps: 73/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378], [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433], [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850], [0.6339, 0.4102, 0.8588, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089], [0.6201, 0.4151, 0.8588, 0.5467, 0.3700, 0.3950, 0.5637, 0.5933], [ nan, nan, 0.9088, 0.3783, 0.4562, 0.2617, 0.6741, 0.5575], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5918, 0.3869, 0.8965, 0.4814, 0.3500, 0.4986, 0.6200, 0.5045], [0.6109, 0.3999, 0.8538, 0.5933, 0.3659, 0.4532, 0.5646, 0.5465], [0.6070, 0.4145, 0.8305, 0.3525, 0.3421, 0.3375, 0.5324, 0.5634], [0.6302, 0.4250, 0.8856, 0.4422, 0.4552, 0.5652, 0.6061, 0.5706], [0.6146, 0.4082, 0.8555, 0.3182, 0.4298, 0.2219, 0.6388, 0.5193], [0.6136, 0.4235, 0.8473, 0.5513, 0.3558, 0.4063, 0.5634, 0.5982], [0.2057, 0.1608, 0.9028, 0.3820, 0.4332, 0.2708, 0.6609, 0.5754], [0.6079, 0.4167, 0.8930, 0.3856, 0.4246, 0.2257, 0.6196, 0.5225]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947], [0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378], [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5412, 0.5433], [0.6184, 0.4079, 0.8838, 0.4617, 0.4837, 0.5650, 0.6175, 0.5850], [0.6339, 0.4102, 0.8587, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089], [0.6202, 0.4151, 0.8587, 0.5467, 0.3700, 0.3950, 0.5638, 0.5933], [0.0000, 0.0000, 0.9087, 0.3783, 0.4563, 0.2617, 0.6741, 0.5575], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.05629452761786524 step: 74 running loss: 0.0007607368597008817 Train Steps: 74/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312], [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333], [ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767], [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.6177, 0.4086, 0.8738, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700], [0.6201, 0.4151, 0.8588, 0.5467, 0.3700, 0.3950, 0.5637, 0.5933], [0.6199, 0.4093, 0.7913, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.5973, 0.3953, 0.8694, 0.4266, 0.3525, 0.4234, 0.5897, 0.5375], [ 0.5847, 0.4172, 0.8889, 0.4558, 0.3766, 0.4846, 0.6057, 0.5288], [-0.0367, -0.0179, 0.7660, 0.3021, 0.4063, 0.2443, 0.5496, 0.5709], [ 0.5803, 0.4091, 0.8079, 0.2855, 0.4667, 0.1643, 0.5777, 0.5374], [ 0.6129, 0.4125, 0.8918, 0.4804, 0.4105, 0.5255, 0.6131, 0.5526], [ 0.6146, 0.4240, 0.8785, 0.4078, 0.3761, 0.5326, 0.6223, 0.5452], [ 0.6074, 0.4237, 0.8493, 0.5601, 0.3660, 0.3887, 0.5702, 0.5769], [ 0.5856, 0.4142, 0.7920, 0.2483, 0.4149, 0.2620, 0.5998, 0.5645]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312], [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6137, 0.5333], [0.0000, 0.0000, 0.7613, 0.3250, 0.4038, 0.2533, 0.5437, 0.5767], [0.6229, 0.4107, 0.8138, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6241, 0.4142, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.6177, 0.4085, 0.8737, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700], [0.6202, 0.4151, 0.8587, 0.5467, 0.3700, 0.3950, 0.5638, 0.5933], [0.6198, 0.4093, 0.7912, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700]]], device='cuda:0') loss_train_step before backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train: 0.05648698155710008 step: 75 running loss: 0.0007531597540946678 Train Steps: 75/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6208, 0.4082, 0.8538, 0.3067, 0.3588, 0.3717, 0.6112, 0.5517], [0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413], [0.6284, 0.4127, 0.8538, 0.5867, 0.4363, 0.5083, 0.6038, 0.5433], [0.6200, 0.3993, 0.8519, 0.4923, 0.3962, 0.4717, 0.6013, 0.5433], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167], [ nan, nan, 0.8488, 0.2300, 0.5563, 0.2100, 0.7390, 0.5679], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.5860, 0.3923, 0.8425, 0.3362, 0.3515, 0.3928, 0.5738, 0.5406], [ 0.6200, 0.4110, 0.8593, 0.2493, 0.5200, 0.1874, 0.6884, 0.5160], [ 0.5664, 0.3781, 0.8495, 0.6168, 0.4182, 0.5069, 0.5700, 0.5070], [ 0.5546, 0.3733, 0.8446, 0.5169, 0.3833, 0.4792, 0.5760, 0.5295], [ 0.6115, 0.4124, 0.8515, 0.4124, 0.3485, 0.3957, 0.5713, 0.5496], [ 0.5548, 0.3851, 0.7570, 0.3027, 0.3813, 0.2792, 0.5600, 0.6106], [-0.0340, -0.0110, 0.8601, 0.2618, 0.5085, 0.2280, 0.7064, 0.5195], [ 0.5388, 0.3658, 0.8529, 0.5009, 0.4163, 0.4864, 0.5184, 0.5185]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6208, 0.4082, 0.8537, 0.3067, 0.3587, 0.3717, 0.6112, 0.5517], [0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413], [0.6284, 0.4127, 0.8537, 0.5867, 0.4363, 0.5083, 0.6037, 0.5433], [0.6200, 0.3993, 0.8519, 0.4923, 0.3963, 0.4717, 0.6012, 0.5433], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167], [0.0000, 0.0000, 0.8487, 0.2300, 0.5562, 0.2100, 0.7390, 0.5679], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.05719406365824398 step: 76 running loss: 0.0007525534691874208 Train Steps: 76/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6361, 0.4076, 0.8862, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297], [0.6185, 0.4042, 0.7700, 0.2250, 0.4062, 0.2117, 0.5763, 0.5150], [0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693], [0.6250, 0.4106, 0.8700, 0.3717, 0.3588, 0.4967, 0.6038, 0.5167], [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350], [0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517], [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519], [0.6120, 0.4014, 0.6863, 0.2817, 0.3700, 0.2783, 0.5513, 0.5667]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5419, 0.3606, 0.9036, 0.5570, 0.3769, 0.4674, 0.6729, 0.5537], [0.5144, 0.3454, 0.7958, 0.2320, 0.4242, 0.1981, 0.5915, 0.5206], [0.5326, 0.3526, 0.8689, 0.5011, 0.3797, 0.4656, 0.5266, 0.5109], [0.5822, 0.3942, 0.8984, 0.3880, 0.3711, 0.5033, 0.6150, 0.5528], [0.5335, 0.3495, 0.8842, 0.5127, 0.3977, 0.4380, 0.5443, 0.5262], [0.5281, 0.3587, 0.8922, 0.4874, 0.4806, 0.4987, 0.5777, 0.5571], [0.5011, 0.3317, 0.8862, 0.5422, 0.4830, 0.5021, 0.5015, 0.5560], [0.5466, 0.3579, 0.7138, 0.2882, 0.3758, 0.2871, 0.5510, 0.5647]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6361, 0.4076, 0.8863, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297], [0.6184, 0.4042, 0.7700, 0.2250, 0.4062, 0.2117, 0.5763, 0.5150], [0.6038, 0.3946, 0.8413, 0.4883, 0.3562, 0.4550, 0.5266, 0.4693], [0.6250, 0.4105, 0.8700, 0.3717, 0.3587, 0.4967, 0.6037, 0.5167], [0.6143, 0.3982, 0.8650, 0.4883, 0.3913, 0.4317, 0.5315, 0.5350], [0.6175, 0.3957, 0.8700, 0.4817, 0.4663, 0.5133, 0.5800, 0.5517], [0.6100, 0.4016, 0.8600, 0.5067, 0.4613, 0.5233, 0.5086, 0.5519], [0.6120, 0.4013, 0.6862, 0.2817, 0.3700, 0.2783, 0.5512, 0.5667]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.05860871459299233 step: 77 running loss: 0.000761152137571329 Train Steps: 77/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726], [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131], [0.6200, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913], [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550], [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864], [0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6068, 0.3924, 0.7669, 0.2712, 0.3886, 0.2821, 0.5187, 0.4813], [0.5069, 0.3377, 0.7737, 0.2816, 0.4305, 0.3110, 0.6087, 0.6188], [0.5604, 0.3725, 0.8664, 0.3686, 0.3614, 0.3860, 0.5699, 0.5144], [0.4626, 0.2921, 0.8657, 0.5072, 0.4164, 0.4981, 0.5736, 0.4907], [0.5271, 0.3523, 0.8900, 0.4604, 0.5042, 0.4929, 0.5649, 0.5500], [0.5475, 0.3650, 0.8750, 0.4847, 0.3954, 0.4822, 0.7045, 0.5844], [0.5740, 0.3863, 0.8746, 0.4546, 0.3666, 0.4736, 0.5707, 0.5636], [0.4890, 0.3238, 0.8717, 0.5288, 0.3855, 0.3552, 0.5640, 0.5294]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726], [0.6196, 0.4094, 0.7563, 0.2817, 0.3938, 0.3183, 0.6012, 0.6183], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131], [0.6199, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913], [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550], [0.6314, 0.4107, 0.8750, 0.5100, 0.3787, 0.4900, 0.7121, 0.5864], [0.6197, 0.4091, 0.8800, 0.4783, 0.3537, 0.4767, 0.5950, 0.5550], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395]]], device='cuda:0') loss_train_step before backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train: 0.060351279433234595 step: 78 running loss: 0.0007737343517081358 Train Steps: 78/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8363, 0.3317, 0.3563, 0.3367, 0.5329, 0.5142], [0.6224, 0.4179, 0.8700, 0.5683, 0.4037, 0.4683, 0.5650, 0.5600], [0.6125, 0.4035, 0.7825, 0.3100, 0.3463, 0.4900, 0.5832, 0.5637], [0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467], [0.6233, 0.4091, 0.8100, 0.2950, 0.3563, 0.3883, 0.6013, 0.5200], [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552], [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[-0.0143, -0.0024, 0.8038, 0.3344, 0.3643, 0.3234, 0.5648, 0.5117], [ 0.6862, 0.4534, 0.8565, 0.5897, 0.4113, 0.4675, 0.5926, 0.5485], [ 0.6735, 0.4393, 0.7570, 0.3208, 0.3560, 0.4950, 0.5838, 0.5183], [ 0.6936, 0.4571, 0.8300, 0.4819, 0.4362, 0.5289, 0.6201, 0.5342], [ 0.6809, 0.4478, 0.7915, 0.3032, 0.3672, 0.4096, 0.6177, 0.5297], [ 0.6235, 0.4041, 0.8658, 0.5046, 0.3969, 0.4643, 0.5237, 0.5457], [-0.0387, -0.0200, 0.8594, 0.2558, 0.5169, 0.2219, 0.7540, 0.5521], [ 0.6800, 0.4418, 0.8547, 0.4494, 0.4336, 0.5273, 0.6129, 0.5608]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8363, 0.3317, 0.3562, 0.3367, 0.5329, 0.5142], [0.6224, 0.4179, 0.8700, 0.5683, 0.4038, 0.4683, 0.5650, 0.5600], [0.6125, 0.4035, 0.7825, 0.3100, 0.3462, 0.4900, 0.5832, 0.5637], [0.6199, 0.4059, 0.8700, 0.4900, 0.4162, 0.5000, 0.6162, 0.5467], [0.6233, 0.4091, 0.8100, 0.2950, 0.3562, 0.3883, 0.6012, 0.5200], [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.0000, 0.0000, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552], [0.6189, 0.4049, 0.8888, 0.4417, 0.4212, 0.5200, 0.5987, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.06108348035195377 step: 79 running loss: 0.0007732086120500476 Train Steps: 79/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413], [0.6177, 0.4086, 0.8738, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700], [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400], [0.6200, 0.4070, 0.8938, 0.4183, 0.3538, 0.4567, 0.6175, 0.5400], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6188, 0.5400]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5224, 0.3419, 0.7550, 0.2004, 0.4276, 0.2089, 0.5920, 0.4913], [0.5525, 0.3680, 0.9134, 0.4232, 0.4112, 0.3087, 0.7105, 0.5515], [0.5902, 0.4045, 0.8558, 0.3962, 0.3823, 0.5594, 0.6005, 0.5678], [0.5608, 0.3880, 0.8351, 0.3319, 0.3604, 0.4914, 0.5836, 0.5657], [0.6617, 0.4348, 0.8837, 0.4245, 0.3588, 0.4808, 0.6075, 0.5278], [0.5215, 0.3512, 0.8832, 0.3702, 0.4036, 0.2688, 0.6154, 0.5268], [0.5126, 0.3345, 0.8717, 0.4826, 0.4417, 0.5923, 0.5912, 0.5300], [0.5494, 0.3744, 0.8171, 0.3323, 0.3662, 0.3569, 0.6079, 0.5482]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6202, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044], [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413], [0.6177, 0.4085, 0.8737, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700], [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400], [0.6200, 0.4070, 0.8938, 0.4183, 0.3537, 0.4567, 0.6175, 0.5400], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6187, 0.5283], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6187, 0.5400]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.06227331746777054 step: 80 running loss: 0.0007784164683471318 Train Steps: 80/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6185, 0.4042, 0.7700, 0.2250, 0.4062, 0.2117, 0.5763, 0.5150], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.6350, 0.4043, 0.8738, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950], [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133], [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413], [0.6286, 0.4055, 0.9000, 0.4717, 0.3763, 0.4683, 0.7018, 0.5494], [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5728, 0.3780, 0.7630, 0.2122, 0.4002, 0.2459, 0.5436, 0.5268], [0.5629, 0.3804, 0.8455, 0.2477, 0.4786, 0.1995, 0.5931, 0.5053], [0.5817, 0.3744, 0.8778, 0.3963, 0.3435, 0.3911, 0.5819, 0.5604], [0.6186, 0.3939, 0.8592, 0.5448, 0.3678, 0.5465, 0.6132, 0.5131], [0.6066, 0.3892, 0.8570, 0.5147, 0.3561, 0.5063, 0.6153, 0.5325], [0.5814, 0.3799, 0.9132, 0.3978, 0.3947, 0.3442, 0.6963, 0.5596], [0.5945, 0.3818, 0.8924, 0.4424, 0.3485, 0.5178, 0.6996, 0.5596], [0.5771, 0.3658, 0.8505, 0.4618, 0.4108, 0.5537, 0.5980, 0.5515]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6184, 0.4042, 0.7700, 0.2250, 0.4062, 0.2117, 0.5763, 0.5150], [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.6350, 0.4043, 0.8737, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950], [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133], [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413], [0.6286, 0.4055, 0.9000, 0.4717, 0.3762, 0.4683, 0.7018, 0.5494], [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.06305107554362621 step: 81 running loss: 0.000778408340044768 Train Steps: 81/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333], [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6117, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142], [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5925, 0.3868, 0.7185, 0.2207, 0.4239, 0.1918, 0.5743, 0.5804], [0.5980, 0.3963, 0.9158, 0.4424, 0.3656, 0.5126, 0.6424, 0.5339], [0.6732, 0.4341, 0.7695, 0.2681, 0.3920, 0.3232, 0.6471, 0.6118], [0.4580, 0.2868, 0.7231, 0.2085, 0.4095, 0.2476, 0.5547, 0.5508], [0.5601, 0.3514, 0.6649, 0.2035, 0.3818, 0.2585, 0.5571, 0.5226], [0.5739, 0.3568, 0.7445, 0.1915, 0.4272, 0.2050, 0.6112, 0.5174], [0.5811, 0.3766, 0.8925, 0.4364, 0.3828, 0.4687, 0.5463, 0.5253], [0.5821, 0.3855, 0.8140, 0.2603, 0.3948, 0.2667, 0.6166, 0.5354]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6154, 0.4112, 0.7038, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6137, 0.5333], [0.6196, 0.4094, 0.7563, 0.2817, 0.3938, 0.3183, 0.6012, 0.6183], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6116, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.6185, 0.4067, 0.8838, 0.4450, 0.4038, 0.4733, 0.5213, 0.5142], [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.06413436621369328 step: 82 running loss: 0.000782126417240162 Train Steps: 82/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333], [0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6262, 0.5367], [0.6216, 0.4167, 0.8588, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6127, 0.4118, 0.8650, 0.5083, 0.4088, 0.5367, 0.5300, 0.5456]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6515, 0.4266, 0.8918, 0.4172, 0.3766, 0.4620, 0.6430, 0.5268], [0.6456, 0.4085, 0.8798, 0.4478, 0.4039, 0.5626, 0.6127, 0.5239], [0.7003, 0.4583, 0.8623, 0.5334, 0.3712, 0.4893, 0.6281, 0.5466], [0.6903, 0.4388, 0.7631, 0.3358, 0.3392, 0.3826, 0.5472, 0.5373], [0.6365, 0.4038, 0.8620, 0.3614, 0.3483, 0.3893, 0.6020, 0.5475], [0.6166, 0.4032, 0.8847, 0.4609, 0.4272, 0.5328, 0.6282, 0.5048], [0.7030, 0.4594, 0.7801, 0.2450, 0.4328, 0.1511, 0.6083, 0.5073], [0.6425, 0.4281, 0.8641, 0.4693, 0.4036, 0.5050, 0.5713, 0.5506]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6137, 0.5333], [0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6263, 0.5367], [0.6216, 0.4167, 0.8587, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.6155, 0.4124, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6127, 0.4118, 0.8650, 0.5083, 0.4087, 0.5367, 0.5300, 0.5456]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.06510589212120976 step: 83 running loss: 0.0007844083388097562 Train Steps: 83/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667], [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393], [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960], [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.1093, 0.0623, 0.7614, 0.2376, 0.4039, 0.2795, 0.6019, 0.5619], [0.6374, 0.4181, 0.8749, 0.4911, 0.3949, 0.4465, 0.5443, 0.5621], [0.7407, 0.4638, 0.8816, 0.4757, 0.3898, 0.5048, 0.6563, 0.4767], [0.7346, 0.4880, 0.8432, 0.5665, 0.4289, 0.4709, 0.6206, 0.5230], [0.7766, 0.4909, 0.8456, 0.2801, 0.4178, 0.2757, 0.6832, 0.5210], [0.6897, 0.4571, 0.8626, 0.4530, 0.4290, 0.4688, 0.5380, 0.4883], [0.6902, 0.4332, 0.8677, 0.4743, 0.3426, 0.3980, 0.5872, 0.5126], [0.7270, 0.4772, 0.8036, 0.2919, 0.3839, 0.2818, 0.6136, 0.5013]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.7525, 0.2291, 0.3837, 0.3017, 0.6050, 0.5667], [0.6163, 0.4001, 0.8788, 0.5033, 0.4013, 0.4633, 0.5337, 0.5767], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6229, 0.4066, 0.8512, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6286, 0.4086, 0.8408, 0.2801, 0.4162, 0.2800, 0.6725, 0.5393], [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960], [0.6236, 0.3965, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6201, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.06704556952172425 step: 84 running loss: 0.0007981615419252886 Train Steps: 84/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467], [0.6263, 0.4030, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771], [0.6267, 0.4080, 0.8438, 0.2633, 0.4763, 0.1800, 0.6259, 0.5240], [ nan, nan, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533], [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850], [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183], [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7320, 0.4737, 0.8611, 0.5556, 0.3902, 0.4397, 0.5956, 0.5358], [0.7077, 0.4610, 0.8938, 0.4741, 0.3591, 0.5033, 0.6193, 0.4845], [0.7522, 0.4808, 0.8343, 0.2710, 0.4589, 0.1764, 0.6038, 0.5168], [0.0814, 0.0406, 0.8817, 0.3188, 0.4925, 0.2216, 0.6584, 0.5543], [0.6683, 0.4413, 0.7192, 0.2160, 0.3753, 0.2583, 0.5710, 0.5413], [0.7060, 0.4613, 0.8823, 0.4435, 0.4619, 0.5227, 0.6069, 0.5665], [0.6551, 0.4286, 0.8835, 0.4538, 0.3918, 0.5613, 0.6043, 0.5227], [0.7188, 0.4664, 0.8615, 0.5288, 0.4406, 0.4852, 0.5693, 0.5099]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6273, 0.4143, 0.8750, 0.5700, 0.3988, 0.4717, 0.6012, 0.5467], [0.6263, 0.4029, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771], [0.6267, 0.4080, 0.8438, 0.2633, 0.4762, 0.1800, 0.6259, 0.5240], [0.0000, 0.0000, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6114, 0.4018, 0.7212, 0.1967, 0.3762, 0.2700, 0.5875, 0.5533], [0.6184, 0.4079, 0.8838, 0.4617, 0.4837, 0.5650, 0.6175, 0.5850], [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183], [0.6202, 0.4053, 0.8637, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133]]], device='cuda:0') loss_train_step before backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0015, device='cuda:0', grad_fn=) loss_train: 0.06853765701816883 step: 85 running loss: 0.0008063253766843392 Train Steps: 85/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6202, 0.4064, 0.7879, 0.2179, 0.4567, 0.1725, 0.5955, 0.5478], [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320], [0.6200, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913], [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519], [0.6173, 0.4114, 0.7325, 0.2500, 0.4213, 0.1917, 0.5338, 0.5700], [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461], [0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6508, 0.4336, 0.8535, 0.2768, 0.4638, 0.2824, 0.6922, 0.5507], [0.6845, 0.4468, 0.7882, 0.2625, 0.4845, 0.1763, 0.5718, 0.5383], [0.6801, 0.4236, 0.9062, 0.4683, 0.3671, 0.4290, 0.6874, 0.5161], [0.6959, 0.4352, 0.8825, 0.5497, 0.4140, 0.4984, 0.5772, 0.4890], [0.6354, 0.4205, 0.8702, 0.5328, 0.4810, 0.5111, 0.4839, 0.5492], [0.5374, 0.3567, 0.7500, 0.2449, 0.4344, 0.2038, 0.5264, 0.5679], [0.6251, 0.4128, 0.8308, 0.3663, 0.3539, 0.4017, 0.5425, 0.5324], [0.6461, 0.3940, 0.8751, 0.5405, 0.4116, 0.4967, 0.6169, 0.5035]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6202, 0.4064, 0.7879, 0.2179, 0.4567, 0.1725, 0.5955, 0.5478], [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320], [0.6199, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913], [0.6100, 0.4016, 0.8600, 0.5067, 0.4613, 0.5233, 0.5086, 0.5519], [0.6173, 0.4114, 0.7325, 0.2500, 0.4212, 0.1917, 0.5337, 0.5700], [0.6058, 0.3978, 0.8288, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461], [0.6308, 0.3990, 0.8687, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.06918519355531316 step: 86 running loss: 0.0008044789948292229 Train Steps: 86/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6353, 0.4128, 0.9138, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6336, 0.4154, 0.8900, 0.2767, 0.4988, 0.2867, 0.7422, 0.5540], [0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6263, 0.4030, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771], [ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6749, 0.4588, 0.8992, 0.3963, 0.4897, 0.3346, 0.7159, 0.6081], [ 0.6886, 0.4441, 0.8908, 0.4452, 0.3752, 0.3636, 0.5976, 0.5500], [ 0.6656, 0.4560, 0.8947, 0.5214, 0.4696, 0.5912, 0.5820, 0.5216], [ 0.6618, 0.4416, 0.8677, 0.3127, 0.5169, 0.3047, 0.7389, 0.5555], [ 0.6127, 0.4198, 0.8645, 0.4757, 0.4118, 0.4427, 0.4806, 0.5549], [ 0.6725, 0.4273, 0.7533, 0.2409, 0.4704, 0.1748, 0.5897, 0.4989], [ 0.6451, 0.4203, 0.8935, 0.5194, 0.3952, 0.5392, 0.6293, 0.4988], [-0.0591, -0.0411, 0.7639, 0.3055, 0.4165, 0.2820, 0.4991, 0.5784]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6353, 0.4128, 0.9137, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.6155, 0.4124, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6336, 0.4154, 0.8900, 0.2767, 0.4988, 0.2867, 0.7422, 0.5540], [0.6127, 0.4084, 0.8700, 0.4467, 0.3988, 0.4317, 0.5013, 0.5471], [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6263, 0.4029, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771], [0.0000, 0.0000, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.06988749319862109 step: 87 running loss: 0.0008033045195243803 Train Steps: 87/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6274, 0.4003, 0.8638, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661], [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528], [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373], [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767], [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6093, 0.3767, 0.8861, 0.6041, 0.4109, 0.4834, 0.6146, 0.5040], [0.5823, 0.3717, 0.9175, 0.5160, 0.4458, 0.5912, 0.6775, 0.5757], [0.6102, 0.4081, 0.7195, 0.2583, 0.4323, 0.2579, 0.5604, 0.5746], [0.5733, 0.3835, 0.9435, 0.3738, 0.4507, 0.3760, 0.7062, 0.5696], [0.6482, 0.4254, 0.8759, 0.5979, 0.4299, 0.4589, 0.5743, 0.5528], [0.6330, 0.4260, 0.9245, 0.4645, 0.4103, 0.3689, 0.5681, 0.5909], [0.6020, 0.3900, 0.8795, 0.5706, 0.4186, 0.5017, 0.7201, 0.5703], [0.5623, 0.3640, 0.8323, 0.2971, 0.4104, 0.2730, 0.5071, 0.5186]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6274, 0.4003, 0.8637, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661], [0.6293, 0.4024, 0.8750, 0.5000, 0.4013, 0.5733, 0.7121, 0.5633], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.6332, 0.4165, 0.9100, 0.3350, 0.4187, 0.3683, 0.7438, 0.5528], [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373], [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767], [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.0706424161180621 step: 88 running loss: 0.0008027547286143421 Train Steps: 88/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118], [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621], [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667], [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617], [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750], [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064], [0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7329, 0.4968, 0.7483, 0.3786, 0.4358, 0.2171, 0.5631, 0.5937], [0.0535, 0.0381, 0.8644, 0.2943, 0.5412, 0.2195, 0.6699, 0.5556], [0.1255, 0.0737, 0.7721, 0.2515, 0.4036, 0.3109, 0.5987, 0.5628], [0.7190, 0.4711, 0.9182, 0.5764, 0.3963, 0.4302, 0.5831, 0.5388], [0.6366, 0.4122, 0.7466, 0.2615, 0.3732, 0.3417, 0.6117, 0.5708], [0.6869, 0.4533, 0.7324, 0.2783, 0.3922, 0.2976, 0.6281, 0.5651], [0.6615, 0.4455, 0.8908, 0.5147, 0.4825, 0.5266, 0.5687, 0.4876], [0.6953, 0.4555, 0.9070, 0.4323, 0.3842, 0.5011, 0.6273, 0.5798]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118], [0.0000, 0.0000, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621], [0.0000, 0.0000, 0.7525, 0.2291, 0.3837, 0.3017, 0.6050, 0.5667], [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617], [0.6132, 0.4066, 0.7259, 0.2402, 0.3587, 0.3300, 0.6000, 0.5600], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750], [0.6102, 0.4005, 0.8687, 0.5100, 0.4812, 0.5400, 0.5404, 0.5064], [0.6200, 0.4112, 0.8863, 0.4100, 0.3638, 0.4917, 0.6087, 0.6050]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.07201781774347182 step: 89 running loss: 0.0008091889634097958 Train Steps: 89/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [ nan, nan, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737], [0.6131, 0.4037, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680], [0.6034, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084], [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882], [0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471], [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6060, 0.3917, 0.6771, 0.3172, 0.3757, 0.2812, 0.5570, 0.5812], [-0.0581, -0.0375, 0.8625, 0.2447, 0.5502, 0.2356, 0.7400, 0.5764], [ 0.5796, 0.3699, 0.6941, 0.2917, 0.3760, 0.2843, 0.5499, 0.5678], [ 0.5950, 0.3760, 0.7373, 0.2582, 0.3706, 0.3330, 0.5954, 0.5064], [ 0.6301, 0.3950, 0.9330, 0.4419, 0.4240, 0.3126, 0.7530, 0.5388], [ 0.6633, 0.4413, 0.8922, 0.4290, 0.3791, 0.3910, 0.5547, 0.5141], [ 0.5718, 0.3831, 0.8884, 0.4709, 0.4058, 0.4407, 0.5156, 0.5442], [ 0.5806, 0.3847, 0.9144, 0.5505, 0.3841, 0.4862, 0.5937, 0.5728]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.0000, 0.0000, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737], [0.6131, 0.4036, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680], [0.6033, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084], [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413], [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882], [0.6127, 0.4084, 0.8700, 0.4467, 0.3988, 0.4317, 0.5013, 0.5471], [0.6222, 0.4171, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.0725348553241929 step: 90 running loss: 0.0008059428369354767 Valid Steps: 10/10 Loss: nan 6.2924 -------------------------------------------------- Epoch: 8 Train Loss: 0.0008 Valid Loss: nan -------------------------------------------------- size of train loader is: 90 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267], [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621], [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400], [0.6092, 0.4001, 0.8638, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064], [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329], [0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656], [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.5809, 0.3892, 0.8763, 0.5048, 0.3556, 0.4534, 0.5933, 0.5725], [ 0.5585, 0.3613, 0.8566, 0.2598, 0.4254, 0.2554, 0.6410, 0.5312], [-0.0281, -0.0072, 0.8385, 0.2609, 0.5153, 0.1945, 0.6909, 0.5716], [ 0.6296, 0.4013, 0.8372, 0.4857, 0.4138, 0.5036, 0.6315, 0.5534], [ 0.5775, 0.3675, 0.8493, 0.4625, 0.4122, 0.5282, 0.5725, 0.5064], [ 0.6405, 0.4147, 0.8342, 0.3335, 0.3355, 0.3601, 0.6054, 0.5336], [ 0.5747, 0.3587, 0.8181, 0.4326, 0.3595, 0.4352, 0.5605, 0.5659], [ 0.5893, 0.3826, 0.8734, 0.4997, 0.3901, 0.4934, 0.6083, 0.5265]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6222, 0.4171, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6266, 0.4067, 0.8587, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267], [0.0000, 0.0000, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621], [0.6201, 0.3970, 0.8413, 0.4950, 0.4412, 0.5183, 0.6087, 0.5400], [0.6092, 0.4001, 0.8637, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064], [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329], [0.6091, 0.3997, 0.8314, 0.4334, 0.3787, 0.4550, 0.5213, 0.5656], [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.0004860513436142355 step: 1 running loss: 0.0004860513436142355 Train Steps: 1/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6127, 0.4115, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495], [0.6228, 0.4119, 0.7938, 0.2233, 0.4674, 0.1773, 0.6188, 0.5433], [0.6248, 0.4032, 0.7738, 0.1900, 0.4813, 0.1400, 0.5941, 0.4904], [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6199, 0.4071, 0.7600, 0.2117, 0.4037, 0.2767, 0.6138, 0.5550], [0.6339, 0.4118, 0.7988, 0.5800, 0.3912, 0.4583, 0.7343, 0.5760], [0.6115, 0.3998, 0.7063, 0.2383, 0.4037, 0.1950, 0.5320, 0.4993]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4761, 0.3098, 0.7138, 0.2520, 0.3472, 0.2872, 0.5280, 0.5661], [0.5404, 0.3577, 0.7892, 0.2008, 0.4592, 0.1701, 0.6062, 0.5569], [0.5778, 0.3732, 0.7722, 0.2057, 0.4636, 0.1201, 0.5917, 0.4919], [0.5289, 0.3369, 0.8543, 0.4854, 0.4132, 0.4918, 0.5679, 0.5458], [0.4861, 0.3146, 0.8276, 0.4051, 0.3561, 0.4725, 0.5445, 0.5689], [0.5397, 0.3539, 0.7541, 0.1826, 0.4015, 0.2716, 0.5978, 0.5660], [0.5419, 0.3511, 0.7921, 0.5433, 0.3567, 0.4557, 0.6956, 0.5577], [0.5144, 0.3353, 0.7089, 0.2007, 0.3883, 0.1958, 0.5168, 0.5069]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6127, 0.4114, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495], [0.6228, 0.4119, 0.7937, 0.2233, 0.4674, 0.1773, 0.6187, 0.5433], [0.6248, 0.4032, 0.7738, 0.1900, 0.4812, 0.1400, 0.5941, 0.4904], [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6199, 0.4071, 0.7600, 0.2117, 0.4038, 0.2767, 0.6137, 0.5550], [0.6339, 0.4118, 0.7987, 0.5800, 0.3913, 0.4583, 0.7343, 0.5760], [0.6115, 0.3998, 0.7063, 0.2383, 0.4038, 0.1950, 0.5320, 0.4993]]], device='cuda:0') loss_train_step before backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0020, device='cuda:0', grad_fn=) loss_train: 0.00245886217453517 step: 2 running loss: 0.001229431087267585 Train Steps: 2/90 Loss: 0.0012 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317], [0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799], [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700], [0.6201, 0.4017, 0.8871, 0.4621, 0.3517, 0.4675, 0.5999, 0.5106], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050], [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783], [0.6304, 0.4024, 0.8925, 0.4800, 0.3937, 0.4817, 0.7485, 0.5297]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.4858, 0.3245, 0.7570, 0.2105, 0.4402, 0.2475, 0.5965, 0.5613], [0.5709, 0.3834, 0.7998, 0.3829, 0.3771, 0.2972, 0.5274, 0.5805], [0.0874, 0.0654, 0.6780, 0.1980, 0.4254, 0.2052, 0.5490, 0.5937], [0.5820, 0.3664, 0.8662, 0.4281, 0.3647, 0.4916, 0.5814, 0.5280], [0.5865, 0.3777, 0.8605, 0.4721, 0.3877, 0.5724, 0.6234, 0.5273], [0.6304, 0.4000, 0.8916, 0.4334, 0.3792, 0.4218, 0.5842, 0.5127], [0.5845, 0.3798, 0.8789, 0.4629, 0.3794, 0.4046, 0.5080, 0.5814], [0.6068, 0.3876, 0.8614, 0.4630, 0.4006, 0.4813, 0.7141, 0.5286]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317], [0.6200, 0.4118, 0.8288, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799], [0.0000, 0.0000, 0.6413, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700], [0.6201, 0.4017, 0.8871, 0.4621, 0.3517, 0.4675, 0.5999, 0.5106], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050], [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783], [0.6304, 0.4024, 0.8925, 0.4800, 0.3938, 0.4817, 0.7485, 0.5297]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.0034466659708414227 step: 3 running loss: 0.0011488886569471408 Train Steps: 3/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117], [0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433], [ nan, nan, 0.8525, 0.2217, 0.5413, 0.2367, 0.7367, 0.5482], [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5710, 0.3786, 0.8432, 0.3172, 0.3542, 0.2946, 0.5317, 0.5189], [0.5667, 0.3979, 0.7542, 0.2634, 0.3320, 0.3626, 0.5838, 0.5578], [0.0320, 0.0410, 0.8086, 0.2045, 0.5000, 0.2239, 0.6892, 0.5512], [0.6161, 0.4025, 0.8187, 0.4901, 0.4073, 0.4758, 0.5308, 0.5540], [0.5866, 0.3672, 0.8515, 0.4788, 0.3924, 0.5120, 0.6089, 0.4946], [0.5726, 0.3708, 0.8269, 0.4869, 0.4068, 0.5184, 0.6734, 0.5451], [0.6295, 0.4060, 0.8382, 0.2308, 0.4869, 0.2442, 0.6998, 0.5377], [0.5761, 0.3858, 0.7478, 0.2148, 0.4222, 0.1888, 0.5715, 0.5357]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117], [0.6175, 0.4091, 0.7862, 0.2800, 0.3638, 0.3583, 0.6187, 0.5433], [0.0000, 0.0000, 0.8525, 0.2217, 0.5412, 0.2367, 0.7367, 0.5482], [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6293, 0.4097, 0.8800, 0.2517, 0.5263, 0.2600, 0.7430, 0.5378], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.004258358589140698 step: 4 running loss: 0.0010645896472851746 Train Steps: 4/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6252, 0.4158, 0.8988, 0.4083, 0.3788, 0.4783, 0.6225, 0.5633], [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082], [0.6200, 0.4070, 0.8938, 0.4183, 0.3538, 0.4567, 0.6175, 0.5400], [0.6346, 0.4092, 0.7712, 0.5917, 0.4037, 0.4767, 0.7343, 0.5725], [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261], [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.6361, 0.4076, 0.8862, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6286, 0.4358, 0.8769, 0.3787, 0.3642, 0.4453, 0.5894, 0.5461], [0.5941, 0.4048, 0.8591, 0.4995, 0.4092, 0.5502, 0.5732, 0.5008], [0.4940, 0.3278, 0.8859, 0.3949, 0.3396, 0.4385, 0.5975, 0.5139], [0.6161, 0.4102, 0.7761, 0.5220, 0.3656, 0.4566, 0.6900, 0.5708], [0.5571, 0.3729, 0.8705, 0.4630, 0.3933, 0.4678, 0.5163, 0.5224], [0.5966, 0.3890, 0.8608, 0.4651, 0.4166, 0.4937, 0.6023, 0.5137], [0.5856, 0.3920, 0.8522, 0.2076, 0.5237, 0.1951, 0.7348, 0.5356], [0.5518, 0.3843, 0.8716, 0.5228, 0.3479, 0.4377, 0.6359, 0.5273]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6252, 0.4158, 0.8988, 0.4083, 0.3787, 0.4783, 0.6225, 0.5633], [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082], [0.6200, 0.4070, 0.8938, 0.4183, 0.3537, 0.4567, 0.6175, 0.5400], [0.6346, 0.4092, 0.7713, 0.5917, 0.4038, 0.4767, 0.7343, 0.5725], [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261], [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.6361, 0.4076, 0.8863, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.005290690081892535 step: 5 running loss: 0.001058138016378507 Train Steps: 5/90 Loss: 0.0011 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6223, 0.4028, 0.8988, 0.4200, 0.3763, 0.5733, 0.6375, 0.5167], [ nan, nan, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6230, 0.4113, 0.7213, 0.1983, 0.4325, 0.2367, 0.6262, 0.5400], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6115, 0.3998, 0.7063, 0.2383, 0.4037, 0.1950, 0.5320, 0.4993], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5628, 0.3675, 0.9012, 0.4321, 0.4039, 0.5772, 0.6425, 0.5173], [0.1123, 0.0837, 0.6882, 0.2115, 0.4484, 0.1851, 0.5827, 0.5779], [0.6376, 0.4141, 0.8935, 0.5407, 0.4235, 0.5582, 0.7549, 0.5473], [0.6193, 0.4202, 0.7262, 0.2125, 0.4546, 0.2231, 0.6338, 0.5522], [0.6757, 0.4559, 0.7992, 0.3261, 0.3754, 0.2913, 0.5048, 0.5636], [0.6086, 0.4059, 0.8714, 0.4880, 0.3842, 0.5257, 0.6139, 0.5058], [0.6429, 0.4291, 0.7146, 0.2307, 0.4096, 0.1962, 0.5411, 0.4932], [0.5703, 0.3764, 0.8226, 0.3077, 0.3508, 0.3935, 0.6427, 0.5397]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6223, 0.4028, 0.8988, 0.4200, 0.3762, 0.5733, 0.6375, 0.5167], [0.0000, 0.0000, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6230, 0.4113, 0.7212, 0.1983, 0.4325, 0.2367, 0.6263, 0.5400], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6115, 0.3998, 0.7063, 0.2383, 0.4038, 0.1950, 0.5320, 0.4993], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.006010964134475216 step: 6 running loss: 0.0010018273557458695 Train Steps: 6/90 Loss: 0.0010 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.6133, 0.4066, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358], [0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656], [0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5988, 0.5667], [0.6246, 0.4028, 0.8738, 0.4867, 0.4088, 0.5667, 0.6362, 0.5200], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6033, 0.3968, 0.8633, 0.4681, 0.3756, 0.3958, 0.5996, 0.5504], [0.6511, 0.4366, 0.8907, 0.5408, 0.3869, 0.4642, 0.6706, 0.4991], [0.6189, 0.4157, 0.7080, 0.2688, 0.3831, 0.2361, 0.5732, 0.5136], [0.6097, 0.4049, 0.8417, 0.4520, 0.3871, 0.4563, 0.5520, 0.5556], [0.6564, 0.4354, 0.7826, 0.2350, 0.4024, 0.2438, 0.6246, 0.5608], [0.6491, 0.4279, 0.8879, 0.4915, 0.4421, 0.5618, 0.6730, 0.5224], [0.6068, 0.4018, 0.7448, 0.1839, 0.4090, 0.2412, 0.6519, 0.5410], [0.5664, 0.3761, 0.8622, 0.4227, 0.4177, 0.5919, 0.6115, 0.5155]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.6133, 0.4065, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358], [0.6091, 0.3997, 0.8314, 0.4334, 0.3787, 0.4550, 0.5213, 0.5656], [0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5987, 0.5667], [0.6246, 0.4028, 0.8737, 0.4867, 0.4087, 0.5667, 0.6363, 0.5200], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.006409419496776536 step: 7 running loss: 0.0009156313566823623 Train Steps: 7/90 Loss: 0.0009 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110], [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6138, 0.5883], [0.6284, 0.4127, 0.8538, 0.5867, 0.4363, 0.5083, 0.6038, 0.5433], [0.6167, 0.4048, 0.6831, 0.3639, 0.3763, 0.3017, 0.5700, 0.5883], [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6184, 0.3991, 0.8760, 0.3445, 0.3885, 0.2568, 0.6579, 0.4637], [0.6569, 0.4371, 0.8507, 0.2505, 0.4504, 0.2141, 0.6291, 0.4659], [0.6265, 0.4320, 0.8225, 0.5594, 0.4027, 0.4503, 0.5940, 0.5949], [0.6272, 0.4296, 0.8969, 0.4458, 0.3914, 0.5446, 0.5957, 0.4927], [0.6098, 0.4292, 0.8974, 0.4494, 0.4426, 0.5651, 0.6509, 0.5711], [0.5869, 0.3878, 0.8679, 0.5695, 0.4349, 0.5059, 0.6164, 0.5257], [0.5942, 0.3922, 0.7184, 0.3401, 0.3705, 0.3252, 0.5932, 0.5651], [0.6090, 0.4050, 0.8767, 0.4538, 0.3760, 0.4140, 0.6232, 0.5329]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6277, 0.4036, 0.8687, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6275, 0.4111, 0.8462, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6186, 0.4129, 0.8900, 0.4567, 0.3938, 0.5417, 0.5734, 0.5110], [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6137, 0.5883], [0.6284, 0.4127, 0.8537, 0.5867, 0.4363, 0.5083, 0.6037, 0.5433], [0.6167, 0.4048, 0.6831, 0.3639, 0.3762, 0.3017, 0.5700, 0.5883], [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.006727247644448653 step: 8 running loss: 0.0008409059555560816 Train Steps: 8/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6138, 0.5883], [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540], [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389], [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631], [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6183, 0.4030, 0.7435, 0.3145, 0.3552, 0.3331, 0.5759, 0.5422], [0.6342, 0.4077, 0.8421, 0.2665, 0.4672, 0.2127, 0.6182, 0.4593], [0.6046, 0.4084, 0.8912, 0.4746, 0.4667, 0.5795, 0.6430, 0.5635], [0.5526, 0.3554, 0.9060, 0.4546, 0.4237, 0.3305, 0.6833, 0.5396], [0.6539, 0.4260, 0.7077, 0.2388, 0.4306, 0.1791, 0.5493, 0.5390], [0.6156, 0.4022, 0.7569, 0.2314, 0.3848, 0.2848, 0.6123, 0.5242], [0.6136, 0.4001, 0.8375, 0.4258, 0.3736, 0.5034, 0.6086, 0.5390], [0.6474, 0.4279, 0.8735, 0.5599, 0.4168, 0.5013, 0.5937, 0.5653]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6275, 0.4111, 0.8462, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6137, 0.5883], [0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540], [0.6154, 0.4112, 0.7038, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389], [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631], [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.0071166820416692644 step: 9 running loss: 0.0007907424490743628 Train Steps: 9/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6064, 0.3953, 0.8738, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6250, 0.4110, 0.7238, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633], [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800], [0.6201, 0.4064, 0.8688, 0.5050, 0.4225, 0.5100, 0.6138, 0.5500], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373], [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6942, 0.4339, 0.8907, 0.4814, 0.3835, 0.5011, 0.5572, 0.5364], [0.6969, 0.4367, 0.8986, 0.5011, 0.3749, 0.4080, 0.6448, 0.5017], [0.5561, 0.3609, 0.7203, 0.2376, 0.4244, 0.2072, 0.5682, 0.5440], [0.6394, 0.4293, 0.7587, 0.3248, 0.3745, 0.2947, 0.5382, 0.5660], [0.6396, 0.4080, 0.8953, 0.5133, 0.4375, 0.5452, 0.6052, 0.5679], [0.5969, 0.4074, 0.7429, 0.3088, 0.4482, 0.2296, 0.5628, 0.5750], [0.6409, 0.4072, 0.8550, 0.6127, 0.4090, 0.4920, 0.5700, 0.5451], [0.6599, 0.4126, 0.7889, 0.1861, 0.4329, 0.2702, 0.6305, 0.5054]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6064, 0.3952, 0.8737, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6250, 0.4110, 0.7237, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633], [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5462, 0.5800], [0.6201, 0.4064, 0.8687, 0.5050, 0.4225, 0.5100, 0.6137, 0.5500], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373], [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.007774444384267554 step: 10 running loss: 0.0007774444384267554 Train Steps: 10/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323], [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999], [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750], [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960], [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6385, 0.4033, 0.8655, 0.4518, 0.4351, 0.5511, 0.6074, 0.5391], [0.6870, 0.4239, 0.8564, 0.5643, 0.3862, 0.4419, 0.5884, 0.5476], [0.5826, 0.3651, 0.7670, 0.2289, 0.4291, 0.1928, 0.6284, 0.5399], [0.6836, 0.4302, 0.8857, 0.5144, 0.3720, 0.4463, 0.5233, 0.5645], [0.6096, 0.3817, 0.8424, 0.4521, 0.3899, 0.4873, 0.5573, 0.5129], [0.6418, 0.4088, 0.7759, 0.3049, 0.4048, 0.2246, 0.5228, 0.5798], [0.6431, 0.4098, 0.8683, 0.4738, 0.4422, 0.4891, 0.5520, 0.5079], [0.6509, 0.4179, 0.7390, 0.3151, 0.3359, 0.3168, 0.5603, 0.5707]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323], [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6068, 0.3963, 0.8650, 0.4317, 0.4038, 0.5083, 0.5253, 0.4999], [0.6182, 0.4099, 0.7812, 0.3000, 0.3938, 0.2367, 0.5325, 0.5750], [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960], [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.008245526551036164 step: 11 running loss: 0.0007495933228214694 Train Steps: 11/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148], [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373], [0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343], [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667], [0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6789, 0.4254, 0.8802, 0.5339, 0.3951, 0.4785, 0.5689, 0.5207], [0.6904, 0.4412, 0.7863, 0.3883, 0.4789, 0.2596, 0.4998, 0.6308], [0.6794, 0.4258, 0.8299, 0.5935, 0.3994, 0.4723, 0.5408, 0.5457], [0.6904, 0.4317, 0.8464, 0.3005, 0.4299, 0.2478, 0.5632, 0.5375], [0.6913, 0.4495, 0.8708, 0.3862, 0.4187, 0.3195, 0.6650, 0.5311], [0.0830, 0.0339, 0.7411, 0.2390, 0.3734, 0.2989, 0.5673, 0.5606], [0.6843, 0.4467, 0.6902, 0.2188, 0.3999, 0.2534, 0.5375, 0.5657], [0.7003, 0.4519, 0.8651, 0.5073, 0.3672, 0.4434, 0.5544, 0.6115]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148], [0.6299, 0.4303, 0.7962, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373], [0.6250, 0.4131, 0.8687, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6275, 0.4003, 0.9100, 0.3783, 0.4387, 0.3133, 0.7058, 0.5343], [0.0000, 0.0000, 0.7525, 0.2291, 0.3837, 0.3017, 0.6050, 0.5667], [0.6187, 0.4103, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6186, 0.4060, 0.8750, 0.5050, 0.3537, 0.4367, 0.5813, 0.6083]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.009264791355235502 step: 12 running loss: 0.0007720659462696252 Train Steps: 12/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6188, 0.5400], [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533], [0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6040, 0.4002, 0.7338, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778], [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667], [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6438, 0.4170, 0.8441, 0.3354, 0.3635, 0.3598, 0.5761, 0.5718], [0.5178, 0.3205, 0.6909, 0.2413, 0.4453, 0.1691, 0.4824, 0.5813], [0.6444, 0.4104, 0.8591, 0.3085, 0.4397, 0.2368, 0.5545, 0.5576], [0.6467, 0.4118, 0.7465, 0.2338, 0.4028, 0.2046, 0.5104, 0.5042], [0.6424, 0.4062, 0.8770, 0.4743, 0.4497, 0.5087, 0.5798, 0.5563], [0.6998, 0.4514, 0.8549, 0.5256, 0.4192, 0.5506, 0.6577, 0.6106], [0.5969, 0.3950, 0.8450, 0.3362, 0.3615, 0.3658, 0.5247, 0.6014], [0.6412, 0.4060, 0.8985, 0.4726, 0.3843, 0.4225, 0.6617, 0.5409]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6187, 0.5400], [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533], [0.6250, 0.4131, 0.8687, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6040, 0.4002, 0.7337, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778], [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667], [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.010133300122106448 step: 13 running loss: 0.000779484624777419 Train Steps: 13/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901], [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400], [0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220], [0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6238, 0.5350], [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950], [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6662, 0.4234, 0.9031, 0.3694, 0.3580, 0.3724, 0.6157, 0.5550], [0.6267, 0.4100, 0.8555, 0.3439, 0.3576, 0.4699, 0.6097, 0.5852], [0.6212, 0.4056, 0.7296, 0.2370, 0.4127, 0.2023, 0.5632, 0.5981], [0.5978, 0.3986, 0.8732, 0.4130, 0.3557, 0.3778, 0.5035, 0.5433], [0.6166, 0.4000, 0.6922, 0.2149, 0.4322, 0.1552, 0.5333, 0.5524], [0.6210, 0.3970, 0.8759, 0.4106, 0.3713, 0.5390, 0.6254, 0.5757], [0.6666, 0.4193, 0.8550, 0.5351, 0.4326, 0.4919, 0.5749, 0.5434], [0.6432, 0.4139, 0.8767, 0.5163, 0.4053, 0.4915, 0.5790, 0.5741]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901], [0.6168, 0.4029, 0.8523, 0.3417, 0.3587, 0.5000, 0.6125, 0.5400], [0.6193, 0.4050, 0.7312, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6082, 0.4024, 0.8737, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220], [0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6237, 0.5350], [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950], [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.010566788056166843 step: 14 running loss: 0.0007547705754404888 Train Steps: 14/90 Loss: 0.0008 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6267, 0.4080, 0.8438, 0.2633, 0.4763, 0.1800, 0.6259, 0.5240], [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103], [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783], [0.6259, 0.4133, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6224, 0.4058, 0.8545, 0.4940, 0.4377, 0.5267, 0.6175, 0.5340], [0.5924, 0.3953, 0.8547, 0.4568, 0.4137, 0.4933, 0.5830, 0.5806], [0.5722, 0.3748, 0.8269, 0.2542, 0.4515, 0.2007, 0.6299, 0.5232], [0.5919, 0.3963, 0.8611, 0.2738, 0.3905, 0.2842, 0.5890, 0.5384], [0.5859, 0.3965, 0.8591, 0.4376, 0.3848, 0.4553, 0.5460, 0.5158], [0.6042, 0.4017, 0.8832, 0.4621, 0.3735, 0.4036, 0.5302, 0.5805], [0.5670, 0.3899, 0.8042, 0.2198, 0.4908, 0.1637, 0.6256, 0.5006], [0.6143, 0.4162, 0.8901, 0.4287, 0.3958, 0.3150, 0.6744, 0.5654]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6267, 0.4080, 0.8438, 0.2633, 0.4762, 0.1800, 0.6259, 0.5240], [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496], [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103], [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783], [0.6259, 0.4132, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.0108872129349038 step: 15 running loss: 0.0007258141956602534 Train Steps: 15/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [ nan, nan, 0.8888, 0.3100, 0.5262, 0.2817, 0.7145, 0.6003], [0.6259, 0.4133, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650], [0.6182, 0.3930, 0.8841, 0.3892, 0.3556, 0.4967, 0.6222, 0.5279], [0.6353, 0.4128, 0.9138, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6015, 0.4170, 0.7197, 0.2435, 0.4179, 0.2499, 0.5913, 0.6008], [0.6438, 0.4261, 0.8888, 0.4595, 0.4274, 0.5851, 0.6316, 0.5029], [0.0754, 0.0515, 0.8658, 0.2984, 0.4927, 0.2718, 0.7197, 0.5535], [0.6106, 0.4292, 0.8134, 0.2247, 0.4864, 0.1641, 0.6272, 0.4792], [0.6354, 0.4273, 0.8239, 0.2510, 0.4067, 0.1870, 0.6059, 0.4785], [0.6247, 0.4177, 0.8501, 0.4693, 0.4100, 0.5023, 0.5431, 0.5304], [0.6193, 0.3932, 0.8688, 0.3921, 0.3346, 0.4652, 0.6326, 0.5106], [0.6273, 0.4316, 0.8989, 0.3456, 0.4486, 0.3181, 0.7250, 0.5764]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6193, 0.4078, 0.7287, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.0000, 0.0000, 0.8888, 0.3100, 0.5263, 0.2817, 0.7145, 0.6003], [0.6259, 0.4132, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6277, 0.4056, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6098, 0.3991, 0.8637, 0.4717, 0.4263, 0.4967, 0.5213, 0.5650], [0.6182, 0.3930, 0.8841, 0.3892, 0.3556, 0.4967, 0.6222, 0.5279], [0.6353, 0.4128, 0.9137, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.011273512180196121 step: 16 running loss: 0.0007045945112622576 Train Steps: 16/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6262, 0.5367], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483], [0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064], [0.6167, 0.4048, 0.6831, 0.3639, 0.3763, 0.3017, 0.5700, 0.5883], [0.6179, 0.4082, 0.6688, 0.2667, 0.3588, 0.3317, 0.5750, 0.5783]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5943, 0.3980, 0.9217, 0.4521, 0.4279, 0.5748, 0.6315, 0.5061], [0.5945, 0.3896, 0.8918, 0.3888, 0.3856, 0.4484, 0.6115, 0.5012], [0.5628, 0.3809, 0.9031, 0.5169, 0.3854, 0.4670, 0.6718, 0.5036], [0.5827, 0.3990, 0.7920, 0.2506, 0.3870, 0.2685, 0.5760, 0.5261], [0.5897, 0.4110, 0.8753, 0.2631, 0.4937, 0.1924, 0.6866, 0.5137], [0.5579, 0.3951, 0.8914, 0.4911, 0.5017, 0.4916, 0.5551, 0.4862], [0.5632, 0.3852, 0.7218, 0.3323, 0.3831, 0.2869, 0.6009, 0.5554], [0.6234, 0.4210, 0.6913, 0.2689, 0.3568, 0.2974, 0.5792, 0.5452]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6263, 0.5367], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483], [0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6128, 0.4022, 0.8737, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064], [0.6167, 0.4048, 0.6831, 0.3639, 0.3762, 0.3017, 0.5700, 0.5883], [0.6179, 0.4082, 0.6687, 0.2667, 0.3587, 0.3317, 0.5750, 0.5783]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.011899679695488885 step: 17 running loss: 0.0006999811585581698 Train Steps: 17/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297], [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6362, 0.5200], [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6160, 0.4093, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290], [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5759, 0.3881, 0.8742, 0.5237, 0.3796, 0.4402, 0.6483, 0.5152], [0.5830, 0.3891, 0.8870, 0.3364, 0.3729, 0.5032, 0.6474, 0.4874], [0.6178, 0.4078, 0.8628, 0.5105, 0.4526, 0.5175, 0.6762, 0.5152], [0.6367, 0.4283, 0.8008, 0.3185, 0.3739, 0.3252, 0.6114, 0.4945], [0.5898, 0.3955, 0.8429, 0.4416, 0.3837, 0.4432, 0.5401, 0.5572], [0.5827, 0.3843, 0.8635, 0.4610, 0.4559, 0.4643, 0.5379, 0.5216], [0.6468, 0.4418, 0.7687, 0.1860, 0.4526, 0.1878, 0.6652, 0.5080], [0.6178, 0.4185, 0.8519, 0.5210, 0.3906, 0.4284, 0.6882, 0.5288]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297], [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6363, 0.5200], [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6160, 0.4092, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290], [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.012323376664426178 step: 18 running loss: 0.0006846320369125655 Train Steps: 18/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913], [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267], [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100], [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290], [ nan, nan, 0.8488, 0.2300, 0.5563, 0.2100, 0.7390, 0.5679], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5797, 0.3699, 0.8009, 0.2417, 0.4690, 0.1885, 0.6042, 0.4845], [0.5711, 0.3744, 0.8285, 0.5885, 0.4178, 0.4771, 0.5702, 0.5147], [0.6353, 0.3908, 0.8537, 0.5339, 0.4054, 0.5166, 0.6063, 0.4817], [0.6587, 0.4327, 0.7637, 0.2214, 0.4382, 0.2271, 0.6492, 0.5154], [0.0846, 0.0291, 0.8546, 0.2416, 0.5450, 0.2666, 0.7472, 0.5351], [0.6410, 0.4245, 0.7340, 0.1888, 0.4015, 0.2636, 0.6165, 0.5394], [0.6169, 0.4116, 0.8886, 0.3509, 0.3904, 0.3113, 0.5855, 0.5371], [0.6510, 0.4297, 0.9147, 0.4437, 0.4104, 0.3270, 0.6731, 0.5441]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6286, 0.4078, 0.8062, 0.2267, 0.4787, 0.1533, 0.5953, 0.4913], [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267], [0.6275, 0.4024, 0.8500, 0.5383, 0.3913, 0.4883, 0.6288, 0.5100], [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290], [0.0000, 0.0000, 0.8487, 0.2300, 0.5562, 0.2100, 0.7390, 0.5679], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478], [0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.012789542262908071 step: 19 running loss: 0.0006731338033109511 Train Steps: 19/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123], [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702], [0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267], [ nan, nan, 0.8463, 0.2550, 0.5850, 0.2133, 0.7129, 0.6072], [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667], [0.6064, 0.3953, 0.8738, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416], [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6144, 0.4006, 0.8502, 0.3668, 0.3638, 0.4422, 0.5844, 0.4963], [0.6249, 0.4046, 0.8022, 0.5589, 0.3946, 0.4815, 0.7036, 0.5593], [0.6223, 0.4128, 0.8770, 0.2965, 0.4358, 0.2833, 0.6508, 0.5138], [0.0960, 0.0492, 0.8665, 0.2693, 0.5726, 0.2490, 0.7453, 0.5857], [0.6294, 0.4175, 0.8569, 0.3660, 0.3810, 0.3529, 0.6021, 0.5679], [0.6125, 0.3823, 0.8703, 0.4571, 0.3776, 0.4691, 0.5518, 0.5235], [0.6238, 0.4022, 0.8976, 0.3624, 0.3795, 0.4454, 0.6836, 0.5143], [0.6391, 0.4209, 0.8698, 0.5133, 0.3780, 0.3417, 0.6381, 0.5009]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6213, 0.4131, 0.8438, 0.3550, 0.3512, 0.4400, 0.5716, 0.5123], [0.6346, 0.4086, 0.7937, 0.5500, 0.3963, 0.4867, 0.7343, 0.5702], [0.6266, 0.4067, 0.8587, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267], [0.0000, 0.0000, 0.8462, 0.2550, 0.5850, 0.2133, 0.7129, 0.6072], [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5987, 0.5667], [0.6064, 0.3952, 0.8737, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416], [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.013119590788846835 step: 20 running loss: 0.0006559795394423418 Train Steps: 20/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6272, 0.4045, 0.8538, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467], [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485], [0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799], [0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133], [0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6296, 0.4074, 0.8948, 0.3214, 0.4477, 0.2529, 0.6697, 0.5555], [0.6054, 0.3959, 0.8685, 0.5957, 0.3754, 0.4682, 0.6372, 0.4977], [0.5450, 0.3340, 0.8945, 0.4939, 0.4314, 0.5339, 0.6315, 0.5598], [0.5763, 0.3613, 0.8921, 0.3362, 0.3809, 0.3682, 0.6608, 0.5188], [0.6269, 0.4156, 0.7311, 0.3101, 0.3836, 0.3070, 0.5511, 0.5822], [0.5670, 0.3752, 0.8318, 0.4177, 0.3885, 0.3006, 0.5604, 0.5971], [0.6255, 0.3709, 0.8807, 0.5453, 0.3992, 0.5174, 0.6570, 0.5380], [0.5794, 0.3761, 0.6974, 0.2540, 0.4204, 0.2369, 0.5726, 0.6080]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6250, 0.4131, 0.8687, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6271, 0.4045, 0.8537, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6199, 0.4059, 0.8700, 0.4900, 0.4162, 0.5000, 0.6162, 0.5467], [0.6255, 0.4017, 0.8687, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6123, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485], [0.6200, 0.4118, 0.8288, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799], [0.6308, 0.3990, 0.8687, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133], [0.6147, 0.4026, 0.6600, 0.2467, 0.4087, 0.2150, 0.5489, 0.5773]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.013877765712095425 step: 21 running loss: 0.0006608459862902583 Train Steps: 21/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033], [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285], [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050], [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717], [0.6325, 0.4165, 0.9000, 0.4617, 0.3813, 0.4900, 0.7485, 0.5447], [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100], [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5650, 0.3551, 0.8819, 0.5313, 0.4009, 0.5310, 0.5723, 0.5404], [0.5824, 0.3737, 0.9007, 0.3910, 0.4633, 0.2645, 0.6485, 0.5586], [0.6109, 0.3947, 0.8851, 0.4863, 0.3972, 0.5470, 0.6387, 0.5904], [0.5914, 0.3826, 0.8671, 0.5249, 0.4320, 0.5236, 0.5967, 0.5469], [0.6290, 0.4103, 0.7343, 0.2679, 0.3852, 0.2887, 0.5911, 0.6182], [0.5890, 0.3860, 0.9081, 0.4825, 0.3769, 0.5022, 0.7166, 0.5768], [0.6258, 0.3809, 0.8503, 0.5562, 0.3911, 0.5099, 0.6123, 0.5244], [0.5949, 0.3755, 0.7443, 0.2458, 0.4314, 0.1738, 0.5385, 0.5607]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033], [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285], [0.6241, 0.4142, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050], [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717], [0.6325, 0.4165, 0.9000, 0.4617, 0.3812, 0.4900, 0.7485, 0.5447], [0.6275, 0.4024, 0.8500, 0.5383, 0.3913, 0.4883, 0.6288, 0.5100], [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.014404586370801553 step: 22 running loss: 0.0006547539259455251 Train Steps: 22/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6289, 0.4032, 0.8419, 0.5446, 0.4075, 0.5017, 0.6312, 0.5117], [0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267], [0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485], [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617], [0.6090, 0.4010, 0.7838, 0.3483, 0.3538, 0.3783, 0.5462, 0.5077], [0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799], [ nan, nan, 0.8850, 0.3000, 0.5363, 0.2250, 0.7343, 0.5771], [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6460, 0.3959, 0.8257, 0.5581, 0.3873, 0.4902, 0.6624, 0.5208], [0.6283, 0.4119, 0.8619, 0.2948, 0.4182, 0.2608, 0.6523, 0.5370], [0.6427, 0.4274, 0.6905, 0.2963, 0.3651, 0.2692, 0.5266, 0.5680], [0.6403, 0.4086, 0.8583, 0.4989, 0.4620, 0.4748, 0.5454, 0.5695], [0.6413, 0.4226, 0.7729, 0.3604, 0.3415, 0.3635, 0.5549, 0.5128], [0.6557, 0.4392, 0.8538, 0.3977, 0.3870, 0.3068, 0.5797, 0.5940], [0.0340, 0.0126, 0.8629, 0.2940, 0.5285, 0.2054, 0.7630, 0.5743], [0.5121, 0.3272, 0.6654, 0.2389, 0.4041, 0.1704, 0.5190, 0.5577]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6289, 0.4031, 0.8419, 0.5446, 0.4075, 0.5017, 0.6313, 0.5117], [0.6266, 0.4067, 0.8587, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267], [0.6123, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485], [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617], [0.6090, 0.4010, 0.7837, 0.3483, 0.3537, 0.3783, 0.5462, 0.5077], [0.6250, 0.4236, 0.8637, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799], [0.0000, 0.0000, 0.8850, 0.3000, 0.5362, 0.2250, 0.7343, 0.5771], [0.6159, 0.4085, 0.6900, 0.2283, 0.4087, 0.1950, 0.5123, 0.5397]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.014939425949705765 step: 23 running loss: 0.0006495402586828593 Train Steps: 23/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6144, 0.4032, 0.8563, 0.3283, 0.3525, 0.4200, 0.5775, 0.5583], [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850], [0.6188, 0.4099, 0.7400, 0.2433, 0.3962, 0.2750, 0.6162, 0.5467], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517], [0.6198, 0.4101, 0.8838, 0.5283, 0.3763, 0.5267, 0.5913, 0.5567], [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6195, 0.3847, 0.8599, 0.3413, 0.3459, 0.3976, 0.5567, 0.5486], [0.6105, 0.3927, 0.8786, 0.4613, 0.4847, 0.5375, 0.6117, 0.5935], [0.6538, 0.4311, 0.7202, 0.2353, 0.3908, 0.2772, 0.5970, 0.5718], [0.6367, 0.4093, 0.8734, 0.4899, 0.4099, 0.5058, 0.6015, 0.5389], [0.6352, 0.3884, 0.8787, 0.5097, 0.3985, 0.4984, 0.6342, 0.4904], [0.6424, 0.4127, 0.8495, 0.4789, 0.4571, 0.4931, 0.5678, 0.5521], [0.6069, 0.3848, 0.8498, 0.5368, 0.3684, 0.4988, 0.5818, 0.5520], [0.6352, 0.3954, 0.8017, 0.2887, 0.4095, 0.2181, 0.5965, 0.5479]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6144, 0.4032, 0.8562, 0.3283, 0.3525, 0.4200, 0.5775, 0.5583], [0.6184, 0.4079, 0.8838, 0.4617, 0.4837, 0.5650, 0.6175, 0.5850], [0.6188, 0.4099, 0.7400, 0.2433, 0.3963, 0.2750, 0.6162, 0.5467], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6175, 0.3957, 0.8700, 0.4817, 0.4663, 0.5133, 0.5800, 0.5517], [0.6198, 0.4101, 0.8838, 0.5283, 0.3762, 0.5267, 0.5913, 0.5567], [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391]]], device='cuda:0') loss_train_step before backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train: 0.015169626043643802 step: 24 running loss: 0.0006320677518184917 Train Steps: 24/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5887, 0.5017], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864], [0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167], [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5988, 0.5700], [0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019], [0.6115, 0.4081, 0.6725, 0.2433, 0.4088, 0.1933, 0.5167, 0.5544]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5821, 0.3851, 0.8686, 0.4562, 0.3777, 0.4854, 0.5571, 0.5129], [0.5867, 0.3747, 0.8297, 0.5226, 0.4066, 0.5395, 0.6983, 0.5600], [0.6231, 0.4036, 0.8612, 0.5096, 0.3793, 0.4699, 0.6989, 0.5680], [0.6329, 0.3993, 0.8323, 0.5190, 0.4639, 0.5174, 0.5098, 0.4993], [0.6596, 0.4292, 0.8100, 0.3271, 0.3733, 0.2925, 0.5534, 0.5063], [0.6017, 0.3856, 0.7856, 0.2735, 0.3719, 0.3316, 0.5911, 0.5630], [0.6342, 0.4121, 0.8748, 0.4167, 0.3754, 0.5315, 0.5385, 0.4856], [0.6165, 0.3957, 0.6680, 0.2330, 0.4212, 0.1719, 0.5023, 0.5374]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5888, 0.5017], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6314, 0.4107, 0.8750, 0.5100, 0.3787, 0.4900, 0.7121, 0.5864], [0.6075, 0.4000, 0.8512, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167], [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5987, 0.5700], [0.6161, 0.4099, 0.8737, 0.4383, 0.3787, 0.5483, 0.5605, 0.5019], [0.6115, 0.4081, 0.6725, 0.2433, 0.4087, 0.1933, 0.5167, 0.5544]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.015483570605283603 step: 25 running loss: 0.0006193428242113441 Train Steps: 25/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378], [0.6339, 0.4118, 0.7988, 0.5800, 0.3912, 0.4583, 0.7343, 0.5760], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [ nan, nan, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733], [0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233], [0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090], [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6852, 0.4632, 0.8482, 0.2990, 0.3883, 0.3077, 0.5581, 0.5466], [ 0.6258, 0.4264, 0.8568, 0.5108, 0.3806, 0.4525, 0.5085, 0.5110], [ 0.6581, 0.4255, 0.7901, 0.5492, 0.3743, 0.4760, 0.7085, 0.5539], [ 0.7054, 0.4497, 0.8871, 0.4031, 0.3478, 0.4665, 0.5962, 0.4949], [-0.0766, -0.0439, 0.6816, 0.1964, 0.4734, 0.1699, 0.5512, 0.5810], [ 0.6778, 0.4353, 0.8170, 0.5715, 0.4026, 0.4640, 0.5526, 0.5143], [ 0.6876, 0.4435, 0.8741, 0.3700, 0.3973, 0.3027, 0.6253, 0.4798], [ 0.6833, 0.4596, 0.8834, 0.4272, 0.3690, 0.3819, 0.5770, 0.5657]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633], [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378], [0.6339, 0.4118, 0.7987, 0.5800, 0.3913, 0.4583, 0.7343, 0.5760], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.0000, 0.0000, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733], [0.6222, 0.3937, 0.8350, 0.5617, 0.4137, 0.4600, 0.5800, 0.5233], [0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090], [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.016324585973052308 step: 26 running loss: 0.0006278686912712426 Train Steps: 26/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6275, 0.4157, 0.8337, 0.5800, 0.3763, 0.4200, 0.5547, 0.6125], [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117], [0.6097, 0.4024, 0.8488, 0.3717, 0.3875, 0.5517, 0.5836, 0.5591], [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881], [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6205, 0.3967, 0.9026, 0.4544, 0.3932, 0.4375, 0.6945, 0.5092], [0.6907, 0.4562, 0.8337, 0.5767, 0.3931, 0.5079, 0.6054, 0.4719], [0.6233, 0.4279, 0.8179, 0.5471, 0.3836, 0.4258, 0.5543, 0.6103], [0.6515, 0.4255, 0.8997, 0.4725, 0.3777, 0.4727, 0.6153, 0.5013], [0.6181, 0.4017, 0.8575, 0.3643, 0.4084, 0.5700, 0.5951, 0.5194], [0.6669, 0.4463, 0.8475, 0.5062, 0.3903, 0.4386, 0.4891, 0.5614], [0.6693, 0.4533, 0.8795, 0.4952, 0.3682, 0.4686, 0.5514, 0.5426], [0.6131, 0.4099, 0.8660, 0.4607, 0.3811, 0.4951, 0.5762, 0.5430]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6275, 0.4157, 0.8338, 0.5800, 0.3762, 0.4200, 0.5547, 0.6125], [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117], [0.6097, 0.4024, 0.8487, 0.3717, 0.3875, 0.5517, 0.5836, 0.5591], [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881], [0.6222, 0.4171, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633], [0.6197, 0.4091, 0.8800, 0.4783, 0.3537, 0.4767, 0.5950, 0.5550]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.016808000102173537 step: 27 running loss: 0.0006225185223027236 Train Steps: 27/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507], [0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427], [0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5563, 0.5533], [ nan, nan, 0.8888, 0.3100, 0.5262, 0.2817, 0.7145, 0.6003], [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533], [0.6222, 0.4169, 0.8638, 0.5650, 0.4313, 0.4783, 0.5637, 0.5633], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6678, 0.4515, 0.6983, 0.1999, 0.3860, 0.2305, 0.5475, 0.5385], [ 0.6567, 0.4500, 0.7093, 0.2060, 0.3911, 0.2487, 0.5784, 0.5428], [ 0.6494, 0.4381, 0.7825, 0.4043, 0.3370, 0.3878, 0.5269, 0.5209], [ 0.5652, 0.3896, 0.8628, 0.5309, 0.3852, 0.4684, 0.5787, 0.5374], [-0.0491, -0.0215, 0.9024, 0.3033, 0.5342, 0.2410, 0.7110, 0.5998], [ 0.7045, 0.4607, 0.7032, 0.2196, 0.4323, 0.1755, 0.5227, 0.5387], [ 0.6341, 0.4296, 0.8694, 0.5635, 0.4121, 0.4887, 0.5844, 0.5403], [ 0.6351, 0.4425, 0.8114, 0.2984, 0.3520, 0.3987, 0.5705, 0.5170]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6072, 0.4029, 0.7038, 0.2150, 0.3913, 0.2267, 0.5516, 0.5507], [0.6187, 0.4103, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427], [0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5562, 0.5533], [0.0000, 0.0000, 0.8888, 0.3100, 0.5263, 0.2817, 0.7145, 0.6003], [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533], [0.6222, 0.4169, 0.8637, 0.5650, 0.4313, 0.4783, 0.5638, 0.5633], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.0173894782201387 step: 28 running loss: 0.0006210527935763821 Train Steps: 28/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6346, 0.4165, 0.9138, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471], [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207], [ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617], [0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050], [0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5899, 0.4050, 0.9158, 0.4174, 0.3807, 0.4620, 0.7336, 0.5475], [0.5687, 0.4059, 0.8109, 0.3149, 0.4253, 0.2654, 0.5626, 0.5792], [0.5833, 0.3953, 0.8511, 0.2648, 0.4415, 0.2807, 0.6835, 0.5603], [0.6309, 0.4139, 0.8280, 0.2610, 0.4369, 0.2403, 0.6261, 0.5160], [0.1151, 0.0865, 0.7205, 0.2217, 0.4265, 0.1935, 0.5202, 0.5690], [0.6392, 0.4385, 0.8658, 0.5408, 0.4172, 0.5399, 0.5582, 0.5038], [0.6263, 0.4299, 0.6765, 0.2299, 0.3918, 0.2112, 0.4994, 0.5611], [0.6386, 0.4270, 0.8344, 0.2350, 0.4449, 0.1997, 0.6108, 0.5236]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6346, 0.4165, 0.9137, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471], [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207], [0.0000, 0.0000, 0.7513, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617], [0.6097, 0.3988, 0.8650, 0.5250, 0.4212, 0.5200, 0.5675, 0.5050], [0.6149, 0.4054, 0.6712, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.018027903337497264 step: 29 running loss: 0.0006216518392240436 Train Steps: 29/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4064, 0.8688, 0.5050, 0.4225, 0.5100, 0.6138, 0.5500], [0.6118, 0.4052, 0.8463, 0.3917, 0.3538, 0.3450, 0.5053, 0.5593], [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6179, 0.4008, 0.8600, 0.4015, 0.3932, 0.2515, 0.5711, 0.5438], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374], [0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183], [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6289, 0.4156, 0.8779, 0.4982, 0.4202, 0.5133, 0.6175, 0.5748], [0.6280, 0.4271, 0.8509, 0.3736, 0.3577, 0.3567, 0.4989, 0.5659], [0.0764, 0.0679, 0.7582, 0.2679, 0.4156, 0.2245, 0.5097, 0.5756], [0.6544, 0.4261, 0.9022, 0.5067, 0.4018, 0.5514, 0.7537, 0.5663], [0.6425, 0.4273, 0.8398, 0.3622, 0.4102, 0.2657, 0.5652, 0.5576], [0.6033, 0.3992, 0.8812, 0.3680, 0.3698, 0.3246, 0.5951, 0.5382], [0.6100, 0.4027, 0.8141, 0.2939, 0.3899, 0.2922, 0.6108, 0.5352], [0.6272, 0.4086, 0.7947, 0.2922, 0.3735, 0.3166, 0.5857, 0.5293]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6201, 0.4064, 0.8687, 0.5050, 0.4225, 0.5100, 0.6137, 0.5500], [0.6118, 0.4052, 0.8462, 0.3917, 0.3537, 0.3450, 0.5053, 0.5593], [0.0000, 0.0000, 0.7515, 0.2708, 0.3988, 0.2267, 0.5163, 0.5567], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6179, 0.4008, 0.8600, 0.4015, 0.3932, 0.2515, 0.5711, 0.5438], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374], [0.6201, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183], [0.6135, 0.3994, 0.7912, 0.3050, 0.3625, 0.3050, 0.5838, 0.5050]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.018376984866335988 step: 30 running loss: 0.0006125661622111996 Train Steps: 30/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350], [0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5082, 0.3386, 0.9000, 0.4524, 0.4091, 0.5285, 0.5743, 0.5588], [0.6539, 0.4280, 0.9310, 0.3621, 0.3786, 0.2677, 0.6350, 0.5476], [0.5480, 0.3625, 0.6927, 0.2695, 0.4049, 0.2346, 0.5865, 0.5690], [0.5836, 0.3790, 0.8909, 0.5103, 0.3678, 0.4543, 0.5583, 0.5349], [0.5760, 0.3655, 0.8916, 0.4884, 0.4742, 0.5025, 0.5968, 0.5890], [0.5642, 0.3593, 0.9249, 0.3504, 0.4558, 0.3376, 0.7363, 0.5549], [0.5540, 0.3576, 0.8712, 0.5523, 0.3851, 0.4578, 0.6114, 0.5569], [0.6044, 0.3882, 0.8784, 0.3101, 0.3934, 0.2648, 0.6318, 0.4963]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5838, 0.5500], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6187, 0.5283], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.6143, 0.3982, 0.8650, 0.4883, 0.3913, 0.4317, 0.5315, 0.5350], [0.6112, 0.4029, 0.8637, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402], [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.019246386946178973 step: 31 running loss: 0.0006208511918122249 Train Steps: 31/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116], [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374], [0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5887, 0.5017], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274], [0.6361, 0.4076, 0.8862, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[-0.0313, -0.0344, 0.6675, 0.2363, 0.4315, 0.1697, 0.5175, 0.5917], [ 0.7018, 0.4575, 0.8709, 0.4227, 0.3885, 0.5999, 0.5903, 0.5508], [ 0.6001, 0.3855, 0.9231, 0.4592, 0.4337, 0.3570, 0.7339, 0.6016], [ 0.5920, 0.3758, 0.8785, 0.3899, 0.3640, 0.3185, 0.6041, 0.5439], [ 0.6057, 0.3949, 0.8959, 0.4820, 0.3706, 0.4992, 0.6088, 0.5532], [ 0.6054, 0.3812, 0.7647, 0.2485, 0.3684, 0.3098, 0.6237, 0.5450], [ 0.6482, 0.4166, 0.8815, 0.3270, 0.4856, 0.1997, 0.6592, 0.5412], [ 0.6420, 0.4174, 0.8970, 0.5579, 0.3575, 0.4708, 0.6795, 0.5610]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116], [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374], [0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5888, 0.5017], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274], [0.6361, 0.4076, 0.8863, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.019706396793480963 step: 32 running loss: 0.0006158248997962801 Train Steps: 32/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6097, 0.4000, 0.7325, 0.2667, 0.3450, 0.3517, 0.5284, 0.5045], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413], [0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100], [0.6289, 0.4024, 0.9088, 0.4567, 0.3937, 0.5633, 0.7058, 0.5609]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5915, 0.3648, 0.8859, 0.3691, 0.3865, 0.2511, 0.6040, 0.4740], [0.5877, 0.3866, 0.7135, 0.2236, 0.3872, 0.2383, 0.5726, 0.5677], [0.5589, 0.3525, 0.7566, 0.2944, 0.3430, 0.3362, 0.5234, 0.5213], [0.5980, 0.3662, 0.8820, 0.4993, 0.4230, 0.5237, 0.6867, 0.5484], [0.5208, 0.3225, 0.9172, 0.3313, 0.4996, 0.2441, 0.7192, 0.5617], [0.5197, 0.3270, 0.7861, 0.2565, 0.4710, 0.1536, 0.5770, 0.5490], [0.5806, 0.3592, 0.8683, 0.5452, 0.3901, 0.4801, 0.6250, 0.5073], [0.5634, 0.3657, 0.9120, 0.4750, 0.3990, 0.5621, 0.7043, 0.5740]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6277, 0.4036, 0.8687, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6187, 0.4103, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6097, 0.4000, 0.7325, 0.2667, 0.3450, 0.3517, 0.5284, 0.5045], [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378], [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413], [0.6245, 0.4100, 0.7763, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6275, 0.4024, 0.8500, 0.5383, 0.3913, 0.4883, 0.6288, 0.5100], [0.6289, 0.4024, 0.9087, 0.4567, 0.3938, 0.5633, 0.7058, 0.5609]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.020707069255877286 step: 33 running loss: 0.0006274869471477965 Train Steps: 33/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6127, 0.4115, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495], [0.6171, 0.4127, 0.8900, 0.4800, 0.4325, 0.5783, 0.5769, 0.5090], [0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947], [0.6226, 0.4103, 0.8575, 0.3450, 0.4388, 0.2067, 0.5787, 0.5383], [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467], [0.6064, 0.3953, 0.8738, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6007, 0.3772, 0.7200, 0.2945, 0.3763, 0.2790, 0.5664, 0.5535], [0.5869, 0.3720, 0.9010, 0.4810, 0.4447, 0.5518, 0.5943, 0.4871], [0.6122, 0.3806, 0.8564, 0.4092, 0.3590, 0.4167, 0.6089, 0.5095], [0.5744, 0.3446, 0.8711, 0.5548, 0.4080, 0.4152, 0.6318, 0.5181], [0.5921, 0.3604, 0.7916, 0.1798, 0.4442, 0.2247, 0.6660, 0.4946], [0.5655, 0.3537, 0.8685, 0.3517, 0.4483, 0.2133, 0.5934, 0.5236], [0.5666, 0.3661, 0.6690, 0.2627, 0.4073, 0.2356, 0.6014, 0.5319], [0.5724, 0.3551, 0.8916, 0.4487, 0.3799, 0.4450, 0.5941, 0.5318]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6127, 0.4114, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495], [0.6171, 0.4127, 0.8900, 0.4800, 0.4325, 0.5783, 0.5769, 0.5090], [0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947], [0.6226, 0.4103, 0.8575, 0.3450, 0.4387, 0.2067, 0.5788, 0.5383], [0.6162, 0.4134, 0.6700, 0.2467, 0.3963, 0.2533, 0.5738, 0.5467], [0.6064, 0.3952, 0.8737, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.021375142212491482 step: 34 running loss: 0.000628680653308573 Train Steps: 34/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167], [ nan, nan, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737], [0.6111, 0.4033, 0.8300, 0.3267, 0.3588, 0.3333, 0.5444, 0.5637], [0.6296, 0.4045, 0.9138, 0.4100, 0.4232, 0.4242, 0.7422, 0.5297], [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917], [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791], [0.6152, 0.4131, 0.6863, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6458, 0.4290, 0.7912, 0.3173, 0.3675, 0.3524, 0.5973, 0.5699], [0.0816, 0.0432, 0.8271, 0.2405, 0.5499, 0.2154, 0.6919, 0.5486], [0.6858, 0.4386, 0.8185, 0.3300, 0.3570, 0.2973, 0.5412, 0.5117], [0.7040, 0.4585, 0.8861, 0.4092, 0.4093, 0.4017, 0.7211, 0.5145], [0.6847, 0.4358, 0.7422, 0.3976, 0.3619, 0.3103, 0.5389, 0.5399], [0.7012, 0.4481, 0.8807, 0.4806, 0.4154, 0.4745, 0.6137, 0.4292], [0.6557, 0.4236, 0.6619, 0.2601, 0.3675, 0.3147, 0.5812, 0.5161], [0.0547, 0.0357, 0.8371, 0.2571, 0.5378, 0.1797, 0.6543, 0.5340]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6037, 0.6167], [0.0000, 0.0000, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737], [0.6111, 0.4033, 0.8300, 0.3267, 0.3587, 0.3333, 0.5444, 0.5637], [0.6296, 0.4045, 0.9137, 0.4100, 0.4232, 0.4242, 0.7422, 0.5297], [0.6179, 0.4118, 0.7278, 0.4237, 0.3587, 0.3400, 0.5675, 0.5917], [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791], [0.6152, 0.4131, 0.6862, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [0.0000, 0.0000, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.022489308787044138 step: 35 running loss: 0.0006425516796298325 Train Steps: 35/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283], [ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617], [0.6133, 0.4066, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.0647, 0.0577, 0.7094, 0.2131, 0.4062, 0.2470, 0.5556, 0.5255], [0.7170, 0.4711, 0.8321, 0.5520, 0.4031, 0.4867, 0.6873, 0.5007], [0.6999, 0.4748, 0.7984, 0.3779, 0.5006, 0.2218, 0.5600, 0.6054], [0.6240, 0.4170, 0.9029, 0.3641, 0.3922, 0.3674, 0.6687, 0.5235], [0.0865, 0.0725, 0.7177, 0.2025, 0.4379, 0.2046, 0.5630, 0.5395], [0.6507, 0.4434, 0.6844, 0.2552, 0.4029, 0.2298, 0.5541, 0.5148], [0.6804, 0.4533, 0.8606, 0.5392, 0.3824, 0.3423, 0.5690, 0.5038], [0.5626, 0.3770, 0.7266, 0.2565, 0.4269, 0.2528, 0.5726, 0.5420]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.7425, 0.2117, 0.3938, 0.2433, 0.5437, 0.5567], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6299, 0.4303, 0.7962, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283], [0.0000, 0.0000, 0.7513, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617], [0.6133, 0.4065, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5688, 0.5567]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.023610311618540436 step: 36 running loss: 0.000655841989403901 Train Steps: 36/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510], [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284], [0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5305, 0.3545, 0.6789, 0.2938, 0.3590, 0.3198, 0.5248, 0.5454], [0.6119, 0.3984, 0.7472, 0.2210, 0.4547, 0.1996, 0.5760, 0.5310], [0.6218, 0.4105, 0.8767, 0.3654, 0.3838, 0.2859, 0.5847, 0.5196], [0.5847, 0.3989, 0.6975, 0.2397, 0.3882, 0.3049, 0.5973, 0.5677], [0.5755, 0.3947, 0.8557, 0.5025, 0.4140, 0.5360, 0.5966, 0.5015], [0.5259, 0.3512, 0.8314, 0.2048, 0.5442, 0.2115, 0.6936, 0.5579], [0.6149, 0.4094, 0.8563, 0.4835, 0.4030, 0.5499, 0.6354, 0.5013], [0.6113, 0.4014, 0.7697, 0.3001, 0.3698, 0.3214, 0.5749, 0.5048]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510], [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284], [0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413], [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217], [0.6135, 0.3994, 0.7912, 0.3050, 0.3625, 0.3050, 0.5838, 0.5050]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.024129573779646307 step: 37 running loss: 0.0006521506426931435 Train Steps: 37/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320], [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960], [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343], [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690], [0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567], [ nan, nan, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5915, 0.4067, 0.7443, 0.3600, 0.3274, 0.4274, 0.5061, 0.5529], [0.6377, 0.4158, 0.8687, 0.4530, 0.3734, 0.4218, 0.6892, 0.5308], [0.5794, 0.3966, 0.8383, 0.4475, 0.4386, 0.5175, 0.5149, 0.4928], [0.6338, 0.4248, 0.8548, 0.3710, 0.4222, 0.3116, 0.6730, 0.5358], [0.7068, 0.4678, 0.7913, 0.5072, 0.4048, 0.5658, 0.7041, 0.5697], [0.6408, 0.4241, 0.8375, 0.3433, 0.3799, 0.2638, 0.5986, 0.4770], [0.6320, 0.4270, 0.6807, 0.1809, 0.3609, 0.2769, 0.5538, 0.5540], [0.0137, 0.0190, 0.8765, 0.3296, 0.4928, 0.2607, 0.7076, 0.5754]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320], [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960], [0.6275, 0.4003, 0.9100, 0.3783, 0.4387, 0.3133, 0.7058, 0.5343], [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690], [0.6277, 0.4036, 0.8687, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6108, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567], [0.0000, 0.0000, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.02468388859415427 step: 38 running loss: 0.0006495760156356386 Train Steps: 38/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6090, 0.4010, 0.7838, 0.3483, 0.3538, 0.3783, 0.5462, 0.5077], [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407], [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233], [ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563], [0.6274, 0.4003, 0.8638, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100], [0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6129, 0.4063, 0.7917, 0.3597, 0.3569, 0.3850, 0.5596, 0.5365], [0.6281, 0.4296, 0.8484, 0.4040, 0.3571, 0.3257, 0.5343, 0.5817], [0.6373, 0.4325, 0.8869, 0.4529, 0.3767, 0.5279, 0.6611, 0.5714], [0.1334, 0.0916, 0.7740, 0.3285, 0.3362, 0.3465, 0.5411, 0.5706], [0.6513, 0.4156, 0.8368, 0.5899, 0.3804, 0.4836, 0.6301, 0.5235], [0.6552, 0.4389, 0.8928, 0.4076, 0.3683, 0.3872, 0.6558, 0.5557], [0.6481, 0.4358, 0.8740, 0.4644, 0.3661, 0.3698, 0.5430, 0.5807], [0.6519, 0.4143, 0.8887, 0.4221, 0.3605, 0.4488, 0.6484, 0.5383]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6090, 0.4010, 0.7837, 0.3483, 0.3537, 0.3783, 0.5462, 0.5077], [0.6133, 0.4094, 0.8495, 0.4028, 0.3587, 0.3200, 0.5003, 0.5407], [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233], [0.0000, 0.0000, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563], [0.6274, 0.4003, 0.8637, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100], [0.6127, 0.4084, 0.8737, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.025601829052902758 step: 39 running loss: 0.0006564571552026349 Train Steps: 39/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631], [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540], [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726], [0.6124, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6216, 0.4167, 0.8588, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667], [0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5923, 0.3935, 0.8309, 0.4015, 0.3470, 0.4915, 0.5611, 0.5772], [0.5971, 0.3880, 0.9268, 0.4489, 0.4108, 0.3010, 0.6907, 0.5608], [0.5750, 0.3796, 0.8649, 0.3878, 0.3541, 0.3999, 0.5692, 0.5792], [0.6181, 0.4089, 0.7601, 0.2960, 0.3571, 0.2836, 0.5407, 0.4981], [0.5764, 0.3802, 0.8338, 0.5028, 0.3890, 0.4537, 0.5403, 0.5809], [0.5548, 0.3543, 0.8884, 0.4880, 0.4015, 0.5276, 0.6515, 0.5027], [0.5754, 0.3802, 0.8593, 0.5455, 0.3952, 0.5068, 0.6076, 0.5875], [0.6041, 0.3931, 0.8534, 0.3662, 0.3729, 0.2801, 0.5366, 0.5386]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631], [0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540], [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726], [0.6123, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6216, 0.4167, 0.8587, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667], [0.6186, 0.4080, 0.8625, 0.3483, 0.3787, 0.2650, 0.5320, 0.5272]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.026024623948615044 step: 40 running loss: 0.0006506155987153761 Train Steps: 40/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633], [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118], [0.6275, 0.4048, 0.8488, 0.2883, 0.4463, 0.2033, 0.6321, 0.5155], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500], [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5710, 0.3669, 0.8803, 0.4861, 0.3399, 0.3925, 0.5360, 0.5629], [0.6074, 0.3811, 0.8732, 0.5031, 0.3837, 0.5618, 0.7136, 0.5447], [0.6312, 0.4340, 0.7334, 0.3560, 0.4082, 0.2233, 0.5435, 0.6089], [0.5749, 0.3547, 0.8671, 0.2773, 0.4278, 0.2215, 0.6111, 0.4986], [0.5160, 0.3339, 0.8919, 0.4469, 0.3652, 0.4611, 0.5324, 0.5568], [0.5851, 0.3807, 0.8481, 0.4560, 0.4351, 0.2759, 0.5394, 0.6058], [0.5667, 0.3640, 0.8894, 0.5041, 0.3626, 0.4124, 0.5844, 0.5541], [0.5533, 0.3655, 0.7546, 0.2004, 0.3957, 0.2607, 0.6060, 0.5405]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5638, 0.5633], [0.6339, 0.4123, 0.8637, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118], [0.6275, 0.4048, 0.8487, 0.2883, 0.4462, 0.2033, 0.6321, 0.5155], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6262, 0.4163, 0.8850, 0.5183, 0.3762, 0.4150, 0.6025, 0.5500], [0.6200, 0.4071, 0.7337, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.02692328830016777 step: 41 running loss: 0.0006566655682967748 Train Steps: 41/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6362, 0.5200], [0.6164, 0.4102, 0.8850, 0.4867, 0.4213, 0.5983, 0.5609, 0.5038], [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901], [ nan, nan, 0.7225, 0.2167, 0.3987, 0.2283, 0.5427, 0.5181], [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540], [0.6205, 0.4081, 0.8950, 0.4017, 0.3788, 0.4700, 0.5963, 0.5667], [0.6339, 0.4149, 0.8800, 0.5000, 0.3900, 0.5283, 0.7541, 0.5424], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6538, 0.4220, 0.8873, 0.3851, 0.3581, 0.5102, 0.6221, 0.5243], [0.6167, 0.4197, 0.9097, 0.5269, 0.4462, 0.5868, 0.5661, 0.5541], [0.6237, 0.4020, 0.9165, 0.4854, 0.4226, 0.4382, 0.4926, 0.5166], [0.0564, 0.0392, 0.7513, 0.2278, 0.3858, 0.2244, 0.5631, 0.5290], [0.6554, 0.4202, 0.9537, 0.4827, 0.4247, 0.2915, 0.6790, 0.5506], [0.6010, 0.3892, 0.9020, 0.4234, 0.3771, 0.4627, 0.5740, 0.5795], [0.6802, 0.4307, 0.9158, 0.5482, 0.3977, 0.5348, 0.6974, 0.5587], [0.6571, 0.4200, 0.7340, 0.2392, 0.4364, 0.1543, 0.5420, 0.5228]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6363, 0.5200], [0.6164, 0.4102, 0.8850, 0.4867, 0.4212, 0.5983, 0.5609, 0.5038], [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901], [0.0000, 0.0000, 0.7225, 0.2167, 0.3988, 0.2283, 0.5427, 0.5181], [0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540], [0.6205, 0.4081, 0.8950, 0.4017, 0.3787, 0.4700, 0.5962, 0.5667], [0.6339, 0.4149, 0.8800, 0.5000, 0.3900, 0.5283, 0.7541, 0.5424], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.027541582996491343 step: 42 running loss: 0.0006557519761069367 Train Steps: 42/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6222, 0.3957, 0.8838, 0.5017, 0.3937, 0.4600, 0.5900, 0.5017], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852], [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954], [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6312, 0.5250], [0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783], [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6118, 0.3809, 0.9111, 0.5268, 0.4022, 0.4456, 0.5834, 0.5014], [0.5783, 0.3738, 0.8171, 0.3106, 0.3702, 0.2482, 0.5296, 0.4924], [0.5396, 0.3483, 0.8902, 0.5831, 0.3964, 0.4131, 0.5367, 0.4819], [0.5588, 0.3638, 0.9016, 0.5124, 0.4678, 0.4860, 0.5316, 0.5854], [0.5850, 0.3821, 0.8373, 0.3984, 0.3518, 0.3892, 0.5544, 0.4998], [0.5832, 0.3674, 0.9191, 0.3663, 0.4033, 0.3351, 0.6651, 0.5223], [0.5833, 0.3871, 0.7874, 0.2756, 0.4619, 0.2136, 0.5767, 0.5614], [0.5861, 0.3766, 0.7723, 0.2376, 0.4008, 0.2322, 0.6142, 0.5345]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6222, 0.3957, 0.8838, 0.5017, 0.3938, 0.4600, 0.5900, 0.5017], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.6140, 0.4070, 0.8700, 0.5000, 0.4613, 0.4900, 0.5260, 0.5852], [0.6076, 0.3952, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954], [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6313, 0.5250], [0.6229, 0.4198, 0.7663, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783], [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.028159943351056427 step: 43 running loss: 0.0006548824035129402 Train Steps: 43/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983], [0.6193, 0.4108, 0.7438, 0.2700, 0.3650, 0.3683, 0.6238, 0.5717], [0.6092, 0.4001, 0.8638, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064], [ nan, nan, 0.8300, 0.3150, 0.3588, 0.3383, 0.5208, 0.5194], [0.6250, 0.4110, 0.7238, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633], [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162], [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343], [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6471, 0.4310, 0.8641, 0.5437, 0.3815, 0.3744, 0.5533, 0.5701], [0.6438, 0.4260, 0.7552, 0.2701, 0.3719, 0.3479, 0.6202, 0.5499], [0.6414, 0.4279, 0.8818, 0.4889, 0.4370, 0.5421, 0.5285, 0.4820], [0.0014, 0.0012, 0.8453, 0.3188, 0.3674, 0.3091, 0.5420, 0.5049], [0.6419, 0.4286, 0.7340, 0.2096, 0.4328, 0.1638, 0.5606, 0.5363], [0.6953, 0.4560, 0.9059, 0.4319, 0.3676, 0.4678, 0.6474, 0.4943], [0.6849, 0.4451, 0.9102, 0.3899, 0.4505, 0.2879, 0.6936, 0.5105], [0.6586, 0.4198, 0.8778, 0.5090, 0.4006, 0.4320, 0.5175, 0.4929]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6260, 0.4214, 0.8537, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983], [0.6193, 0.4108, 0.7437, 0.2700, 0.3650, 0.3683, 0.6237, 0.5717], [0.6092, 0.4001, 0.8637, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064], [0.0000, 0.0000, 0.8300, 0.3150, 0.3587, 0.3383, 0.5208, 0.5194], [0.6250, 0.4110, 0.7237, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633], [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162], [0.6275, 0.4003, 0.9100, 0.3783, 0.4387, 0.3133, 0.7058, 0.5343], [0.6143, 0.3982, 0.8650, 0.4883, 0.3913, 0.4317, 0.5315, 0.5350]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.028655284026172012 step: 44 running loss: 0.000651256455140273 Train Steps: 44/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6183, 0.4076, 0.8838, 0.4517, 0.3813, 0.4483, 0.5775, 0.5633], [0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954], [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6362, 0.5000], [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6124, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426], [0.6221, 0.4107, 0.7788, 0.3033, 0.3950, 0.2817, 0.6075, 0.5517]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5749, 0.3847, 0.8832, 0.4350, 0.3910, 0.4564, 0.5788, 0.5423], [0.5752, 0.3708, 0.8686, 0.4065, 0.4086, 0.2554, 0.5023, 0.4715], [0.5742, 0.3650, 0.8813, 0.5068, 0.3964, 0.5120, 0.6576, 0.4739], [0.6020, 0.3978, 0.8577, 0.5034, 0.5040, 0.5041, 0.5080, 0.5240], [0.6640, 0.4409, 0.7202, 0.2139, 0.4073, 0.2258, 0.5346, 0.5368], [0.6279, 0.4088, 0.8483, 0.4965, 0.4003, 0.4486, 0.5340, 0.5275], [0.6190, 0.3872, 0.8941, 0.4202, 0.3754, 0.3232, 0.5813, 0.4973], [0.6085, 0.4068, 0.8025, 0.2739, 0.4027, 0.2712, 0.6047, 0.5237]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6183, 0.4076, 0.8838, 0.4517, 0.3812, 0.4483, 0.5775, 0.5633], [0.6137, 0.4038, 0.8562, 0.4050, 0.3812, 0.2550, 0.5106, 0.4954], [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6363, 0.5000], [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6123, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426], [0.6221, 0.4107, 0.7788, 0.3033, 0.3950, 0.2817, 0.6075, 0.5517]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.02907089513610117 step: 45 running loss: 0.0006460198919133593 Train Steps: 45/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391], [0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6424, 0.4252, 0.8419, 0.5430, 0.4097, 0.4271, 0.5317, 0.5429], [0.6416, 0.4061, 0.8005, 0.2153, 0.4281, 0.2328, 0.6118, 0.5114], [0.5964, 0.3905, 0.8354, 0.4923, 0.4562, 0.5312, 0.5099, 0.5059], [0.6142, 0.4033, 0.8649, 0.3768, 0.3471, 0.3857, 0.5197, 0.4894], [0.6468, 0.4079, 0.8238, 0.5167, 0.3851, 0.5406, 0.6945, 0.5502], [0.5346, 0.3638, 0.7716, 0.1978, 0.4523, 0.1814, 0.6084, 0.5259], [0.5847, 0.3879, 0.8422, 0.2714, 0.4162, 0.2223, 0.5703, 0.5198], [0.5986, 0.3831, 0.8542, 0.4812, 0.4572, 0.5328, 0.5836, 0.4991]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6075, 0.4000, 0.8512, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6082, 0.4024, 0.8737, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.6199, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391], [0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.02950749092269689 step: 46 running loss: 0.0006414671939716715 Train Steps: 46/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6127, 0.4115, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495], [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364], [ nan, nan, 0.8938, 0.2850, 0.4662, 0.3117, 0.7406, 0.5528], [0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297], [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240], [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507], [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6545, 0.4337, 0.8315, 0.5277, 0.4499, 0.5490, 0.5029, 0.5197], [0.6328, 0.4295, 0.6849, 0.2833, 0.3487, 0.3037, 0.4949, 0.5662], [0.6140, 0.4039, 0.7645, 0.2126, 0.4507, 0.2166, 0.6423, 0.5386], [0.1316, 0.0792, 0.9086, 0.2893, 0.4728, 0.2985, 0.7566, 0.5720], [0.6985, 0.4589, 0.8860, 0.4478, 0.4048, 0.3236, 0.6257, 0.5292], [0.5595, 0.3648, 0.7252, 0.2200, 0.4176, 0.1788, 0.5110, 0.5336], [0.6185, 0.4166, 0.6774, 0.2080, 0.3846, 0.2352, 0.5337, 0.5579], [0.6581, 0.4324, 0.8878, 0.4624, 0.3838, 0.4349, 0.6888, 0.5846]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6075, 0.4000, 0.8512, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6127, 0.4114, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495], [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364], [0.0000, 0.0000, 0.8938, 0.2850, 0.4663, 0.3117, 0.7406, 0.5528], [0.6364, 0.4165, 0.9087, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297], [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240], [0.6072, 0.4029, 0.7038, 0.2150, 0.3913, 0.2267, 0.5516, 0.5507], [0.6346, 0.4144, 0.9087, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.030339945456944406 step: 47 running loss: 0.0006455307544030725 Train Steps: 47/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6197, 0.4118, 0.8688, 0.5517, 0.4037, 0.5233, 0.5875, 0.5600], [0.6289, 0.4032, 0.8419, 0.5446, 0.4075, 0.5017, 0.6312, 0.5117], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6353, 0.4128, 0.8488, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6063, 0.5617]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6954, 0.4573, 0.8934, 0.3729, 0.4420, 0.2353, 0.6405, 0.5110], [0.6154, 0.4141, 0.7470, 0.3557, 0.3407, 0.4372, 0.5124, 0.5594], [0.6399, 0.4215, 0.8431, 0.5455, 0.3859, 0.5617, 0.5871, 0.5451], [0.6166, 0.3978, 0.8250, 0.5295, 0.3898, 0.5311, 0.6521, 0.5186], [0.6482, 0.4284, 0.8307, 0.3686, 0.3432, 0.4072, 0.6018, 0.5702], [0.6427, 0.4318, 0.8174, 0.4371, 0.4419, 0.2847, 0.5541, 0.6283], [0.4102, 0.2716, 0.8553, 0.2385, 0.5360, 0.1767, 0.7007, 0.5699], [0.5646, 0.3775, 0.8764, 0.4672, 0.3465, 0.4999, 0.6117, 0.5606]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986], [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466], [0.6197, 0.4118, 0.8687, 0.5517, 0.4038, 0.5233, 0.5875, 0.5600], [0.6289, 0.4031, 0.8419, 0.5446, 0.4075, 0.5017, 0.6313, 0.5117], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6353, 0.4128, 0.8487, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6062, 0.5617]]], device='cuda:0') loss_train_step before backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0016, device='cuda:0', grad_fn=) loss_train: 0.03189117752481252 step: 48 running loss: 0.0006643995317669275 Train Steps: 48/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136], [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350], [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167], [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5837, 0.5583], [0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400], [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367], [ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6015, 0.4091, 0.6792, 0.2112, 0.3803, 0.2605, 0.5657, 0.5407], [ 0.6865, 0.4520, 0.8424, 0.5097, 0.4201, 0.4669, 0.5732, 0.5527], [ 0.6460, 0.4397, 0.7416, 0.2950, 0.3965, 0.2750, 0.6165, 0.6558], [ 0.6428, 0.4430, 0.8711, 0.5092, 0.4494, 0.5808, 0.6035, 0.5408], [ 0.6505, 0.4395, 0.8684, 0.4644, 0.4637, 0.5359, 0.6041, 0.5740], [ 0.6279, 0.4122, 0.8126, 0.2524, 0.4543, 0.1920, 0.6404, 0.5525], [ 0.6940, 0.4552, 0.8476, 0.3173, 0.3913, 0.2747, 0.6630, 0.5650], [-0.0232, -0.0028, 0.7549, 0.2287, 0.4013, 0.2235, 0.5537, 0.5746]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6043, 0.4022, 0.6888, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136], [0.6126, 0.3954, 0.8537, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350], [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167], [0.6155, 0.4124, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5838, 0.5583], [0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400], [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367], [0.0000, 0.0000, 0.7425, 0.2117, 0.3938, 0.2433, 0.5437, 0.5567]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.032499423483386636 step: 49 running loss: 0.0006632535404772783 Train Steps: 49/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6263, 0.4233, 0.7924, 0.4626, 0.3788, 0.2883, 0.5573, 0.6047], [0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050], [0.6122, 0.3993, 0.8738, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927], [0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5887, 0.5017], [0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5563, 0.5533], [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142], [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899], [ nan, nan, 0.8850, 0.3000, 0.5363, 0.2250, 0.7343, 0.5771]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6716, 0.4577, 0.7748, 0.4710, 0.3777, 0.2858, 0.5700, 0.6375], [0.6477, 0.4321, 0.8409, 0.5273, 0.4271, 0.5287, 0.5504, 0.5211], [0.6724, 0.4409, 0.8604, 0.4746, 0.4422, 0.5025, 0.5418, 0.5029], [0.6832, 0.4455, 0.8589, 0.4731, 0.3588, 0.5001, 0.5931, 0.5311], [0.6424, 0.4260, 0.8493, 0.5241, 0.3827, 0.4609, 0.5877, 0.5796], [0.6399, 0.4265, 0.8529, 0.4411, 0.3920, 0.4770, 0.5210, 0.5491], [0.6415, 0.4153, 0.8961, 0.4563, 0.3989, 0.4180, 0.7343, 0.6127], [0.0013, 0.0094, 0.9064, 0.2967, 0.5239, 0.2192, 0.7298, 0.5917]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6263, 0.4232, 0.7924, 0.4626, 0.3787, 0.2883, 0.5573, 0.6047], [0.6097, 0.3988, 0.8650, 0.5250, 0.4212, 0.5200, 0.5675, 0.5050], [0.6122, 0.3993, 0.8737, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927], [0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5888, 0.5017], [0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5562, 0.5533], [0.6185, 0.4067, 0.8838, 0.4450, 0.4038, 0.4733, 0.5213, 0.5142], [0.6346, 0.4144, 0.9087, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899], [0.0000, 0.0000, 0.8850, 0.3000, 0.5362, 0.2250, 0.7343, 0.5771]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.03302222432103008 step: 50 running loss: 0.0006604444864206017 Train Steps: 50/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5988, 0.5700], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [ nan, nan, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702], [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5637, 0.5083], [0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006], [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540], [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6350, 0.4143, 0.7841, 0.2905, 0.3769, 0.3715, 0.5888, 0.5981], [0.6583, 0.4429, 0.8623, 0.4535, 0.3923, 0.4489, 0.5495, 0.5690], [0.0326, 0.0221, 0.9258, 0.3567, 0.5297, 0.2380, 0.7122, 0.5763], [0.6316, 0.4160, 0.8865, 0.4939, 0.3749, 0.4176, 0.6057, 0.5378], [0.5989, 0.4009, 0.8659, 0.5195, 0.3898, 0.4810, 0.5427, 0.5191], [0.6340, 0.4147, 0.8485, 0.5868, 0.3658, 0.4756, 0.6203, 0.5119], [0.6492, 0.4303, 0.9090, 0.4505, 0.4247, 0.3199, 0.6731, 0.5689], [0.7051, 0.4617, 0.7650, 0.2274, 0.4712, 0.1993, 0.5996, 0.5446]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5987, 0.5700], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.0000, 0.0000, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702], [0.6236, 0.3965, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5638, 0.5083], [0.6310, 0.4017, 0.8562, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006], [0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540], [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.03343431546818465 step: 51 running loss: 0.0006555748131016598 Train Steps: 51/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165], [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050], [0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071], [0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283], [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5903, 0.3858, 0.8913, 0.3479, 0.4020, 0.3025, 0.5814, 0.5338], [0.6145, 0.3955, 0.9317, 0.4338, 0.3939, 0.4503, 0.6172, 0.5075], [0.5838, 0.3638, 0.8573, 0.3198, 0.3734, 0.3658, 0.6123, 0.5476], [0.6290, 0.4002, 0.8091, 0.3085, 0.3937, 0.3147, 0.5827, 0.5231], [0.6167, 0.4136, 0.7696, 0.2070, 0.4055, 0.2694, 0.5670, 0.5264], [0.5940, 0.3831, 0.8313, 0.5702, 0.4039, 0.4745, 0.7099, 0.5393], [0.6499, 0.4372, 0.9018, 0.4441, 0.3836, 0.3688, 0.5787, 0.5647], [0.6392, 0.4080, 0.7350, 0.1964, 0.4205, 0.2529, 0.5691, 0.5825]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165], [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6135, 0.3994, 0.7912, 0.3050, 0.3625, 0.3050, 0.5838, 0.5050], [0.6163, 0.4114, 0.7650, 0.2017, 0.3762, 0.2867, 0.5631, 0.5071], [0.6311, 0.3998, 0.7975, 0.5767, 0.3837, 0.4850, 0.7327, 0.5343], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283], [0.6108, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.03381663607433438 step: 52 running loss: 0.0006503199245064304 Train Steps: 52/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933], [0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433], [0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773], [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6216, 0.4100, 0.8862, 0.5867, 0.4946, 0.4252, 0.5894, 0.5776], [0.5798, 0.3730, 0.8235, 0.2799, 0.3553, 0.3580, 0.5904, 0.5176], [0.5872, 0.3870, 0.6935, 0.2447, 0.4144, 0.2064, 0.5287, 0.5627], [0.5558, 0.3519, 0.9112, 0.5062, 0.3737, 0.4437, 0.5028, 0.5373], [0.5895, 0.3918, 0.8638, 0.3317, 0.3426, 0.3596, 0.5723, 0.5447], [0.5498, 0.3511, 0.7831, 0.2351, 0.3763, 0.3139, 0.6081, 0.5010], [0.5845, 0.3780, 0.9155, 0.5267, 0.3823, 0.3203, 0.6370, 0.4975], [0.5984, 0.3738, 0.8927, 0.2421, 0.5714, 0.2009, 0.7444, 0.5272]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933], [0.6175, 0.4091, 0.7862, 0.2800, 0.3638, 0.3583, 0.6187, 0.5433], [0.6147, 0.4026, 0.6600, 0.2467, 0.4087, 0.2150, 0.5489, 0.5773], [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.03452799213118851 step: 53 running loss: 0.0006514715496450663 Train Steps: 53/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986], [0.6092, 0.4001, 0.8638, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064], [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431], [0.6218, 0.4098, 0.7238, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350], [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5949, 0.3693, 0.9166, 0.3965, 0.4438, 0.2190, 0.6329, 0.4888], [0.5678, 0.3612, 0.8888, 0.4901, 0.4240, 0.5450, 0.5531, 0.5003], [0.5673, 0.3601, 0.8953, 0.4152, 0.3538, 0.3924, 0.5423, 0.4920], [0.6909, 0.4426, 0.7072, 0.2401, 0.4204, 0.1919, 0.5422, 0.5530], [0.5453, 0.3431, 0.8851, 0.4871, 0.4364, 0.5039, 0.5110, 0.5051], [0.6636, 0.4255, 0.7338, 0.2045, 0.4271, 0.2490, 0.6436, 0.5372], [0.5560, 0.3329, 0.8886, 0.5402, 0.3658, 0.4380, 0.6252, 0.4615], [0.6067, 0.3854, 0.8926, 0.4606, 0.3783, 0.4371, 0.5586, 0.5382]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986], [0.6092, 0.4001, 0.8637, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064], [0.6082, 0.4024, 0.8737, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136], [0.6154, 0.4112, 0.7038, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431], [0.6218, 0.4098, 0.7237, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350], [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.035225964384153485 step: 54 running loss: 0.0006523326737806201 Train Steps: 54/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153], [0.6059, 0.4002, 0.7562, 0.2767, 0.3538, 0.3033, 0.5529, 0.5455], [0.6173, 0.4114, 0.7325, 0.2500, 0.4213, 0.1917, 0.5338, 0.5700], [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6058, 0.3986, 0.8324, 0.4626, 0.3838, 0.4983, 0.5147, 0.5466], [0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267], [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5942, 0.3785, 0.8978, 0.5211, 0.4668, 0.5194, 0.6236, 0.4956], [0.5834, 0.3925, 0.7657, 0.2807, 0.3593, 0.2938, 0.5746, 0.5117], [0.6970, 0.4616, 0.7445, 0.2467, 0.4266, 0.1788, 0.5442, 0.5388], [0.5240, 0.3442, 0.8705, 0.4321, 0.3640, 0.5122, 0.5731, 0.4977], [0.5492, 0.3548, 0.8602, 0.3845, 0.3575, 0.3699, 0.6041, 0.5329], [0.5581, 0.3591, 0.8418, 0.4721, 0.3978, 0.4798, 0.5361, 0.5162], [0.6528, 0.4251, 0.8704, 0.2919, 0.4347, 0.2784, 0.6402, 0.5054], [0.6352, 0.3991, 0.9248, 0.4779, 0.3932, 0.5056, 0.6813, 0.4867]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153], [0.6059, 0.4002, 0.7563, 0.2767, 0.3537, 0.3033, 0.5529, 0.5455], [0.6173, 0.4114, 0.7325, 0.2500, 0.4212, 0.1917, 0.5337, 0.5700], [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6058, 0.3986, 0.8324, 0.4626, 0.3837, 0.4983, 0.5147, 0.5466], [0.6266, 0.4067, 0.8587, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267], [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.03592075948836282 step: 55 running loss: 0.0006531047179702331 Train Steps: 55/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6250, 0.4106, 0.8700, 0.3717, 0.3588, 0.4967, 0.6038, 0.5167], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717], [0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773], [0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6009, 0.3886, 0.8640, 0.3660, 0.3473, 0.4980, 0.6004, 0.5025], [0.6008, 0.3929, 0.8754, 0.4582, 0.3570, 0.5216, 0.5903, 0.4871], [0.6102, 0.4029, 0.8723, 0.4841, 0.4216, 0.4752, 0.5486, 0.5360], [0.6062, 0.4069, 0.6612, 0.2374, 0.3979, 0.2247, 0.5234, 0.5572], [0.5861, 0.3793, 0.8570, 0.3509, 0.3809, 0.2627, 0.6151, 0.4455], [0.6013, 0.3983, 0.8778, 0.4348, 0.4101, 0.5217, 0.5876, 0.5433], [0.6018, 0.3773, 0.8865, 0.4585, 0.3473, 0.3775, 0.6076, 0.4694], [0.6657, 0.4308, 0.8050, 0.2541, 0.4369, 0.2434, 0.6707, 0.5185]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6250, 0.4105, 0.8700, 0.3717, 0.3587, 0.4967, 0.6037, 0.5167], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5412, 0.5717], [0.6147, 0.4026, 0.6600, 0.2467, 0.4087, 0.2150, 0.5489, 0.5773], [0.6277, 0.4036, 0.8687, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.6189, 0.4049, 0.8888, 0.4417, 0.4212, 0.5200, 0.5987, 0.5633], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.03618784763966687 step: 56 running loss: 0.0006462115649940513 Train Steps: 56/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650], [0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517], [0.6275, 0.4157, 0.8337, 0.5800, 0.3763, 0.4200, 0.5547, 0.6125], [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717], [0.6200, 0.3999, 0.8653, 0.5207, 0.4100, 0.5125, 0.5975, 0.5103], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6214, 0.4112, 0.7838, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.1162, 0.0801, 0.6802, 0.1951, 0.4182, 0.2635, 0.5322, 0.5675], [0.6565, 0.4317, 0.8646, 0.4593, 0.4474, 0.4991, 0.5942, 0.5406], [0.6423, 0.4291, 0.8379, 0.5595, 0.3659, 0.4292, 0.5599, 0.6017], [0.6656, 0.4248, 0.8112, 0.2486, 0.4119, 0.2329, 0.6516, 0.5067], [0.6346, 0.4177, 0.8700, 0.4794, 0.4136, 0.4643, 0.5676, 0.5569], [0.6783, 0.4386, 0.8775, 0.5302, 0.3981, 0.5214, 0.5827, 0.4985], [0.6109, 0.3950, 0.8578, 0.5303, 0.3491, 0.3746, 0.5703, 0.5402], [0.6370, 0.4200, 0.7620, 0.1908, 0.3480, 0.3000, 0.5853, 0.4956]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650], [0.6175, 0.3957, 0.8700, 0.4817, 0.4663, 0.5133, 0.5800, 0.5517], [0.6275, 0.4157, 0.8338, 0.5800, 0.3762, 0.4200, 0.5547, 0.6125], [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5412, 0.5717], [0.6200, 0.3999, 0.8653, 0.5207, 0.4100, 0.5125, 0.5975, 0.5103], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6214, 0.4112, 0.7837, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.036844901071162894 step: 57 running loss: 0.0006464017731782964 Train Steps: 57/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4064, 0.8688, 0.5050, 0.4225, 0.5100, 0.6138, 0.5500], [ nan, nan, 0.8850, 0.2817, 0.5112, 0.2183, 0.7184, 0.5436], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626], [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533], [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413], [0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6270, 0.4258, 0.8462, 0.4879, 0.3983, 0.5099, 0.5488, 0.5659], [0.1573, 0.1195, 0.8400, 0.2562, 0.4836, 0.2424, 0.6582, 0.5480], [0.6321, 0.4317, 0.7994, 0.2551, 0.4191, 0.2894, 0.6537, 0.5727], [0.6228, 0.4126, 0.8361, 0.6025, 0.3768, 0.4785, 0.5576, 0.4819], [0.5831, 0.3975, 0.6814, 0.1729, 0.3564, 0.2846, 0.5472, 0.5592], [0.6615, 0.4358, 0.8663, 0.3155, 0.4706, 0.2533, 0.6759, 0.5524], [0.5915, 0.4250, 0.7890, 0.3555, 0.3250, 0.5123, 0.5517, 0.5376], [0.6395, 0.4154, 0.8463, 0.2810, 0.4706, 0.2473, 0.6771, 0.5393]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6201, 0.4064, 0.8687, 0.5050, 0.4225, 0.5100, 0.6137, 0.5500], [0.0000, 0.0000, 0.8850, 0.2817, 0.5113, 0.2183, 0.7184, 0.5436], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626], [0.6114, 0.4018, 0.7212, 0.1967, 0.3762, 0.2700, 0.5875, 0.5533], [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413], [0.6132, 0.4118, 0.8200, 0.3633, 0.3562, 0.5400, 0.5787, 0.5136], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343]]], device='cuda:0') loss_train_step before backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0013, device='cuda:0', grad_fn=) loss_train: 0.038141772645758465 step: 58 running loss: 0.0006576167697544563 Train Steps: 58/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6111, 0.4033, 0.8300, 0.3267, 0.3588, 0.3333, 0.5444, 0.5637], [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633], [0.6200, 0.4049, 0.8638, 0.5617, 0.4125, 0.5100, 0.6013, 0.5317], [0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6201, 0.4102, 0.7288, 0.2417, 0.4150, 0.2383, 0.6100, 0.5500], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395], [0.6087, 0.3976, 0.8337, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5434, 0.3604, 0.8125, 0.3093, 0.3506, 0.3417, 0.5186, 0.5918], [0.6032, 0.4228, 0.8550, 0.4658, 0.3610, 0.3783, 0.5531, 0.5995], [0.6350, 0.4280, 0.8495, 0.5566, 0.4133, 0.5111, 0.6015, 0.5680], [0.6236, 0.4199, 0.8116, 0.2509, 0.4350, 0.2018, 0.6264, 0.5068], [0.5966, 0.3983, 0.7309, 0.2129, 0.3987, 0.2585, 0.5930, 0.5883], [0.6126, 0.3992, 0.8214, 0.5925, 0.3858, 0.5042, 0.5894, 0.5219], [0.5975, 0.3883, 0.7192, 0.2058, 0.3947, 0.2623, 0.5913, 0.5678], [0.6132, 0.4144, 0.8182, 0.3650, 0.3688, 0.3395, 0.5849, 0.5806]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6111, 0.4033, 0.8300, 0.3267, 0.3587, 0.3333, 0.5444, 0.5637], [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5638, 0.5633], [0.6199, 0.4049, 0.8637, 0.5617, 0.4125, 0.5100, 0.6012, 0.5317], [0.6277, 0.4056, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6201, 0.4102, 0.7287, 0.2417, 0.4150, 0.2383, 0.6100, 0.5500], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395], [0.6087, 0.3976, 0.8338, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.0386340361146722 step: 59 running loss: 0.000654814171435122 Train Steps: 59/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567], [0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995], [0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183], [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633], [0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207], [0.6131, 0.4037, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6260, 0.4173, 0.7693, 0.2477, 0.4896, 0.1354, 0.5837, 0.5601], [0.5821, 0.3951, 0.7618, 0.2472, 0.3904, 0.3180, 0.5785, 0.5780], [0.5168, 0.3570, 0.7499, 0.2007, 0.4158, 0.2174, 0.6111, 0.5105], [0.6055, 0.4041, 0.8060, 0.3109, 0.3993, 0.2829, 0.6012, 0.5408], [0.6270, 0.4130, 0.8786, 0.5441, 0.4146, 0.5860, 0.7014, 0.5775], [0.5542, 0.3892, 0.7627, 0.2894, 0.3791, 0.4210, 0.5780, 0.5523], [0.5910, 0.3917, 0.6935, 0.2902, 0.3789, 0.2751, 0.5223, 0.5864], [0.6000, 0.3858, 0.8019, 0.2483, 0.4373, 0.2467, 0.6337, 0.5463]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6245, 0.4100, 0.7763, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567], [0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995], [0.6201, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183], [0.6293, 0.4024, 0.8750, 0.5000, 0.4013, 0.5733, 0.7121, 0.5633], [0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207], [0.6131, 0.4036, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.03914449064177461 step: 60 running loss: 0.0006524081773629102 Train Steps: 60/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183], [0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319], [0.6126, 0.4039, 0.8237, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138], [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683], [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986], [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6002, 0.3741, 0.7528, 0.2012, 0.4403, 0.1921, 0.5857, 0.5105], [0.5763, 0.3811, 0.7150, 0.3156, 0.3673, 0.3497, 0.5242, 0.5349], [0.6692, 0.4403, 0.8189, 0.4075, 0.3779, 0.3896, 0.6108, 0.6077], [0.5852, 0.3847, 0.8953, 0.3522, 0.4302, 0.3928, 0.7393, 0.5533], [0.5822, 0.3798, 0.8676, 0.5529, 0.3981, 0.3941, 0.5389, 0.5724], [0.5916, 0.3839, 0.7662, 0.2230, 0.4605, 0.1916, 0.6009, 0.5561], [0.5540, 0.3630, 0.8852, 0.3655, 0.4472, 0.2216, 0.6218, 0.5047], [0.6438, 0.4159, 0.8962, 0.4108, 0.3776, 0.4650, 0.6316, 0.4925]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6213, 0.4001, 0.7713, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183], [0.6109, 0.4041, 0.6975, 0.3167, 0.3512, 0.3383, 0.5153, 0.5319], [0.6126, 0.4038, 0.8238, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138], [0.6332, 0.4165, 0.9100, 0.3350, 0.4187, 0.3683, 0.7438, 0.5528], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5412, 0.5683], [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986], [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.039610637846635655 step: 61 running loss: 0.0006493547187973058 Train Steps: 61/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274], [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064], [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6262, 0.5367], [0.6332, 0.4118, 0.9238, 0.4267, 0.4012, 0.4733, 0.7525, 0.5436], [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6490, 0.4308, 0.8600, 0.3135, 0.4776, 0.1786, 0.6367, 0.5065], [0.5685, 0.3786, 0.8593, 0.4983, 0.4870, 0.5088, 0.5418, 0.4965], [0.5677, 0.3706, 0.8188, 0.2409, 0.4533, 0.2450, 0.7117, 0.5441], [0.6425, 0.4133, 0.8852, 0.4869, 0.4178, 0.5740, 0.6153, 0.5150], [0.6038, 0.3772, 0.9128, 0.4450, 0.4081, 0.4684, 0.7507, 0.5351], [0.6241, 0.3928, 0.8778, 0.4593, 0.4418, 0.5083, 0.5830, 0.5158], [0.6153, 0.3958, 0.8522, 0.3992, 0.3652, 0.3736, 0.5988, 0.5468], [0.5823, 0.3788, 0.8807, 0.3593, 0.3960, 0.2584, 0.6143, 0.5197]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274], [0.6102, 0.4005, 0.8687, 0.5100, 0.4812, 0.5400, 0.5404, 0.5064], [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6263, 0.5367], [0.6332, 0.4118, 0.9237, 0.4267, 0.4013, 0.4733, 0.7525, 0.5436], [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5838, 0.5500], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6187, 0.5283]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.039949122903635725 step: 62 running loss: 0.0006443406919941246 Train Steps: 62/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6250, 0.4110, 0.7238, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633], [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717], [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6489, 0.4172, 0.8604, 0.5817, 0.3890, 0.5051, 0.6722, 0.5223], [0.5758, 0.3802, 0.7537, 0.2132, 0.4380, 0.1834, 0.5846, 0.5585], [0.6042, 0.4036, 0.8379, 0.2920, 0.4279, 0.2024, 0.5808, 0.5277], [0.6568, 0.4249, 0.9017, 0.5003, 0.4414, 0.4782, 0.5668, 0.5598], [0.3651, 0.2423, 0.7194, 0.2475, 0.4385, 0.2289, 0.5711, 0.5761], [0.6485, 0.4133, 0.9225, 0.3724, 0.4294, 0.2090, 0.6064, 0.4847], [0.5770, 0.3819, 0.7592, 0.1862, 0.4164, 0.2548, 0.6278, 0.5392], [0.6558, 0.4108, 0.8697, 0.5434, 0.4043, 0.5355, 0.7380, 0.5651]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6250, 0.4110, 0.7237, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633], [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411], [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5412, 0.5717], [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6273, 0.4110, 0.8900, 0.3817, 0.4187, 0.2167, 0.5858, 0.4835], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795]]], device='cuda:0') loss_train_step before backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train: 0.04169243582873605 step: 63 running loss: 0.000661784695694223 Train Steps: 63/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6317, 0.4038, 0.8287, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183], [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817], [0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136], [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378], [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6751, 0.4236, 0.8554, 0.5983, 0.3808, 0.4436, 0.6582, 0.5019], [0.6391, 0.4155, 0.8110, 0.3076, 0.3625, 0.3694, 0.5915, 0.5181], [0.6277, 0.4008, 0.9097, 0.3904, 0.4020, 0.2135, 0.5999, 0.5297], [0.6436, 0.3957, 0.8956, 0.4783, 0.4681, 0.5160, 0.6527, 0.5159], [0.5659, 0.3881, 0.8140, 0.3132, 0.4450, 0.2331, 0.6016, 0.5673], [0.6419, 0.4246, 0.8367, 0.3678, 0.3640, 0.4855, 0.6123, 0.5136], [0.6364, 0.4119, 0.9123, 0.4822, 0.4031, 0.3257, 0.7460, 0.5331], [0.6155, 0.3943, 0.8747, 0.4187, 0.4091, 0.5571, 0.5951, 0.5081]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6317, 0.4038, 0.8288, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986], [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103], [0.6226, 0.4098, 0.8913, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6251, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6087, 0.5183], [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817], [0.6132, 0.4118, 0.8200, 0.3633, 0.3562, 0.5400, 0.5787, 0.5136], [0.6307, 0.4029, 0.8988, 0.4817, 0.3938, 0.3500, 0.7311, 0.5378], [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.04215108629432507 step: 64 running loss: 0.0006586107233488292 Train Steps: 64/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833], [0.6273, 0.4105, 0.8988, 0.4517, 0.3912, 0.2550, 0.5894, 0.4811], [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667], [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6188, 0.5433], [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6841, 0.4391, 0.8785, 0.4640, 0.3834, 0.4200, 0.5930, 0.5421], [0.6607, 0.4358, 0.8822, 0.5192, 0.4290, 0.5394, 0.6074, 0.5179], [0.6769, 0.4259, 0.8845, 0.5061, 0.3831, 0.4829, 0.6011, 0.4728], [0.6597, 0.4213, 0.9095, 0.4465, 0.4110, 0.2541, 0.5965, 0.4808], [0.0435, 0.0366, 0.7480, 0.2466, 0.3787, 0.2960, 0.6180, 0.5568], [0.7205, 0.4638, 0.9050, 0.4852, 0.3870, 0.5248, 0.6364, 0.5229], [0.7102, 0.4620, 0.9185, 0.4474, 0.4126, 0.3109, 0.6806, 0.5438], [0.7577, 0.4880, 0.9010, 0.5120, 0.4162, 0.5258, 0.6563, 0.4690]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6037, 0.4833], [0.6273, 0.4105, 0.8988, 0.4517, 0.3913, 0.2550, 0.5894, 0.4811], [0.0000, 0.0000, 0.7525, 0.2291, 0.3837, 0.3017, 0.6050, 0.5667], [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6187, 0.5433], [0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758]]], device='cuda:0') loss_train_step before backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0011, device='cuda:0', grad_fn=) loss_train: 0.04324785657809116 step: 65 running loss: 0.0006653516396629409 Train Steps: 65/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6286, 0.4060, 0.9188, 0.4333, 0.3675, 0.4167, 0.7034, 0.5528], [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6197, 0.4118, 0.8688, 0.5517, 0.4037, 0.5233, 0.5875, 0.5600], [0.6261, 0.3987, 0.8688, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309], [0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6728, 0.4450, 0.9098, 0.4187, 0.3732, 0.4188, 0.6945, 0.5405], [0.6582, 0.4256, 0.9019, 0.4799, 0.3953, 0.5400, 0.7138, 0.5375], [0.6464, 0.4291, 0.7980, 0.2633, 0.4291, 0.1830, 0.5571, 0.5232], [0.6475, 0.4214, 0.8394, 0.5691, 0.3625, 0.5164, 0.6335, 0.5250], [0.6215, 0.4021, 0.8771, 0.5360, 0.3976, 0.5311, 0.5583, 0.5489], [0.6519, 0.4227, 0.8778, 0.4887, 0.4175, 0.5327, 0.6750, 0.5320], [0.6482, 0.4330, 0.8629, 0.2958, 0.4095, 0.2383, 0.6013, 0.5273], [0.6709, 0.4361, 0.7736, 0.2770, 0.4107, 0.2393, 0.5732, 0.5426]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6286, 0.4060, 0.9187, 0.4333, 0.3675, 0.4167, 0.7034, 0.5528], [0.6339, 0.4102, 0.9087, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6197, 0.4118, 0.8687, 0.5517, 0.4038, 0.5233, 0.5875, 0.5600], [0.6261, 0.3987, 0.8687, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309], [0.6250, 0.4131, 0.8687, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6229, 0.4066, 0.7613, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.043584351253230125 step: 66 running loss: 0.0006603689583822746 Train Steps: 66/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6205, 0.4081, 0.8950, 0.4017, 0.3788, 0.4700, 0.5963, 0.5667], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5787, 0.5550], [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082], [0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6228, 0.4119, 0.7938, 0.2233, 0.4674, 0.1773, 0.6188, 0.5433]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6590, 0.4183, 0.7758, 0.2653, 0.3696, 0.3135, 0.6233, 0.5215], [0.6834, 0.4374, 0.8891, 0.4067, 0.3751, 0.4616, 0.6058, 0.5631], [0.6540, 0.4188, 0.8939, 0.4728, 0.4269, 0.5779, 0.6053, 0.5424], [0.6287, 0.4122, 0.8230, 0.2927, 0.3532, 0.3614, 0.5817, 0.5574], [0.6505, 0.4088, 0.8768, 0.5240, 0.4169, 0.5441, 0.6076, 0.5048], [0.6724, 0.4390, 0.8796, 0.4794, 0.3679, 0.4075, 0.5526, 0.5108], [0.6069, 0.3898, 0.8456, 0.5323, 0.4040, 0.5576, 0.7046, 0.5654], [0.7121, 0.4588, 0.8121, 0.2477, 0.4546, 0.1666, 0.6290, 0.5217]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6205, 0.4081, 0.8950, 0.4017, 0.3787, 0.4700, 0.5962, 0.5667], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5788, 0.5550], [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082], [0.6108, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6228, 0.4119, 0.7937, 0.2233, 0.4674, 0.1773, 0.6187, 0.5433]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.04411605268251151 step: 67 running loss: 0.0006584485475001718 Train Steps: 67/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131], [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413], [0.6199, 0.4065, 0.7598, 0.2385, 0.4317, 0.1981, 0.5933, 0.5221], [0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995], [0.6145, 0.4007, 0.8775, 0.4533, 0.4562, 0.5533, 0.6088, 0.5533], [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329], [0.6075, 0.4007, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6283, 0.4003, 0.8561, 0.3963, 0.3301, 0.3903, 0.6070, 0.5052], [0.6088, 0.3770, 0.9110, 0.3335, 0.4759, 0.2409, 0.7457, 0.5498], [0.6425, 0.4072, 0.7553, 0.2335, 0.4060, 0.2231, 0.5940, 0.5234], [0.6522, 0.4164, 0.7659, 0.1988, 0.3875, 0.2213, 0.6210, 0.5014], [0.5815, 0.3730, 0.8650, 0.4474, 0.4370, 0.5787, 0.6060, 0.5581], [0.6791, 0.4359, 0.8828, 0.4949, 0.3584, 0.4545, 0.5194, 0.5556], [0.6427, 0.4100, 0.8443, 0.3498, 0.3293, 0.3790, 0.6097, 0.5322], [0.6070, 0.3958, 0.8130, 0.4942, 0.4050, 0.5089, 0.5191, 0.5420]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131], [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413], [0.6199, 0.4065, 0.7598, 0.2385, 0.4317, 0.1981, 0.5933, 0.5221], [0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995], [0.6145, 0.4007, 0.8775, 0.4533, 0.4563, 0.5533, 0.6087, 0.5533], [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329], [0.6075, 0.4006, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.044399865000741556 step: 68 running loss: 0.0006529391911873758 Train Steps: 68/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.6205, 0.4081, 0.8950, 0.4017, 0.3788, 0.4700, 0.5963, 0.5667], [ nan, nan, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717], [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058], [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6684, 0.4103, 0.8983, 0.4467, 0.3564, 0.4676, 0.6509, 0.4979], [0.6697, 0.4285, 0.8877, 0.3976, 0.3835, 0.4953, 0.6155, 0.5776], [0.1189, 0.0600, 0.6698, 0.2783, 0.3735, 0.2384, 0.5544, 0.5748], [0.6611, 0.4208, 0.7464, 0.2118, 0.4308, 0.2101, 0.6222, 0.5398], [0.6449, 0.4069, 0.7567, 0.2398, 0.3928, 0.2952, 0.5959, 0.5886], [0.6412, 0.4086, 0.8399, 0.4462, 0.3675, 0.4937, 0.5392, 0.5161], [0.6842, 0.4399, 0.8795, 0.5051, 0.3818, 0.4660, 0.5229, 0.5636], [0.6966, 0.4450, 0.7876, 0.2467, 0.4578, 0.2063, 0.6285, 0.5483]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.6205, 0.4081, 0.8950, 0.4017, 0.3787, 0.4700, 0.5962, 0.5667], [0.0000, 0.0000, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717], [0.6084, 0.4005, 0.8400, 0.4317, 0.3762, 0.4750, 0.5476, 0.5058], [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.04514229335472919 step: 69 running loss: 0.0006542361355757854 Train Steps: 69/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033], [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6362, 0.5200], [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332], [0.6216, 0.4167, 0.8588, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667], [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400], [0.6268, 0.4061, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300], [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6063, 0.5567], [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5932, 0.3854, 0.8746, 0.4924, 0.4156, 0.5271, 0.5691, 0.5149], [0.6065, 0.3985, 0.8727, 0.3528, 0.3805, 0.4974, 0.6261, 0.5165], [0.5919, 0.3854, 0.8122, 0.5360, 0.3914, 0.4945, 0.6972, 0.5451], [0.5852, 0.3936, 0.8503, 0.5523, 0.4109, 0.5134, 0.5696, 0.5734], [0.5774, 0.3899, 0.8421, 0.3239, 0.3606, 0.4490, 0.6091, 0.5548], [0.6649, 0.4226, 0.8385, 0.2488, 0.4642, 0.1997, 0.6237, 0.5406], [0.5808, 0.3709, 0.8631, 0.4307, 0.4711, 0.4975, 0.5966, 0.5669], [0.5930, 0.3836, 0.8309, 0.5664, 0.4003, 0.4188, 0.5687, 0.5385]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033], [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6363, 0.5200], [0.6289, 0.4019, 0.8112, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332], [0.6216, 0.4167, 0.8587, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667], [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400], [0.6268, 0.4060, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300], [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6062, 0.5567], [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.045491363474866375 step: 70 running loss: 0.0006498766210695197 Train Steps: 70/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6063, 0.5617], [0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183], [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366], [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650], [0.6198, 0.4130, 0.8762, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103], [0.6371, 0.4092, 0.8337, 0.5850, 0.3950, 0.5117, 0.6559, 0.5262], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750], [0.6224, 0.4097, 0.7438, 0.2267, 0.3850, 0.2850, 0.5988, 0.5250]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6027, 0.3895, 0.9023, 0.4631, 0.3711, 0.4540, 0.5978, 0.5551], [0.5901, 0.3676, 0.8123, 0.2965, 0.3895, 0.2630, 0.5769, 0.5181], [0.5757, 0.3723, 0.7930, 0.5494, 0.3922, 0.4865, 0.6760, 0.5425], [0.5815, 0.3743, 0.8654, 0.4666, 0.4310, 0.4858, 0.5209, 0.5583], [0.6159, 0.4109, 0.8653, 0.4103, 0.3742, 0.4880, 0.5513, 0.5205], [0.6212, 0.4156, 0.8455, 0.5615, 0.4003, 0.5054, 0.6516, 0.5298], [0.6269, 0.3967, 0.7233, 0.2441, 0.3980, 0.2832, 0.6001, 0.5731], [0.5960, 0.3751, 0.7550, 0.2161, 0.4011, 0.2795, 0.5930, 0.5292]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6062, 0.5617], [0.6201, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183], [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366], [0.6098, 0.3991, 0.8637, 0.4717, 0.4263, 0.4967, 0.5213, 0.5650], [0.6198, 0.4130, 0.8763, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103], [0.6371, 0.4092, 0.8338, 0.5850, 0.3950, 0.5117, 0.6559, 0.5262], [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750], [0.6224, 0.4097, 0.7437, 0.2267, 0.3850, 0.2850, 0.5987, 0.5250]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.04574897524435073 step: 71 running loss: 0.0006443517640049399 Train Steps: 71/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461], [0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693], [0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517], [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417], [0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5637, 0.5083], [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183], [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611], [0.6261, 0.4066, 0.8325, 0.2150, 0.4763, 0.2667, 0.7002, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5633, 0.3836, 0.8420, 0.5337, 0.4440, 0.5154, 0.5364, 0.5381], [0.5384, 0.3566, 0.8358, 0.4855, 0.3616, 0.4613, 0.5191, 0.4777], [0.5452, 0.3580, 0.8498, 0.4636, 0.4535, 0.5147, 0.5744, 0.5431], [0.5719, 0.3858, 0.7744, 0.2861, 0.3919, 0.3032, 0.6182, 0.5462], [0.5891, 0.3935, 0.8548, 0.5052, 0.3834, 0.4553, 0.5580, 0.5031], [0.5469, 0.3664, 0.7412, 0.2916, 0.4010, 0.3196, 0.6024, 0.6164], [0.5542, 0.3592, 0.8591, 0.4615, 0.4228, 0.4564, 0.5400, 0.5590], [0.5844, 0.3841, 0.8278, 0.2302, 0.4739, 0.2515, 0.7062, 0.5385]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461], [0.6038, 0.3946, 0.8413, 0.4883, 0.3562, 0.4550, 0.5266, 0.4693], [0.6175, 0.3957, 0.8700, 0.4817, 0.4663, 0.5133, 0.5800, 0.5517], [0.6243, 0.4128, 0.7763, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417], [0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5638, 0.5083], [0.6196, 0.4094, 0.7563, 0.2817, 0.3938, 0.3183, 0.6012, 0.6183], [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611], [0.6261, 0.4066, 0.8325, 0.2150, 0.4762, 0.2667, 0.7002, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.04638030182104558 step: 72 running loss: 0.0006441708586256331 Train Steps: 72/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6136, 0.4117, 0.8700, 0.5167, 0.4188, 0.5083, 0.5147, 0.5495], [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333], [0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220], [0.6353, 0.4128, 0.8488, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123], [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5476, 0.3818, 0.8540, 0.5142, 0.4185, 0.5273, 0.5357, 0.5329], [0.5659, 0.3972, 0.8821, 0.4454, 0.3786, 0.5081, 0.6200, 0.5237], [0.5426, 0.3779, 0.8634, 0.3470, 0.3503, 0.4327, 0.5906, 0.5590], [0.5297, 0.3636, 0.7151, 0.2196, 0.4366, 0.1792, 0.5518, 0.5310], [0.6560, 0.4389, 0.8403, 0.2511, 0.5428, 0.1872, 0.6646, 0.5584], [0.5384, 0.3503, 0.8167, 0.5637, 0.4109, 0.4789, 0.5900, 0.5254], [0.5101, 0.3595, 0.7190, 0.2721, 0.3746, 0.2856, 0.5514, 0.5176], [0.5184, 0.3575, 0.7436, 0.2908, 0.4046, 0.3418, 0.6030, 0.6151]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6136, 0.4117, 0.8700, 0.5167, 0.4187, 0.5083, 0.5147, 0.5495], [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6137, 0.5333], [0.6219, 0.4097, 0.8737, 0.3400, 0.3562, 0.4117, 0.5975, 0.5683], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220], [0.6353, 0.4128, 0.8487, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540], [0.6222, 0.3937, 0.8350, 0.5617, 0.4137, 0.4600, 0.5800, 0.5233], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123], [0.6196, 0.4094, 0.7563, 0.2817, 0.3938, 0.3183, 0.6012, 0.6183]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.04742685717064887 step: 73 running loss: 0.0006496829749403955 Train Steps: 73/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200], [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633], [0.6339, 0.4118, 0.7988, 0.5800, 0.3912, 0.4583, 0.7343, 0.5760], [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5360, 0.3843, 0.8365, 0.4999, 0.4472, 0.5498, 0.5199, 0.5358], [0.5817, 0.4067, 0.7153, 0.1968, 0.3909, 0.2734, 0.5824, 0.5477], [0.6057, 0.4180, 0.7590, 0.2297, 0.4045, 0.2917, 0.6042, 0.5432], [0.5751, 0.4026, 0.7735, 0.2757, 0.4548, 0.1920, 0.5601, 0.5316], [0.5739, 0.4009, 0.8505, 0.3059, 0.4595, 0.2439, 0.5841, 0.5173], [0.5428, 0.3725, 0.8629, 0.4832, 0.4021, 0.5831, 0.6896, 0.5574], [0.5653, 0.3907, 0.7998, 0.5568, 0.3800, 0.4650, 0.6964, 0.5689], [0.5964, 0.4303, 0.8868, 0.4445, 0.3831, 0.4982, 0.6010, 0.5317]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6114, 0.4018, 0.7212, 0.1967, 0.3762, 0.2700, 0.5875, 0.5533], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200], [0.6293, 0.4024, 0.8750, 0.5000, 0.4013, 0.5733, 0.7121, 0.5633], [0.6339, 0.4118, 0.7987, 0.5800, 0.3913, 0.4583, 0.7343, 0.5760], [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6137, 0.5333]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.047969518578611314 step: 74 running loss: 0.0006482367375488015 Train Steps: 74/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6126, 0.4067, 0.8638, 0.5383, 0.4188, 0.4850, 0.5016, 0.5392], [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483], [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350], [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6268, 0.4061, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300], [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5686, 0.4101, 0.8524, 0.5625, 0.4094, 0.4891, 0.5374, 0.5442], [0.5875, 0.4116, 0.7930, 0.2905, 0.4642, 0.1922, 0.5773, 0.5451], [0.5920, 0.3973, 0.8621, 0.3971, 0.3795, 0.4870, 0.6047, 0.5376], [0.5645, 0.3967, 0.7930, 0.2855, 0.4020, 0.2773, 0.5988, 0.5453], [0.5812, 0.3969, 0.8747, 0.3446, 0.3547, 0.4323, 0.6985, 0.5203], [0.5914, 0.3975, 0.8695, 0.4539, 0.4608, 0.5580, 0.6136, 0.5605], [0.5825, 0.3929, 0.8245, 0.2393, 0.4464, 0.2194, 0.6340, 0.5284], [0.5679, 0.3847, 0.8464, 0.4783, 0.4435, 0.5161, 0.5504, 0.5069]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6126, 0.4067, 0.8637, 0.5383, 0.4187, 0.4850, 0.5016, 0.5392], [0.6229, 0.4107, 0.8138, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483], [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350], [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285], [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6268, 0.4060, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300], [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.04832018641172908 step: 75 running loss: 0.0006442691521563878 Train Steps: 75/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6286, 0.4060, 0.9188, 0.4333, 0.3675, 0.4167, 0.7034, 0.5528], [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082], [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301], [ nan, nan, 0.8300, 0.3150, 0.3588, 0.3383, 0.5208, 0.5194], [0.6206, 0.4001, 0.8900, 0.3933, 0.3588, 0.3567, 0.5837, 0.5083], [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896], [ nan, nan, 0.8525, 0.2217, 0.5413, 0.2367, 0.7367, 0.5482], [0.6218, 0.4137, 0.7263, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6626, 0.4414, 0.9109, 0.4295, 0.3842, 0.4193, 0.7023, 0.5545], [0.6793, 0.4529, 0.8732, 0.5148, 0.4264, 0.5669, 0.6052, 0.5096], [0.6709, 0.4413, 0.8750, 0.5038, 0.4249, 0.5001, 0.6196, 0.5342], [0.2023, 0.1378, 0.8027, 0.3239, 0.3842, 0.3492, 0.5233, 0.5351], [0.6833, 0.4441, 0.8951, 0.4070, 0.3554, 0.3622, 0.5769, 0.4957], [0.6971, 0.4723, 0.8174, 0.3886, 0.3557, 0.3141, 0.5261, 0.5929], [0.1040, 0.0812, 0.8375, 0.2226, 0.5503, 0.2352, 0.7235, 0.5569], [0.6915, 0.4451, 0.7427, 0.2376, 0.4066, 0.2849, 0.5986, 0.5743]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6286, 0.4060, 0.9187, 0.4333, 0.3675, 0.4167, 0.7034, 0.5528], [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082], [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301], [0.0000, 0.0000, 0.8300, 0.3150, 0.3587, 0.3383, 0.5208, 0.5194], [0.6206, 0.4001, 0.8900, 0.3933, 0.3587, 0.3567, 0.5838, 0.5083], [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896], [0.0000, 0.0000, 0.8525, 0.2217, 0.5412, 0.2367, 0.7367, 0.5482], [0.6218, 0.4137, 0.7262, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783]]], device='cuda:0') loss_train_step before backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0019, device='cuda:0', grad_fn=) loss_train: 0.05019753027590923 step: 76 running loss: 0.0006604938194198583 Train Steps: 76/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667], [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519], [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5988, 0.5700], [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517], [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6749, 0.4230, 0.8694, 0.3523, 0.3925, 0.2650, 0.6397, 0.4864], [0.0600, 0.0301, 0.7583, 0.2349, 0.3915, 0.3052, 0.6079, 0.5664], [0.6410, 0.4068, 0.8578, 0.5020, 0.4575, 0.5220, 0.5322, 0.5538], [0.6767, 0.4299, 0.8070, 0.2836, 0.3702, 0.3566, 0.6059, 0.5776], [0.6658, 0.4451, 0.7451, 0.1958, 0.4262, 0.2458, 0.6349, 0.5657], [0.6659, 0.4249, 0.8828, 0.5640, 0.3632, 0.3914, 0.6147, 0.4700], [0.6506, 0.4232, 0.8870, 0.4793, 0.3805, 0.4489, 0.5319, 0.5572], [0.6846, 0.4381, 0.8686, 0.3619, 0.3694, 0.3212, 0.6116, 0.5322]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6277, 0.4036, 0.8687, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742], [0.0000, 0.0000, 0.7525, 0.2291, 0.3837, 0.3017, 0.6050, 0.5667], [0.6100, 0.4016, 0.8600, 0.5067, 0.4613, 0.5233, 0.5086, 0.5519], [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5987, 0.5700], [0.6200, 0.4071, 0.7337, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517], [0.6272, 0.4071, 0.8737, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.05063753016293049 step: 77 running loss: 0.0006576302618562401 Train Steps: 77/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683], [0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767], [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285], [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6171, 0.3931, 0.7081, 0.2966, 0.3679, 0.2957, 0.5455, 0.5772], [0.6125, 0.3821, 0.8967, 0.4685, 0.4851, 0.4934, 0.5463, 0.5613], [0.6223, 0.3868, 0.8138, 0.2293, 0.4097, 0.2930, 0.6357, 0.5563], [0.6632, 0.4119, 0.7575, 0.2619, 0.4331, 0.2198, 0.5652, 0.5772], [0.6501, 0.3907, 0.9321, 0.4798, 0.3699, 0.3962, 0.6038, 0.5200], [0.5769, 0.3612, 0.9468, 0.4648, 0.3979, 0.5583, 0.7056, 0.5172], [0.6279, 0.3926, 0.8856, 0.3573, 0.3740, 0.2902, 0.5336, 0.5701], [0.5986, 0.3918, 0.8499, 0.5497, 0.4113, 0.4810, 0.5777, 0.6208]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683], [0.6230, 0.4152, 0.7588, 0.2283, 0.4013, 0.2883, 0.6200, 0.5767], [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6236, 0.3965, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183], [0.6254, 0.3993, 0.8988, 0.4767, 0.3988, 0.5517, 0.6955, 0.5285], [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.051187454839237034 step: 78 running loss: 0.0006562494210158595 Train Steps: 78/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719], [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6750, 0.4192, 0.8474, 0.3967, 0.3528, 0.4086, 0.5324, 0.5612], [0.6266, 0.3706, 0.8872, 0.4848, 0.4445, 0.4700, 0.5373, 0.5792], [0.6138, 0.3691, 0.8315, 0.2530, 0.4299, 0.2373, 0.6471, 0.5371], [0.6447, 0.3941, 0.7982, 0.3344, 0.3555, 0.3328, 0.5823, 0.5228], [0.6479, 0.4008, 0.8742, 0.4860, 0.3814, 0.4307, 0.5192, 0.5254], [0.6559, 0.4105, 0.8597, 0.4973, 0.4129, 0.5586, 0.7078, 0.5791], [0.6423, 0.3960, 0.9124, 0.5206, 0.3707, 0.3737, 0.6229, 0.4838], [0.6428, 0.3967, 0.9083, 0.4797, 0.4606, 0.5630, 0.6257, 0.5711]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719], [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750], [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6143, 0.3982, 0.8650, 0.4883, 0.3913, 0.4317, 0.5315, 0.5350], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.051443345699226484 step: 79 running loss: 0.0006511815911294491 Train Steps: 79/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901], [0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533], [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550], [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459], [0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343], [0.6268, 0.4061, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6421, 0.4114, 0.9054, 0.4475, 0.4030, 0.4379, 0.4815, 0.5277], [0.6290, 0.3797, 0.8638, 0.5687, 0.4455, 0.4976, 0.5647, 0.5426], [0.6579, 0.4121, 0.7245, 0.2258, 0.3609, 0.3025, 0.5843, 0.5721], [0.6677, 0.4169, 0.8331, 0.2470, 0.4378, 0.1644, 0.6066, 0.5146], [0.6490, 0.4068, 0.9010, 0.4690, 0.3478, 0.4634, 0.5632, 0.5723], [0.6050, 0.3680, 0.8381, 0.5526, 0.3784, 0.4725, 0.6788, 0.5530], [0.6591, 0.4004, 0.8229, 0.5532, 0.3631, 0.4770, 0.6958, 0.5461], [0.5582, 0.3386, 0.8518, 0.2422, 0.4435, 0.2046, 0.6203, 0.5380]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901], [0.6229, 0.4066, 0.8512, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350], [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533], [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6197, 0.4091, 0.8800, 0.4783, 0.3537, 0.4767, 0.5950, 0.5550], [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459], [0.6311, 0.3998, 0.7975, 0.5767, 0.3837, 0.4850, 0.7327, 0.5343], [0.6268, 0.4060, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.052004264638526365 step: 80 running loss: 0.0006500533079815795 Train Steps: 80/90 Loss: 0.0007 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350], [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817], [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400], [0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5737, 0.5700], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183], [0.6199, 0.4093, 0.7913, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5703, 0.3669, 0.7581, 0.2540, 0.4432, 0.1998, 0.5945, 0.5471], [0.6378, 0.4003, 0.7889, 0.2524, 0.4271, 0.1770, 0.6109, 0.5023], [0.5915, 0.3625, 0.7947, 0.5689, 0.4100, 0.5077, 0.5169, 0.5048], [0.6458, 0.4033, 0.8496, 0.4939, 0.4427, 0.5757, 0.5995, 0.5437], [0.6419, 0.4080, 0.8402, 0.5368, 0.3544, 0.4809, 0.5785, 0.5598], [0.6207, 0.3765, 0.8509, 0.2847, 0.4755, 0.2091, 0.7005, 0.5387], [0.6051, 0.3781, 0.7718, 0.2943, 0.3607, 0.2736, 0.5701, 0.5217], [0.5617, 0.3639, 0.7565, 0.2403, 0.4141, 0.2534, 0.6069, 0.5797]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350], [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817], [0.6185, 0.4098, 0.8838, 0.4900, 0.4538, 0.5800, 0.6288, 0.5400], [0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5738, 0.5700], [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343], [0.6201, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183], [0.6198, 0.4093, 0.7912, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.052478661498753354 step: 81 running loss: 0.0006478847098611525 Train Steps: 81/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6164, 0.4119, 0.7913, 0.2650, 0.3538, 0.3500, 0.5614, 0.5038], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667], [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268], [0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433], [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5892, 0.3835, 0.8204, 0.6024, 0.3746, 0.4807, 0.6545, 0.5273], [0.6529, 0.4333, 0.7754, 0.2854, 0.3490, 0.3485, 0.5996, 0.5021], [0.5644, 0.3742, 0.7539, 0.2913, 0.3867, 0.2633, 0.5590, 0.5586], [0.6384, 0.4279, 0.8462, 0.5287, 0.4235, 0.5200, 0.5983, 0.5212], [0.6365, 0.4300, 0.8416, 0.3824, 0.3755, 0.3318, 0.6057, 0.5722], [0.6088, 0.3886, 0.8468, 0.4973, 0.4447, 0.4895, 0.6455, 0.5237], [0.6316, 0.4297, 0.7667, 0.2996, 0.3584, 0.3493, 0.6122, 0.5377], [0.6340, 0.4251, 0.8641, 0.4704, 0.4122, 0.4273, 0.5046, 0.5069]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297], [0.6164, 0.4119, 0.7912, 0.2650, 0.3537, 0.3500, 0.5614, 0.5038], [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650], [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202], [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5987, 0.5667], [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268], [0.6175, 0.4091, 0.7862, 0.2800, 0.3638, 0.3583, 0.6187, 0.5433], [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.05284168719663285 step: 82 running loss: 0.0006444108194711323 Train Steps: 82/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517], [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5988, 0.5283], [0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795], [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183], [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501], [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400], [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6312, 0.4297, 0.8587, 0.4907, 0.4612, 0.5042, 0.5684, 0.5371], [0.5988, 0.4089, 0.8584, 0.5445, 0.3812, 0.4816, 0.5975, 0.5158], [0.5220, 0.3592, 0.6690, 0.2524, 0.3919, 0.1902, 0.5083, 0.5494], [0.5786, 0.3829, 0.8285, 0.5459, 0.3944, 0.5311, 0.7094, 0.5628], [0.5903, 0.3994, 0.8665, 0.5041, 0.4634, 0.5658, 0.6238, 0.5060], [0.6346, 0.4473, 0.7965, 0.2709, 0.4465, 0.2053, 0.5946, 0.5359], [0.6350, 0.4314, 0.8643, 0.5173, 0.4638, 0.5775, 0.6126, 0.5268], [0.6126, 0.4190, 0.7661, 0.2302, 0.4490, 0.2217, 0.6696, 0.5299]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6175, 0.3957, 0.8700, 0.4817, 0.4663, 0.5133, 0.5800, 0.5517], [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5987, 0.5283], [0.6149, 0.4054, 0.6712, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795], [0.6251, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6087, 0.5183], [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501], [0.6185, 0.4098, 0.8838, 0.4900, 0.4538, 0.5800, 0.6288, 0.5400], [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.05330092628719285 step: 83 running loss: 0.000642179834785456 Train Steps: 83/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6218, 0.4137, 0.7263, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783], [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050], [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667], [0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585], [0.6199, 0.4093, 0.7913, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700], [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400], [0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5973, 0.3945, 0.7225, 0.2356, 0.4205, 0.2714, 0.6090, 0.5509], [0.6181, 0.4025, 0.7789, 0.3133, 0.3741, 0.3125, 0.5720, 0.4842], [0.6007, 0.4061, 0.8564, 0.5378, 0.4134, 0.5479, 0.7513, 0.5355], [0.6377, 0.4411, 0.8541, 0.3740, 0.3877, 0.3519, 0.6080, 0.5572], [0.6030, 0.4121, 0.8575, 0.4374, 0.4640, 0.5614, 0.5938, 0.5109], [0.5986, 0.4167, 0.7761, 0.2509, 0.4511, 0.2576, 0.6155, 0.5611], [0.6257, 0.4321, 0.8652, 0.4166, 0.3818, 0.4734, 0.6011, 0.5235], [0.6366, 0.4268, 0.8353, 0.5934, 0.3804, 0.4744, 0.6389, 0.4864]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6218, 0.4137, 0.7262, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783], [0.6135, 0.3994, 0.7912, 0.3050, 0.3625, 0.3050, 0.5838, 0.5050], [0.6339, 0.4123, 0.8637, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436], [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5987, 0.5667], [0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585], [0.6198, 0.4093, 0.7912, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700], [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6137, 0.5400], [0.6310, 0.4017, 0.8562, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.05356978881172836 step: 84 running loss: 0.0006377355810920042 Train Steps: 84/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272], [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240], [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879], [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804], [0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413], [0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5875, 0.4149, 0.8459, 0.3509, 0.3870, 0.2874, 0.5411, 0.5442], [0.5924, 0.4037, 0.7505, 0.2112, 0.4286, 0.1813, 0.5500, 0.5437], [0.5634, 0.4030, 0.8751, 0.4425, 0.4029, 0.5792, 0.6001, 0.5064], [0.5981, 0.4082, 0.8929, 0.4612, 0.3577, 0.4834, 0.6200, 0.4828], [0.6246, 0.4228, 0.7950, 0.5650, 0.3801, 0.5068, 0.7198, 0.5421], [0.5945, 0.4179, 0.8718, 0.4739, 0.4445, 0.5150, 0.5740, 0.5743], [0.6081, 0.4048, 0.9095, 0.3811, 0.4746, 0.2754, 0.7238, 0.5498], [0.5983, 0.4079, 0.8501, 0.3613, 0.3965, 0.2732, 0.6135, 0.4916]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6186, 0.4080, 0.8625, 0.3483, 0.3787, 0.2650, 0.5320, 0.5272], [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240], [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879], [0.6263, 0.4038, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804], [0.6311, 0.3998, 0.7975, 0.5767, 0.3837, 0.4850, 0.7327, 0.5343], [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617], [0.6271, 0.4040, 0.9137, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413], [0.6277, 0.4036, 0.8687, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742]]], device='cuda:0') loss_train_step before backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train: 0.05381043939269148 step: 85 running loss: 0.0006330639928551939 Train Steps: 85/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [ nan, nan, 0.7335, 0.2569, 0.3788, 0.2667, 0.5066, 0.5578], [0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667], [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097], [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [ nan, nan, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650], [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6852, 0.4618, 0.8708, 0.3944, 0.3783, 0.4155, 0.6161, 0.5445], [0.0702, 0.0619, 0.7505, 0.2546, 0.3969, 0.2535, 0.5469, 0.5487], [0.6904, 0.4733, 0.6993, 0.2870, 0.3562, 0.3299, 0.5718, 0.5590], [0.7018, 0.4831, 0.8874, 0.4605, 0.4533, 0.6048, 0.5969, 0.4998], [0.6675, 0.4485, 0.9028, 0.4753, 0.4376, 0.5313, 0.6724, 0.5408], [0.0645, 0.0471, 0.7440, 0.2061, 0.4404, 0.2277, 0.5841, 0.5455], [0.6645, 0.4337, 0.8776, 0.4873, 0.4567, 0.4745, 0.5866, 0.5565], [0.6834, 0.4593, 0.8890, 0.4265, 0.3717, 0.3716, 0.6853, 0.5162]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650], [0.0000, 0.0000, 0.7335, 0.2569, 0.3787, 0.2667, 0.5066, 0.5578], [0.6134, 0.4090, 0.6926, 0.2819, 0.3537, 0.3233, 0.5562, 0.5667], [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097], [0.6241, 0.4142, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.0000, 0.0000, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650], [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100]]], device='cuda:0') loss_train_step before backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0012, device='cuda:0', grad_fn=) loss_train: 0.055031666153809056 step: 86 running loss: 0.0006399030948117332 Train Steps: 86/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397], [0.6346, 0.4092, 0.7712, 0.5917, 0.4037, 0.4767, 0.7343, 0.5725], [0.6218, 0.4098, 0.7238, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350], [0.6157, 0.4102, 0.8513, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890], [0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167], [0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917], [0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6089, 0.4066, 0.8930, 0.4500, 0.3558, 0.3774, 0.5126, 0.5316], [0.6178, 0.4133, 0.8086, 0.5496, 0.3828, 0.4906, 0.7146, 0.5772], [0.5650, 0.3830, 0.7388, 0.1678, 0.4170, 0.2644, 0.6357, 0.5449], [0.5711, 0.3814, 0.8569, 0.3553, 0.3582, 0.3664, 0.5219, 0.5682], [0.6151, 0.3987, 0.7894, 0.2035, 0.4397, 0.1833, 0.6035, 0.5115], [0.6286, 0.4139, 0.8668, 0.5642, 0.4487, 0.4700, 0.5509, 0.5796], [0.5903, 0.3813, 0.9292, 0.4573, 0.3781, 0.5267, 0.6417, 0.4760], [0.6194, 0.4071, 0.9080, 0.4125, 0.3576, 0.3366, 0.5864, 0.5319]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6127, 0.4084, 0.8737, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397], [0.6346, 0.4092, 0.7713, 0.5917, 0.4038, 0.4767, 0.7343, 0.5725], [0.6218, 0.4098, 0.7237, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350], [0.6157, 0.4102, 0.8512, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890], [0.6204, 0.4007, 0.7837, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167], [0.6248, 0.4185, 0.8500, 0.5767, 0.4462, 0.4550, 0.5612, 0.5917], [0.6264, 0.4049, 0.8988, 0.4633, 0.3812, 0.4983, 0.6326, 0.4843], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.05537027501850389 step: 87 running loss: 0.000636439942741424 Train Steps: 87/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170], [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6289, 0.4024, 0.9088, 0.4567, 0.3937, 0.5633, 0.7058, 0.5609], [0.6261, 0.4066, 0.8325, 0.2150, 0.4763, 0.2667, 0.7002, 0.5633], [ nan, nan, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650], [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683], [0.6272, 0.4045, 0.8538, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6273, 0.3890, 0.8337, 0.2448, 0.4812, 0.1961, 0.6621, 0.5329], [ 0.5398, 0.3412, 0.7003, 0.2040, 0.4109, 0.1969, 0.5295, 0.5867], [ 0.6335, 0.3931, 0.9081, 0.4608, 0.3957, 0.5802, 0.7179, 0.5608], [ 0.6215, 0.3847, 0.8332, 0.2021, 0.4697, 0.2656, 0.6955, 0.5724], [ 0.0028, -0.0193, 0.7317, 0.1922, 0.4107, 0.2410, 0.5504, 0.5703], [ 0.6439, 0.4225, 0.8855, 0.4874, 0.3613, 0.4436, 0.5802, 0.5788], [ 0.6505, 0.4059, 0.8815, 0.5170, 0.3708, 0.3837, 0.5383, 0.5823], [ 0.6387, 0.4065, 0.8629, 0.5829, 0.3632, 0.4534, 0.5999, 0.4838]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170], [0.6154, 0.4112, 0.7038, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600], [0.6289, 0.4024, 0.9087, 0.4567, 0.3938, 0.5633, 0.7058, 0.5609], [0.6261, 0.4066, 0.8325, 0.2150, 0.4762, 0.2667, 0.7002, 0.5633], [0.0000, 0.0000, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650], [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5412, 0.5683], [0.6271, 0.4045, 0.8537, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.05571220855927095 step: 88 running loss: 0.0006330932790826245 Train Steps: 88/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285], [0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683], [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383], [0.6111, 0.4033, 0.8300, 0.3267, 0.3588, 0.3333, 0.5444, 0.5637], [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407], [0.6125, 0.4076, 0.8488, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6336, 0.3862, 0.7997, 0.4040, 0.3665, 0.4314, 0.5084, 0.5298], [0.6450, 0.4073, 0.8902, 0.3394, 0.3426, 0.4122, 0.5901, 0.5705], [0.6138, 0.3758, 0.9187, 0.4917, 0.3973, 0.5376, 0.7533, 0.5515], [0.6018, 0.3784, 0.7607, 0.2297, 0.4623, 0.1590, 0.5976, 0.5293], [0.6231, 0.3946, 0.8374, 0.3192, 0.3491, 0.3294, 0.5278, 0.5643], [0.5974, 0.3757, 0.8539, 0.3863, 0.3613, 0.3103, 0.4865, 0.5473], [0.5614, 0.3451, 0.8446, 0.3562, 0.3489, 0.3618, 0.4804, 0.5477], [0.6163, 0.3803, 0.9144, 0.4768, 0.4139, 0.5283, 0.6034, 0.5241]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285], [0.6219, 0.4097, 0.8737, 0.3400, 0.3562, 0.4117, 0.5975, 0.5683], [0.6339, 0.4102, 0.9087, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383], [0.6111, 0.4033, 0.8300, 0.3267, 0.3587, 0.3333, 0.5444, 0.5637], [0.6133, 0.4094, 0.8495, 0.4028, 0.3587, 0.3200, 0.5003, 0.5407], [0.6125, 0.4076, 0.8487, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.056046031415462494 step: 89 running loss: 0.0006297306900613763 Train Steps: 89/90 Loss: 0.0006 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8850, 0.3000, 0.5363, 0.2250, 0.7343, 0.5771], [ nan, nan, 0.7192, 0.2346, 0.4037, 0.2050, 0.5138, 0.5650], [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320], [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6138, 0.5450], [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833], [0.6263, 0.4030, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771], [0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.0543, 0.0123, 0.8911, 0.3049, 0.5193, 0.2527, 0.7156, 0.5693], [0.0740, 0.0303, 0.7232, 0.2378, 0.4109, 0.2229, 0.4969, 0.5576], [0.6706, 0.4200, 0.8034, 0.2314, 0.4377, 0.2683, 0.6421, 0.5472], [0.6664, 0.4229, 0.8999, 0.4158, 0.4024, 0.2327, 0.5504, 0.5444], [0.6982, 0.4263, 0.9126, 0.4696, 0.4135, 0.5241, 0.6159, 0.5539], [0.6612, 0.4072, 0.6893, 0.3068, 0.3739, 0.2782, 0.5454, 0.5953], [0.6631, 0.4017, 0.9254, 0.4943, 0.3901, 0.5292, 0.6213, 0.4722], [0.6907, 0.4428, 0.7078, 0.2828, 0.3454, 0.3258, 0.5317, 0.5711]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8850, 0.3000, 0.5362, 0.2250, 0.7343, 0.5771], [0.0000, 0.0000, 0.7192, 0.2346, 0.4038, 0.2050, 0.5138, 0.5650], [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320], [0.6226, 0.4098, 0.8913, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6137, 0.5450], [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833], [0.6263, 0.4029, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771], [0.6134, 0.4090, 0.6926, 0.2819, 0.3537, 0.3233, 0.5562, 0.5667]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.05668294767383486 step: 90 running loss: 0.0006298105297092762 Valid Steps: 10/10 Loss: nan 06 -------------------------------------------------- Epoch: 9 Train Loss: 0.0006 Valid Loss: nan -------------------------------------------------- size of train loader is: 90 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748], [0.6200, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367], [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267], [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058], [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6203, 0.4021, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6319, 0.4033, 0.8382, 0.5868, 0.4032, 0.4758, 0.5705, 0.6225], [0.6193, 0.4029, 0.8496, 0.5472, 0.3828, 0.3973, 0.6909, 0.5676], [0.6050, 0.3735, 0.8999, 0.4544, 0.3672, 0.5176, 0.5923, 0.5468], [0.5715, 0.3618, 0.8084, 0.2102, 0.4502, 0.1593, 0.6312, 0.5402], [0.6272, 0.3849, 0.8458, 0.4138, 0.3584, 0.4622, 0.5293, 0.5167], [0.6106, 0.3786, 0.9130, 0.4429, 0.3443, 0.4480, 0.6162, 0.4848], [0.6164, 0.3759, 0.8053, 0.2452, 0.4003, 0.2702, 0.5980, 0.5262], [0.6215, 0.3952, 0.8878, 0.5006, 0.3623, 0.3742, 0.5542, 0.5449]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117], [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748], [0.6199, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367], [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267], [0.6084, 0.4005, 0.8400, 0.4317, 0.3762, 0.4750, 0.5476, 0.5058], [0.6263, 0.4038, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6203, 0.4020, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.00033010990591719747 step: 1 running loss: 0.00033010990591719747 Train Steps: 1/90 Loss: 0.0003 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290], [0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843], [0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567], [0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[-0.0342, -0.0272, 0.8486, 0.2634, 0.4950, 0.2271, 0.7346, 0.5626], [ 0.6159, 0.4097, 0.8308, 0.5721, 0.3762, 0.4762, 0.6822, 0.5432], [ 0.5889, 0.3787, 0.8285, 0.4872, 0.3594, 0.4427, 0.5375, 0.4913], [ 0.6269, 0.4067, 0.8038, 0.3009, 0.3414, 0.3860, 0.6229, 0.5287], [ 0.6134, 0.3922, 0.8882, 0.4636, 0.3768, 0.4819, 0.6375, 0.4844], [ 0.6016, 0.3828, 0.8553, 0.4822, 0.4793, 0.4714, 0.5879, 0.5612], [ 0.6456, 0.4502, 0.8314, 0.4961, 0.4227, 0.2749, 0.5583, 0.5983], [ 0.5594, 0.3836, 0.7677, 0.3099, 0.3410, 0.2879, 0.4798, 0.5510]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6038, 0.3946, 0.8413, 0.4883, 0.3562, 0.4550, 0.5266, 0.4693], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290], [0.6264, 0.4049, 0.8988, 0.4633, 0.3812, 0.4983, 0.6326, 0.4843], [0.6112, 0.4029, 0.8637, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567], [0.6278, 0.4253, 0.8875, 0.5017, 0.4112, 0.2750, 0.5413, 0.6196], [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.0007069317507557571 step: 2 running loss: 0.00035346587537787855 Train Steps: 2/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064], [0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6238, 0.5350], [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6060, 0.3924, 0.8450, 0.5717, 0.4200, 0.5217, 0.5253, 0.4752], [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550], [0.6104, 0.4029, 0.8738, 0.4900, 0.4088, 0.4533, 0.5070, 0.5510]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5574, 0.3672, 0.7911, 0.3417, 0.3739, 0.2880, 0.5773, 0.5484], [0.5857, 0.4212, 0.7881, 0.5672, 0.3776, 0.4325, 0.5911, 0.6166], [0.5877, 0.4114, 0.8576, 0.5004, 0.4741, 0.4730, 0.5572, 0.5137], [0.5858, 0.3907, 0.8654, 0.4103, 0.3588, 0.5215, 0.6651, 0.5400], [0.6192, 0.4234, 0.7952, 0.2539, 0.4438, 0.1704, 0.6434, 0.4868], [0.5735, 0.3856, 0.8259, 0.5744, 0.4119, 0.4878, 0.5524, 0.4930], [0.6191, 0.4261, 0.8729, 0.4594, 0.4652, 0.4792, 0.6136, 0.5484], [0.6088, 0.4250, 0.8484, 0.4860, 0.4031, 0.4150, 0.5401, 0.5447]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6128, 0.4022, 0.8737, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064], [0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6237, 0.5350], [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869], [0.6060, 0.3924, 0.8450, 0.5717, 0.4200, 0.5217, 0.5253, 0.4752], [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550], [0.6104, 0.4029, 0.8737, 0.4900, 0.4087, 0.4533, 0.5070, 0.5510]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.0011995541281066835 step: 3 running loss: 0.00039985137603556115 Train Steps: 3/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116], [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334], [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395], [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578], [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864], [0.6083, 0.3957, 0.8638, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5633, 0.3993, 0.8716, 0.4412, 0.3910, 0.5889, 0.5730, 0.5090], [0.6544, 0.4508, 0.7246, 0.2789, 0.4351, 0.2671, 0.5843, 0.6092], [0.5153, 0.3684, 0.8457, 0.2617, 0.5448, 0.2269, 0.7136, 0.5446], [0.5525, 0.3710, 0.9015, 0.4977, 0.3737, 0.4679, 0.6343, 0.5163], [0.6168, 0.4171, 0.7332, 0.2345, 0.4105, 0.2477, 0.5846, 0.5238], [0.6235, 0.4353, 0.6781, 0.2241, 0.4157, 0.1958, 0.5455, 0.5421], [0.5919, 0.4037, 0.8823, 0.5418, 0.3836, 0.4859, 0.7154, 0.5731], [0.5329, 0.3702, 0.8681, 0.5204, 0.4421, 0.5144, 0.5354, 0.4807]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116], [0.6193, 0.4078, 0.7287, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334], [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395], [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578], [0.6314, 0.4107, 0.8750, 0.5100, 0.3787, 0.4900, 0.7121, 0.5864], [0.6083, 0.3957, 0.8637, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.001964781025890261 step: 4 running loss: 0.0004911952564725652 Train Steps: 4/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650], [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082], [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167], [0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183], [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5787, 0.5600], [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500], [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6021, 0.4175, 0.8881, 0.4239, 0.3671, 0.4225, 0.6052, 0.5651], [0.6143, 0.4123, 0.8668, 0.5317, 0.4316, 0.5661, 0.6044, 0.5064], [0.5993, 0.4155, 0.8203, 0.3317, 0.3824, 0.3246, 0.5612, 0.5325], [0.6159, 0.4190, 0.7533, 0.2214, 0.4518, 0.1878, 0.5800, 0.5094], [0.6217, 0.4305, 0.8700, 0.4981, 0.4964, 0.5245, 0.5738, 0.5462], [0.6181, 0.4288, 0.8747, 0.5019, 0.3862, 0.4134, 0.5927, 0.5407], [0.5921, 0.3948, 0.7740, 0.3583, 0.3586, 0.3819, 0.5415, 0.5482], [0.6055, 0.4110, 0.7385, 0.2399, 0.3841, 0.3300, 0.6114, 0.5342]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6203, 0.4096, 0.8863, 0.4267, 0.3537, 0.4117, 0.6025, 0.5650], [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082], [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167], [0.6213, 0.4001, 0.7713, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183], [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5788, 0.5600], [0.6262, 0.4163, 0.8850, 0.5183, 0.3762, 0.4150, 0.6025, 0.5500], [0.6102, 0.4001, 0.7738, 0.3583, 0.3462, 0.3800, 0.5524, 0.5689], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334]]], device='cuda:0') loss_train_step before backward: tensor(0.0001, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0001, device='cuda:0', grad_fn=) loss_train: 0.0021081157465232536 step: 5 running loss: 0.0004216231493046507 Train Steps: 5/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6206, 0.4001, 0.8900, 0.3933, 0.3588, 0.3567, 0.5837, 0.5083], [ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220], [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166], [0.6152, 0.4131, 0.6863, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427], [0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019], [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6132, 0.4043, 0.9107, 0.4013, 0.3892, 0.3707, 0.5962, 0.5043], [0.0139, 0.0169, 0.7506, 0.2564, 0.4062, 0.2713, 0.5084, 0.5668], [0.6754, 0.4624, 0.7043, 0.2228, 0.4741, 0.1872, 0.5517, 0.5211], [0.6439, 0.4191, 0.8781, 0.5365, 0.4522, 0.5445, 0.6816, 0.5261], [0.6293, 0.4285, 0.6871, 0.2469, 0.3909, 0.3598, 0.5914, 0.5490], [0.6005, 0.4066, 0.7894, 0.4102, 0.3767, 0.3983, 0.5329, 0.5549], [0.6179, 0.4132, 0.9037, 0.4317, 0.3969, 0.5640, 0.5869, 0.5120], [0.6048, 0.4038, 0.9222, 0.3448, 0.4445, 0.3862, 0.7535, 0.5470]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6206, 0.4001, 0.8900, 0.3933, 0.3587, 0.3567, 0.5838, 0.5083], [0.0000, 0.0000, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220], [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166], [0.6152, 0.4131, 0.6862, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305], [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427], [0.6161, 0.4099, 0.8737, 0.4383, 0.3787, 0.5483, 0.5605, 0.5019], [0.6332, 0.4165, 0.9100, 0.3350, 0.4187, 0.3683, 0.7438, 0.5528]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.002486242665327154 step: 6 running loss: 0.0004143737775545257 Train Steps: 6/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6095, 0.3970, 0.8688, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940], [0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550], [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6171, 0.4000, 0.8989, 0.4624, 0.4872, 0.5078, 0.5293, 0.5201], [0.6218, 0.4176, 0.8943, 0.4584, 0.3685, 0.3988, 0.5197, 0.5399], [0.6205, 0.4179, 0.7013, 0.2469, 0.4153, 0.2526, 0.5713, 0.5631], [0.6313, 0.3969, 0.8684, 0.5323, 0.4041, 0.4980, 0.6304, 0.5513], [0.6323, 0.4256, 0.6733, 0.2832, 0.3674, 0.3227, 0.5471, 0.5779], [0.6289, 0.3980, 0.8613, 0.5977, 0.3881, 0.5231, 0.6237, 0.4799], [0.6523, 0.4320, 0.8302, 0.2181, 0.4760, 0.2077, 0.6532, 0.5335], [0.6019, 0.3899, 0.8802, 0.3725, 0.3655, 0.4080, 0.6073, 0.5136]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6095, 0.3970, 0.8687, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940], [0.6127, 0.4084, 0.8737, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5362, 0.5550], [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377], [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267], [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.0027903580776182935 step: 7 running loss: 0.0003986225825168991 Train Steps: 7/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633], [0.6110, 0.4047, 0.8700, 0.4483, 0.3713, 0.3967, 0.5088, 0.5517], [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650], [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297], [0.6226, 0.4103, 0.8575, 0.3450, 0.4388, 0.2067, 0.5787, 0.5383], [0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983], [0.6259, 0.4133, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6655, 0.4275, 0.8717, 0.5391, 0.4057, 0.5384, 0.5896, 0.5467], [0.5990, 0.3900, 0.8523, 0.4329, 0.3639, 0.4221, 0.4961, 0.5334], [0.6059, 0.3806, 0.8612, 0.4520, 0.4181, 0.5171, 0.5301, 0.5463], [0.6200, 0.3915, 0.8680, 0.5255, 0.3656, 0.4592, 0.6387, 0.5337], [0.6142, 0.4028, 0.8463, 0.3289, 0.4391, 0.2495, 0.5600, 0.5419], [0.6230, 0.4021, 0.8543, 0.5149, 0.3631, 0.4069, 0.5589, 0.5789], [0.6494, 0.4203, 0.8158, 0.2036, 0.4982, 0.1942, 0.6319, 0.4887], [0.6183, 0.3849, 0.8135, 0.5679, 0.3800, 0.4807, 0.5747, 0.5301]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6192, 0.4128, 0.8512, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633], [0.6110, 0.4047, 0.8700, 0.4483, 0.3713, 0.3967, 0.5088, 0.5517], [0.6098, 0.3991, 0.8637, 0.4717, 0.4263, 0.4967, 0.5213, 0.5650], [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297], [0.6226, 0.4103, 0.8575, 0.3450, 0.4387, 0.2067, 0.5788, 0.5383], [0.6260, 0.4214, 0.8537, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983], [0.6259, 0.4132, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.003120287772617303 step: 8 running loss: 0.00039003597157716285 Train Steps: 8/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6250, 0.4103, 0.8950, 0.4400, 0.3912, 0.5650, 0.6050, 0.5133], [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6083, 0.3957, 0.8638, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6907, 0.4464, 0.8934, 0.4423, 0.3799, 0.5879, 0.5853, 0.5208], [0.6341, 0.4092, 0.8970, 0.4907, 0.3774, 0.4600, 0.5645, 0.5343], [0.6480, 0.3964, 0.8914, 0.4762, 0.3615, 0.3755, 0.6009, 0.5030], [0.6196, 0.3916, 0.8647, 0.4987, 0.4383, 0.5247, 0.5082, 0.4973], [0.6607, 0.4309, 0.7277, 0.1843, 0.4020, 0.2626, 0.5911, 0.5482], [0.6550, 0.4048, 0.8476, 0.5480, 0.3676, 0.4864, 0.7078, 0.5719], [0.6345, 0.4082, 0.8107, 0.2881, 0.4759, 0.1987, 0.5459, 0.5407], [0.6141, 0.3883, 0.8778, 0.4642, 0.3770, 0.4266, 0.5691, 0.5419]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6250, 0.4103, 0.8950, 0.4400, 0.3913, 0.5650, 0.6050, 0.5133], [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6083, 0.3957, 0.8637, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760], [0.6229, 0.4107, 0.8138, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400], [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.0034082017227774486 step: 9 running loss: 0.0003786890803086054 Train Steps: 9/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8300, 0.3150, 0.3588, 0.3383, 0.5208, 0.5194], [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719], [0.6125, 0.4035, 0.7825, 0.3100, 0.3463, 0.4900, 0.5832, 0.5637], [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6173, 0.4114, 0.7325, 0.2500, 0.4213, 0.1917, 0.5338, 0.5700], [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019], [0.6081, 0.3950, 0.8538, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.0015, -0.0031, 0.7993, 0.3112, 0.3631, 0.3143, 0.5292, 0.5382], [ 0.6202, 0.3914, 0.8323, 0.4142, 0.3660, 0.4016, 0.5376, 0.5640], [ 0.6413, 0.4078, 0.7725, 0.3255, 0.3465, 0.4617, 0.6007, 0.5419], [ 0.6826, 0.4191, 0.8940, 0.4635, 0.3621, 0.4411, 0.6330, 0.4861], [ 0.6830, 0.4369, 0.7711, 0.2781, 0.4563, 0.1491, 0.5781, 0.5318], [ 0.6227, 0.3945, 0.7264, 0.2345, 0.4269, 0.1709, 0.5418, 0.5762], [ 0.6921, 0.4178, 0.8874, 0.4984, 0.3737, 0.3784, 0.6353, 0.5010], [ 0.6811, 0.4307, 0.8601, 0.4722, 0.3757, 0.4662, 0.5190, 0.4957]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8300, 0.3150, 0.3587, 0.3383, 0.5208, 0.5194], [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719], [0.6125, 0.4035, 0.7825, 0.3100, 0.3462, 0.4900, 0.5832, 0.5637], [0.6263, 0.4038, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6173, 0.4114, 0.7325, 0.2500, 0.4212, 0.1917, 0.5337, 0.5700], [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019], [0.6081, 0.3950, 0.8537, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.0039045417652232572 step: 10 running loss: 0.0003904541765223257 Train Steps: 10/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6115, 0.4005, 0.8838, 0.3867, 0.3763, 0.4700, 0.5800, 0.5550], [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494], [0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921], [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312], [0.6261, 0.4066, 0.8325, 0.2150, 0.4763, 0.2667, 0.7002, 0.5633], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6111, 0.3954, 0.8494, 0.5584, 0.3472, 0.3372, 0.5197, 0.5469], [0.5739, 0.3684, 0.8690, 0.3896, 0.3697, 0.4466, 0.5364, 0.5605], [0.6279, 0.3980, 0.8315, 0.5804, 0.3685, 0.3994, 0.6520, 0.5557], [0.6394, 0.4043, 0.8134, 0.2342, 0.3990, 0.2662, 0.6132, 0.5075], [0.5666, 0.3551, 0.8625, 0.4179, 0.3445, 0.4102, 0.5713, 0.5328], [0.6321, 0.3944, 0.8275, 0.2332, 0.4824, 0.2432, 0.6758, 0.5559], [0.6157, 0.3995, 0.7224, 0.1973, 0.4033, 0.2224, 0.5934, 0.5523], [0.6081, 0.3950, 0.8581, 0.3306, 0.4382, 0.1855, 0.5523, 0.5056]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6115, 0.4005, 0.8838, 0.3867, 0.3762, 0.4700, 0.5800, 0.5550], [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494], [0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921], [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312], [0.6261, 0.4066, 0.8325, 0.2150, 0.4762, 0.2667, 0.7002, 0.5633], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.0043209217401454225 step: 11 running loss: 0.0003928110672859475 Train Steps: 11/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6250, 0.4013, 0.8525, 0.5417, 0.4037, 0.5117, 0.6325, 0.5017], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550], [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609], [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533], [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862], [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519], [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6466, 0.4067, 0.8561, 0.5692, 0.3907, 0.4857, 0.6159, 0.4922], [0.5812, 0.3606, 0.8624, 0.5676, 0.3698, 0.4190, 0.5720, 0.5218], [0.6026, 0.3910, 0.8887, 0.4900, 0.3473, 0.4561, 0.5905, 0.5322], [0.6874, 0.4356, 0.8931, 0.3268, 0.3932, 0.2698, 0.6987, 0.5500], [0.6209, 0.3994, 0.7293, 0.2280, 0.3702, 0.2973, 0.6075, 0.5366], [0.6020, 0.3824, 0.9123, 0.4212, 0.3538, 0.4350, 0.6238, 0.4767], [0.6339, 0.4077, 0.8634, 0.5291, 0.4464, 0.4912, 0.4958, 0.5456], [0.6225, 0.3932, 0.8367, 0.2665, 0.4396, 0.2139, 0.6829, 0.5355]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6250, 0.4013, 0.8525, 0.5417, 0.4038, 0.5117, 0.6325, 0.5017], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6197, 0.4091, 0.8800, 0.4783, 0.3537, 0.4767, 0.5950, 0.5550], [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609], [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533], [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862], [0.6100, 0.4016, 0.8600, 0.5067, 0.4613, 0.5233, 0.5086, 0.5519], [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.004716772076790221 step: 12 running loss: 0.0003930643397325184 Train Steps: 12/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517], [0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272], [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835], [0.6048, 0.3928, 0.8538, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719], [0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148], [0.6205, 0.4081, 0.8950, 0.4017, 0.3788, 0.4700, 0.5963, 0.5667], [ nan, nan, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819], [0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6150, 0.3924, 0.8681, 0.4778, 0.4595, 0.5068, 0.6044, 0.5595], [0.6532, 0.4270, 0.8590, 0.3559, 0.3754, 0.2603, 0.5552, 0.5468], [0.6754, 0.4331, 0.9025, 0.3900, 0.4097, 0.2071, 0.6262, 0.5093], [0.6148, 0.4029, 0.8485, 0.5439, 0.3948, 0.4812, 0.5710, 0.5134], [0.6420, 0.4042, 0.8782, 0.5306, 0.3821, 0.4582, 0.6024, 0.5224], [0.6530, 0.4177, 0.8900, 0.3976, 0.3744, 0.4545, 0.6184, 0.5700], [0.0528, 0.0365, 0.6978, 0.2690, 0.3843, 0.2045, 0.5725, 0.5857], [0.6206, 0.4146, 0.7602, 0.2088, 0.3685, 0.2788, 0.6094, 0.5118]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6175, 0.3957, 0.8700, 0.4817, 0.4663, 0.5133, 0.5800, 0.5517], [0.6186, 0.4080, 0.8625, 0.3483, 0.3787, 0.2650, 0.5320, 0.5272], [0.6273, 0.4110, 0.8900, 0.3817, 0.4187, 0.2167, 0.5858, 0.4835], [0.6048, 0.3928, 0.8537, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719], [0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148], [0.6205, 0.4081, 0.8950, 0.4017, 0.3787, 0.4700, 0.5962, 0.5667], [0.0000, 0.0000, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819], [0.6163, 0.4114, 0.7650, 0.2017, 0.3762, 0.2867, 0.5631, 0.5071]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.00511384145647753 step: 13 running loss: 0.0003933724197290408 Train Steps: 13/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913], [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136], [0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967], [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200], [ nan, nan, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650], [0.6193, 0.3930, 0.8949, 0.4437, 0.3852, 0.5435, 0.6263, 0.5263], [0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6139, 0.3975, 0.7991, 0.2557, 0.4762, 0.1688, 0.6188, 0.5154], [ 0.6325, 0.4076, 0.8653, 0.3249, 0.3729, 0.3577, 0.6646, 0.5236], [ 0.5999, 0.3938, 0.6965, 0.2217, 0.3897, 0.2614, 0.5847, 0.5240], [ 0.6546, 0.4266, 0.9047, 0.4479, 0.4026, 0.5428, 0.6200, 0.5899], [ 0.6073, 0.4046, 0.8692, 0.3171, 0.4470, 0.2431, 0.6499, 0.5239], [-0.0289, -0.0041, 0.7068, 0.2008, 0.4259, 0.2260, 0.5732, 0.5807], [ 0.5991, 0.3933, 0.8810, 0.4519, 0.3926, 0.5543, 0.6291, 0.5463], [ 0.6007, 0.4034, 0.8795, 0.4790, 0.3675, 0.4355, 0.5861, 0.5176]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6286, 0.4078, 0.8062, 0.2267, 0.4787, 0.1533, 0.5953, 0.4913], [0.6255, 0.4017, 0.8687, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901], [0.6043, 0.4022, 0.6888, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136], [0.6199, 0.4102, 0.8950, 0.4417, 0.4013, 0.5367, 0.6112, 0.5967], [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200], [0.0000, 0.0000, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650], [0.6193, 0.3930, 0.8949, 0.4437, 0.3852, 0.5435, 0.6263, 0.5263], [0.6108, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117]]], device='cuda:0') loss_train_step before backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train: 0.005352006017346866 step: 14 running loss: 0.00038228614409620477 Train Steps: 14/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5787, 0.5550], [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967], [0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748], [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285], [0.6246, 0.4008, 0.8757, 0.5088, 0.4101, 0.5392, 0.6644, 0.5133], [ nan, nan, 0.8938, 0.2850, 0.4662, 0.3117, 0.7406, 0.5528]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6304, 0.4434, 0.8698, 0.3517, 0.3991, 0.2967, 0.6378, 0.5092], [0.6057, 0.4039, 0.8408, 0.5369, 0.3770, 0.4544, 0.5819, 0.5145], [0.6102, 0.4243, 0.8122, 0.2548, 0.3479, 0.3743, 0.5690, 0.5382], [0.6166, 0.4169, 0.8540, 0.4114, 0.3589, 0.5225, 0.5643, 0.5030], [0.5879, 0.4055, 0.9096, 0.4521, 0.4069, 0.2987, 0.7183, 0.5450], [0.5678, 0.3852, 0.8754, 0.3586, 0.4418, 0.2730, 0.6236, 0.5102], [0.6362, 0.4257, 0.8712, 0.5049, 0.3944, 0.5445, 0.6412, 0.5012], [0.0803, 0.0648, 0.8858, 0.2733, 0.4625, 0.3084, 0.7475, 0.5582]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183], [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398], [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5788, 0.5550], [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967], [0.6329, 0.4196, 0.9237, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748], [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285], [0.6246, 0.4008, 0.8757, 0.5088, 0.4101, 0.5392, 0.6644, 0.5133], [0.0000, 0.0000, 0.8938, 0.2850, 0.4663, 0.3117, 0.7406, 0.5528]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.005812907431391068 step: 15 running loss: 0.00038752716209273786 Train Steps: 15/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6233, 0.4091, 0.8100, 0.2950, 0.3563, 0.3883, 0.6013, 0.5200], [0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320], [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892], [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533], [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.0263, 0.0256, 0.6932, 0.1956, 0.4421, 0.2121, 0.5592, 0.5668], [0.5857, 0.3967, 0.7539, 0.1916, 0.4034, 0.2748, 0.6176, 0.5288], [0.6044, 0.4116, 0.8246, 0.3053, 0.3672, 0.4163, 0.6139, 0.5282], [0.6559, 0.4351, 0.9174, 0.4294, 0.3850, 0.5451, 0.6995, 0.5243], [0.6037, 0.4098, 0.8734, 0.4678, 0.3612, 0.3758, 0.5414, 0.5704], [0.5970, 0.4016, 0.8845, 0.4692, 0.4716, 0.5759, 0.6031, 0.5291], [0.6008, 0.4041, 0.8873, 0.4476, 0.3542, 0.3747, 0.5981, 0.5399], [0.6103, 0.4158, 0.8827, 0.3146, 0.4869, 0.2184, 0.6485, 0.5120]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6233, 0.4091, 0.8100, 0.2950, 0.3562, 0.3883, 0.6012, 0.5200], [0.6250, 0.3993, 0.9137, 0.4333, 0.3762, 0.5217, 0.6995, 0.5320], [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892], [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6252, 0.4162, 0.8662, 0.4467, 0.3625, 0.3567, 0.6037, 0.5533], [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.006098556055803783 step: 16 running loss: 0.00038115975348773645 Train Steps: 16/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783], [0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869], [0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5787, 0.5550], [0.6115, 0.3998, 0.7063, 0.2383, 0.4037, 0.1950, 0.5320, 0.4993], [0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843], [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5833, 0.3842, 0.8821, 0.4733, 0.3704, 0.4449, 0.5607, 0.5791], [0.5857, 0.3952, 0.8461, 0.5062, 0.4464, 0.5468, 0.5452, 0.5381], [0.5606, 0.3757, 0.8263, 0.1889, 0.4795, 0.2077, 0.6426, 0.4731], [0.5283, 0.3615, 0.8104, 0.3245, 0.4075, 0.2385, 0.5421, 0.5131], [0.5866, 0.3970, 0.8141, 0.2517, 0.3591, 0.3805, 0.5958, 0.5653], [0.5578, 0.3751, 0.7063, 0.2147, 0.4096, 0.2207, 0.5562, 0.5014], [0.6209, 0.3995, 0.8981, 0.4447, 0.3939, 0.5215, 0.6618, 0.4925], [0.5742, 0.4028, 0.8075, 0.3001, 0.4348, 0.2643, 0.5919, 0.5839]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783], [0.6075, 0.4000, 0.8512, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6276, 0.4095, 0.8238, 0.2250, 0.4663, 0.1783, 0.6171, 0.4869], [0.6143, 0.4040, 0.8238, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973], [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5788, 0.5550], [0.6115, 0.3998, 0.7063, 0.2383, 0.4038, 0.1950, 0.5320, 0.4993], [0.6264, 0.4049, 0.8988, 0.4633, 0.3812, 0.4983, 0.6326, 0.4843], [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.00670615334820468 step: 17 running loss: 0.00039447960871792236 Train Steps: 17/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6075, 0.4007, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280], [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575], [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700], [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343], [0.6289, 0.4024, 0.9088, 0.4567, 0.3937, 0.5633, 0.7058, 0.5609], [0.6250, 0.4146, 0.8838, 0.3933, 0.3588, 0.4283, 0.6162, 0.5367]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6088, 0.4146, 0.8243, 0.4837, 0.4125, 0.5085, 0.4962, 0.5246], [0.5454, 0.3526, 0.7500, 0.1850, 0.4105, 0.2336, 0.5899, 0.5240], [0.6472, 0.4219, 0.8957, 0.2913, 0.4292, 0.4209, 0.6909, 0.5476], [0.6005, 0.4060, 0.8789, 0.4457, 0.3934, 0.4989, 0.5180, 0.5573], [0.6391, 0.4281, 0.9070, 0.4751, 0.3749, 0.4296, 0.6496, 0.5202], [0.6069, 0.3982, 0.7994, 0.5490, 0.3786, 0.4836, 0.7198, 0.5246], [0.5879, 0.3833, 0.8961, 0.4360, 0.4143, 0.5762, 0.6696, 0.5471], [0.6208, 0.4172, 0.8873, 0.3753, 0.3730, 0.4114, 0.5894, 0.5254]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6075, 0.4006, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280], [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575], [0.6111, 0.3995, 0.8788, 0.4567, 0.3812, 0.4833, 0.5450, 0.5700], [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274], [0.6311, 0.3998, 0.7975, 0.5767, 0.3837, 0.4850, 0.7327, 0.5343], [0.6289, 0.4024, 0.9087, 0.4567, 0.3938, 0.5633, 0.7058, 0.5609], [0.6250, 0.4146, 0.8838, 0.3933, 0.3587, 0.4283, 0.6162, 0.5367]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.007039182659354992 step: 18 running loss: 0.00039106570329749957 Train Steps: 18/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6228, 0.4119, 0.7938, 0.2233, 0.4674, 0.1773, 0.6188, 0.5433], [0.6198, 0.4101, 0.8838, 0.5283, 0.3763, 0.5267, 0.5913, 0.5567], [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6357, 0.4118, 0.8400, 0.2500, 0.5413, 0.1633, 0.6725, 0.5586], [ nan, nan, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.5953, 0.3860, 0.8574, 0.4379, 0.4426, 0.2795, 0.5448, 0.6258], [ 0.6346, 0.3960, 0.8043, 0.1938, 0.4684, 0.1988, 0.6204, 0.5387], [ 0.6455, 0.4134, 0.8872, 0.5100, 0.3674, 0.5438, 0.5928, 0.5578], [ 0.6155, 0.3911, 0.8685, 0.5148, 0.4183, 0.5489, 0.5997, 0.5189], [ 0.5396, 0.3379, 0.7125, 0.1986, 0.4258, 0.1941, 0.5334, 0.5251], [ 0.6262, 0.3989, 0.8644, 0.2575, 0.4861, 0.2410, 0.6567, 0.5455], [ 0.6301, 0.3905, 0.8583, 0.2206, 0.5306, 0.2024, 0.6591, 0.5488], [ 0.0209, -0.0096, 0.7140, 0.2022, 0.4335, 0.1930, 0.5282, 0.5127]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6228, 0.4119, 0.7937, 0.2233, 0.4674, 0.1773, 0.6187, 0.5433], [0.6198, 0.4101, 0.8838, 0.5283, 0.3762, 0.5267, 0.5913, 0.5567], [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134], [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6357, 0.4118, 0.8400, 0.2500, 0.5412, 0.1633, 0.6725, 0.5586], [0.0000, 0.0000, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.00753981365414802 step: 19 running loss: 0.00039683229758673787 Train Steps: 19/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617], [0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467], [0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050], [0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567], [0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6439, 0.4007, 0.8235, 0.3386, 0.3773, 0.2824, 0.5642, 0.5577], [0.6291, 0.3961, 0.8844, 0.4833, 0.4212, 0.4785, 0.6076, 0.5477], [0.6298, 0.3989, 0.8760, 0.5110, 0.4461, 0.4928, 0.5428, 0.5114], [0.6189, 0.3918, 0.7877, 0.2282, 0.3809, 0.2835, 0.5803, 0.5645], [0.6044, 0.4022, 0.7587, 0.2884, 0.3577, 0.4086, 0.5835, 0.5339], [0.6266, 0.3994, 0.8935, 0.4803, 0.4208, 0.5055, 0.5916, 0.5261], [0.6269, 0.3987, 0.7843, 0.2327, 0.4596, 0.1697, 0.6159, 0.5375], [0.6304, 0.4008, 0.8788, 0.4662, 0.3646, 0.4020, 0.5589, 0.5101]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617], [0.6199, 0.4059, 0.8700, 0.4900, 0.4162, 0.5000, 0.6162, 0.5467], [0.6097, 0.3988, 0.8650, 0.5250, 0.4212, 0.5200, 0.5675, 0.5050], [0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567], [0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250], [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400], [0.6108, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117]]], device='cuda:0') loss_train_step before backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train: 0.007740936329355463 step: 20 running loss: 0.00038704681646777316 Train Steps: 20/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823], [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100], [0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090], [ nan, nan, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220], [0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196], [0.6198, 0.4130, 0.8762, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290], [ nan, nan, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6662, 0.4223, 0.7406, 0.3587, 0.3473, 0.3636, 0.5362, 0.5764], [ 0.6988, 0.4409, 0.8367, 0.5339, 0.4200, 0.4791, 0.6176, 0.5213], [ 0.6454, 0.4078, 0.8821, 0.3713, 0.4065, 0.2698, 0.6423, 0.5060], [ 0.0193, -0.0011, 0.7563, 0.2482, 0.3874, 0.2615, 0.5023, 0.5327], [ 0.6599, 0.4391, 0.8735, 0.4986, 0.4344, 0.2519, 0.5618, 0.6114], [ 0.6724, 0.4471, 0.8753, 0.4173, 0.3714, 0.4785, 0.5753, 0.5135], [ 0.6454, 0.4219, 0.8223, 0.3016, 0.3528, 0.3894, 0.6167, 0.5270], [ 0.1428, 0.0750, 0.8914, 0.3256, 0.5149, 0.1905, 0.6792, 0.5704]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6170, 0.4102, 0.7468, 0.3695, 0.3462, 0.3767, 0.5238, 0.5823], [0.6299, 0.4008, 0.8450, 0.5350, 0.4212, 0.5000, 0.6350, 0.5100], [0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090], [0.0000, 0.0000, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220], [0.6278, 0.4253, 0.8875, 0.5017, 0.4112, 0.2750, 0.5413, 0.6196], [0.6198, 0.4130, 0.8763, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290], [0.0000, 0.0000, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.008517837704857811 step: 21 running loss: 0.0004056113192789434 Train Steps: 21/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693], [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967], [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500], [0.6272, 0.4045, 0.8538, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5720, 0.3776, 0.8345, 0.4755, 0.3489, 0.4654, 0.5330, 0.5079], [0.5836, 0.3819, 0.8806, 0.5104, 0.3676, 0.3222, 0.6156, 0.5192], [0.6031, 0.3888, 0.7692, 0.2578, 0.3459, 0.3895, 0.6086, 0.5580], [0.6142, 0.4031, 0.8418, 0.5832, 0.3648, 0.4312, 0.6062, 0.4846], [0.6351, 0.4040, 0.8955, 0.3888, 0.4027, 0.2019, 0.5870, 0.4993], [0.5938, 0.3881, 0.8417, 0.3796, 0.3461, 0.3908, 0.5894, 0.5815], [0.5953, 0.3816, 0.7752, 0.2554, 0.4761, 0.1240, 0.5810, 0.5446], [0.5647, 0.3757, 0.8465, 0.4529, 0.4370, 0.2333, 0.5333, 0.6345]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6038, 0.3946, 0.8413, 0.4883, 0.3562, 0.4550, 0.5266, 0.4693], [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967], [0.6197, 0.4051, 0.7812, 0.2650, 0.3512, 0.4050, 0.6112, 0.5500], [0.6271, 0.4045, 0.8537, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6273, 0.4110, 0.8900, 0.3817, 0.4187, 0.2167, 0.5858, 0.4835], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6245, 0.4100, 0.7763, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.008867989556165412 step: 22 running loss: 0.0004030904343711551 Train Steps: 22/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783], [0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050], [0.6296, 0.4045, 0.9138, 0.4100, 0.4232, 0.4242, 0.7422, 0.5297], [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835], [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350], [0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6024, 0.3981, 0.8128, 0.2590, 0.4556, 0.2491, 0.6595, 0.5584], [0.5549, 0.3711, 0.8749, 0.5209, 0.3631, 0.4099, 0.4960, 0.5748], [0.5520, 0.3803, 0.8692, 0.4278, 0.3688, 0.4868, 0.5845, 0.5944], [0.6277, 0.4208, 0.8766, 0.4481, 0.3989, 0.4240, 0.6891, 0.5475], [0.6073, 0.4031, 0.8811, 0.4218, 0.4144, 0.2082, 0.5623, 0.4811], [0.5931, 0.3978, 0.7811, 0.3136, 0.3934, 0.2683, 0.5637, 0.5376], [0.6119, 0.4190, 0.8744, 0.4121, 0.3878, 0.2531, 0.6042, 0.5076], [0.5991, 0.3918, 0.8556, 0.3521, 0.4053, 0.2951, 0.6678, 0.5565]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783], [0.6200, 0.4112, 0.8863, 0.4100, 0.3638, 0.4917, 0.6087, 0.6050], [0.6296, 0.4045, 0.9137, 0.4100, 0.4232, 0.4242, 0.7422, 0.5297], [0.6273, 0.4110, 0.8900, 0.3817, 0.4187, 0.2167, 0.5858, 0.4835], [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350], [0.6277, 0.4117, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.009468808741075918 step: 23 running loss: 0.00041168733656851816 Train Steps: 23/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267], [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412], [0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500], [0.6087, 0.3951, 0.8387, 0.5833, 0.4188, 0.4933, 0.5146, 0.4830], [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301], [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6148, 0.4041, 0.8155, 0.2686, 0.4448, 0.1657, 0.6586, 0.5228], [0.6045, 0.4073, 0.8558, 0.5662, 0.4406, 0.4801, 0.5450, 0.5344], [0.6265, 0.4334, 0.8728, 0.4929, 0.3422, 0.3535, 0.5344, 0.5387], [0.6185, 0.4159, 0.8433, 0.5764, 0.3751, 0.4169, 0.5593, 0.4895], [0.5972, 0.3902, 0.8796, 0.4573, 0.4146, 0.5011, 0.6148, 0.5498], [0.6242, 0.4118, 0.8383, 0.6012, 0.3855, 0.4754, 0.5471, 0.5078], [0.6271, 0.4017, 0.8617, 0.5176, 0.3983, 0.4823, 0.6515, 0.5281], [0.6338, 0.4183, 0.8594, 0.3378, 0.4281, 0.2242, 0.6443, 0.5128]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267], [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412], [0.6127, 0.4084, 0.8737, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5838, 0.5500], [0.6087, 0.3951, 0.8388, 0.5833, 0.4187, 0.4933, 0.5146, 0.4830], [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301], [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.00983630126575008 step: 24 running loss: 0.00040984588607292 Train Steps: 24/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167], [0.6357, 0.4097, 0.9038, 0.3883, 0.4213, 0.2950, 0.6686, 0.5390], [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6100, 0.4071, 0.7601, 0.3444, 0.3400, 0.4117, 0.5625, 0.5617], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5478, 0.3742, 0.7595, 0.3074, 0.3959, 0.2678, 0.5664, 0.5896], [0.5971, 0.4027, 0.8927, 0.3894, 0.4067, 0.2946, 0.6613, 0.5152], [0.6388, 0.4374, 0.8103, 0.4226, 0.4771, 0.2424, 0.5317, 0.6015], [0.6420, 0.4268, 0.8663, 0.2420, 0.5460, 0.2147, 0.7460, 0.5467], [0.5977, 0.3995, 0.7534, 0.2410, 0.3814, 0.2896, 0.6142, 0.5233], [0.6387, 0.4289, 0.8900, 0.4843, 0.3351, 0.3626, 0.6269, 0.4801], [0.5766, 0.3897, 0.7768, 0.3512, 0.3272, 0.4412, 0.5664, 0.5261], [0.5779, 0.3848, 0.8936, 0.4680, 0.4346, 0.6008, 0.6111, 0.5183]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167], [0.6357, 0.4097, 0.9038, 0.3883, 0.4212, 0.2950, 0.6686, 0.5390], [0.6299, 0.4303, 0.7962, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6100, 0.4071, 0.7601, 0.3444, 0.3400, 0.4117, 0.5625, 0.5617], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.010191787354415283 step: 25 running loss: 0.00040767149417661133 Train Steps: 25/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748], [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267], [0.6138, 0.4101, 0.8800, 0.5083, 0.4637, 0.5950, 0.5587, 0.5077], [0.6157, 0.4102, 0.8513, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890], [0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494], [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6597, 0.4380, 0.9125, 0.4130, 0.4238, 0.3909, 0.7084, 0.5772], [0.6539, 0.4436, 0.8975, 0.3660, 0.3861, 0.2707, 0.6218, 0.4967], [0.6279, 0.4259, 0.8352, 0.5556, 0.3844, 0.4193, 0.7031, 0.5480], [0.6050, 0.4060, 0.8342, 0.5771, 0.4128, 0.4695, 0.5517, 0.5103], [0.5663, 0.3743, 0.8819, 0.5145, 0.4564, 0.5949, 0.5561, 0.5101], [0.6380, 0.4306, 0.8561, 0.3709, 0.3507, 0.3565, 0.5027, 0.5586], [0.6158, 0.4169, 0.9179, 0.3553, 0.4339, 0.3942, 0.7244, 0.5386], [0.6019, 0.4118, 0.8663, 0.4772, 0.4382, 0.4870, 0.5078, 0.5601]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945], [0.6277, 0.4117, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748], [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267], [0.6138, 0.4101, 0.8800, 0.5083, 0.4638, 0.5950, 0.5587, 0.5077], [0.6157, 0.4102, 0.8512, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890], [0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494], [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822]]], device='cuda:0') loss_train_step before backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train: 0.010435777367092669 step: 26 running loss: 0.0004013760525804873 Train Steps: 26/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450], [0.6148, 0.3996, 0.8488, 0.3867, 0.3488, 0.4067, 0.5863, 0.5000], [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459], [0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319], [0.6164, 0.4119, 0.7913, 0.2650, 0.3538, 0.3500, 0.5614, 0.5038], [0.6059, 0.4002, 0.7562, 0.2767, 0.3538, 0.3033, 0.5529, 0.5455], [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050], [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6733, 0.4420, 0.9006, 0.4592, 0.3952, 0.4471, 0.6059, 0.5509], [0.6782, 0.4406, 0.8676, 0.3867, 0.3818, 0.4237, 0.6056, 0.5240], [0.6645, 0.4411, 0.8261, 0.5779, 0.4138, 0.5102, 0.7137, 0.5565], [0.6322, 0.4215, 0.7219, 0.3003, 0.3706, 0.3627, 0.5216, 0.5281], [0.6336, 0.4272, 0.8046, 0.2458, 0.3731, 0.3714, 0.5971, 0.5173], [0.6348, 0.4317, 0.7630, 0.2643, 0.3831, 0.3224, 0.5698, 0.5473], [0.6039, 0.4015, 0.9008, 0.5119, 0.4727, 0.5480, 0.5999, 0.5222], [0.6702, 0.4475, 0.9149, 0.4061, 0.4231, 0.2565, 0.5952, 0.5360]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450], [0.6148, 0.3996, 0.8487, 0.3867, 0.3487, 0.4067, 0.5863, 0.5000], [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459], [0.6109, 0.4041, 0.6975, 0.3167, 0.3512, 0.3383, 0.5153, 0.5319], [0.6164, 0.4119, 0.7912, 0.2650, 0.3537, 0.3500, 0.5614, 0.5038], [0.6059, 0.4002, 0.7563, 0.2767, 0.3537, 0.3033, 0.5529, 0.5455], [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050], [0.6226, 0.4098, 0.8913, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.011048498621676117 step: 27 running loss: 0.000409203652654671 Train Steps: 27/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148], [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467], [0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6048, 0.3928, 0.8538, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719], [0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6253, 0.3900, 0.8995, 0.5372, 0.4009, 0.5019, 0.6102, 0.5218], [0.6573, 0.4399, 0.6953, 0.2518, 0.4098, 0.2716, 0.5794, 0.5534], [0.6543, 0.4098, 0.8893, 0.5400, 0.4585, 0.5274, 0.6245, 0.5081], [0.6368, 0.4064, 0.7912, 0.2724, 0.4639, 0.2089, 0.5825, 0.5365], [0.6131, 0.3953, 0.8726, 0.5570, 0.4134, 0.5330, 0.5629, 0.5069], [0.6109, 0.4030, 0.7297, 0.1998, 0.4061, 0.2826, 0.6012, 0.5700], [0.6804, 0.4367, 0.6936, 0.2261, 0.4120, 0.2187, 0.5295, 0.5860], [0.6497, 0.4117, 0.8310, 0.2360, 0.4834, 0.2101, 0.6582, 0.5428]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148], [0.6162, 0.4134, 0.6700, 0.2467, 0.3963, 0.2533, 0.5738, 0.5467], [0.6219, 0.3934, 0.8687, 0.5267, 0.4313, 0.4967, 0.5987, 0.4983], [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351], [0.6048, 0.3928, 0.8537, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719], [0.6187, 0.4103, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6149, 0.4054, 0.6712, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695], [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.01147236383985728 step: 28 running loss: 0.0004097272799949029 Train Steps: 28/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6200, 0.4070, 0.8938, 0.4183, 0.3538, 0.4567, 0.6175, 0.5400], [0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090], [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183], [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901], [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402], [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167], [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6964, 0.4375, 0.8261, 0.2212, 0.4650, 0.2734, 0.7027, 0.5774], [0.6946, 0.4328, 0.8898, 0.4038, 0.3624, 0.4662, 0.6211, 0.5375], [0.6542, 0.4034, 0.8801, 0.3598, 0.4098, 0.3013, 0.6400, 0.5122], [0.6437, 0.4128, 0.8703, 0.4686, 0.4730, 0.5749, 0.6123, 0.5324], [0.6431, 0.4101, 0.8832, 0.4444, 0.4106, 0.4485, 0.5275, 0.5117], [0.6653, 0.4102, 0.8407, 0.5278, 0.4112, 0.4669, 0.5808, 0.5438], [0.6775, 0.4272, 0.8181, 0.3206, 0.3716, 0.3246, 0.5615, 0.5419], [0.0300, 0.0195, 0.8411, 0.2661, 0.5341, 0.2151, 0.6922, 0.5651]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6200, 0.4070, 0.8938, 0.4183, 0.3537, 0.4567, 0.6175, 0.5400], [0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090], [0.6251, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6087, 0.5183], [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901], [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402], [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167], [0.0000, 0.0000, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.011934660928091034 step: 29 running loss: 0.00041154003200313913 Train Steps: 29/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6339, 0.4118, 0.7988, 0.5800, 0.3912, 0.4583, 0.7343, 0.5760], [0.6205, 0.4081, 0.8950, 0.4017, 0.3788, 0.4700, 0.5963, 0.5667], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683], [ nan, nan, 0.8363, 0.3317, 0.3563, 0.3367, 0.5329, 0.5142], [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726], [ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633], [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.7444, 0.4550, 0.7838, 0.5614, 0.3935, 0.4228, 0.7239, 0.5780], [0.7114, 0.4385, 0.8891, 0.3816, 0.3997, 0.4399, 0.6062, 0.5688], [0.6936, 0.4127, 0.8160, 0.2357, 0.4294, 0.2486, 0.6374, 0.5281], [0.6937, 0.4294, 0.8518, 0.4698, 0.5013, 0.4684, 0.5379, 0.5678], [0.1745, 0.0923, 0.8023, 0.3025, 0.3639, 0.3292, 0.5370, 0.5189], [0.6800, 0.4200, 0.7463, 0.2655, 0.3864, 0.2591, 0.5379, 0.4902], [0.0922, 0.0367, 0.7440, 0.2449, 0.3967, 0.2344, 0.4970, 0.5574], [0.6934, 0.4381, 0.8870, 0.4631, 0.4642, 0.5533, 0.5820, 0.5214]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6339, 0.4118, 0.7987, 0.5800, 0.3913, 0.4583, 0.7343, 0.5760], [0.6205, 0.4081, 0.8950, 0.4017, 0.3787, 0.4700, 0.5962, 0.5667], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683], [0.0000, 0.0000, 0.8363, 0.3317, 0.3562, 0.3367, 0.5329, 0.5142], [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726], [0.0000, 0.0000, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633], [0.6155, 0.4124, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084]]], device='cuda:0') loss_train_step before backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0017, device='cuda:0', grad_fn=) loss_train: 0.013682589953532442 step: 30 running loss: 0.00045608633178441474 Train Steps: 30/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6199, 0.4071, 0.7600, 0.2117, 0.4037, 0.2767, 0.6138, 0.5550], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901], [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6087, 0.3951, 0.8387, 0.5833, 0.4188, 0.4933, 0.5146, 0.4830], [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116], [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6379, 0.3909, 0.7579, 0.2245, 0.4173, 0.2712, 0.6210, 0.5547], [0.6233, 0.3771, 0.8362, 0.4750, 0.3749, 0.4593, 0.5709, 0.5568], [0.5827, 0.3658, 0.7819, 0.3057, 0.3846, 0.2551, 0.5367, 0.5088], [0.6191, 0.3872, 0.8765, 0.5207, 0.4641, 0.4718, 0.5482, 0.5878], [0.6211, 0.3721, 0.8128, 0.2735, 0.4169, 0.2571, 0.6458, 0.5189], [0.6149, 0.3738, 0.8328, 0.6057, 0.4157, 0.4810, 0.5209, 0.4928], [0.6080, 0.3842, 0.7540, 0.1936, 0.3971, 0.2233, 0.5777, 0.4945], [0.5913, 0.3738, 0.8921, 0.5489, 0.3768, 0.4280, 0.5715, 0.5629]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6199, 0.4071, 0.7600, 0.2117, 0.4038, 0.2767, 0.6137, 0.5550], [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901], [0.6140, 0.4070, 0.8700, 0.5000, 0.4613, 0.4900, 0.5260, 0.5852], [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212], [0.6087, 0.3951, 0.8388, 0.5833, 0.4187, 0.4933, 0.5146, 0.4830], [0.6165, 0.4106, 0.7575, 0.1733, 0.3837, 0.2650, 0.5680, 0.5116], [0.6222, 0.4171, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.014052209124201909 step: 31 running loss: 0.0004532970685226422 Train Steps: 31/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012], [0.6198, 0.4101, 0.8838, 0.5283, 0.3763, 0.5267, 0.5913, 0.5567], [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5963, 0.6217], [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250], [ nan, nan, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737], [0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148], [0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5737, 0.5700]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6495, 0.4011, 0.8692, 0.5537, 0.3627, 0.4384, 0.6152, 0.4814], [ 0.5820, 0.3635, 0.8739, 0.5224, 0.3707, 0.5128, 0.5727, 0.5245], [ 0.6267, 0.4012, 0.7082, 0.2621, 0.4269, 0.2425, 0.5802, 0.5959], [ 0.6077, 0.3814, 0.8682, 0.5002, 0.3490, 0.4299, 0.5728, 0.5763], [ 0.6103, 0.3933, 0.8777, 0.4750, 0.4192, 0.5271, 0.5952, 0.5091], [-0.0732, -0.0449, 0.8521, 0.2557, 0.5387, 0.2475, 0.7223, 0.5523], [ 0.5827, 0.3624, 0.8614, 0.5299, 0.3894, 0.4689, 0.5755, 0.4910], [ 0.6196, 0.3949, 0.8621, 0.5466, 0.3767, 0.4803, 0.5679, 0.5404]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012], [0.6198, 0.4101, 0.8838, 0.5283, 0.3762, 0.5267, 0.5913, 0.5567], [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5962, 0.6217], [0.6186, 0.4060, 0.8750, 0.5050, 0.3537, 0.4367, 0.5813, 0.6083], [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250], [0.0000, 0.0000, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737], [0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148], [0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5738, 0.5700]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.014462272200034931 step: 32 running loss: 0.0004519460062510916 Train Steps: 32/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767], [0.6109, 0.4015, 0.7668, 0.3639, 0.3513, 0.3667, 0.5200, 0.5641], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672], [0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285], [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5832, 0.3911, 0.8822, 0.4977, 0.3909, 0.4613, 0.5244, 0.5735], [0.5820, 0.3953, 0.7600, 0.3603, 0.3348, 0.3778, 0.5203, 0.5651], [0.5541, 0.3639, 0.8830, 0.5053, 0.3868, 0.5514, 0.7251, 0.5361], [0.5904, 0.3968, 0.8927, 0.4745, 0.3561, 0.3798, 0.6333, 0.5047], [0.6538, 0.4274, 0.8706, 0.5482, 0.3517, 0.4422, 0.6079, 0.4647], [0.5155, 0.3559, 0.8503, 0.5779, 0.4409, 0.4510, 0.5537, 0.5804], [0.5931, 0.3919, 0.7732, 0.4311, 0.3554, 0.4521, 0.5282, 0.5202], [0.5779, 0.3841, 0.8554, 0.3345, 0.3512, 0.5036, 0.6164, 0.5412]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6163, 0.4001, 0.8788, 0.5033, 0.4013, 0.4633, 0.5337, 0.5767], [0.6109, 0.4015, 0.7668, 0.3639, 0.3512, 0.3667, 0.5200, 0.5641], [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378], [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051], [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672], [0.6248, 0.4185, 0.8500, 0.5767, 0.4462, 0.4550, 0.5612, 0.5917], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285], [0.6168, 0.4029, 0.8523, 0.3417, 0.3587, 0.5000, 0.6125, 0.5400]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.015006558765890077 step: 33 running loss: 0.000454744205026972 Train Steps: 33/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200], [0.6104, 0.4029, 0.8738, 0.4900, 0.4088, 0.4533, 0.5070, 0.5510], [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947], [0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183], [0.6200, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393], [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167], [ nan, nan, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6451, 0.4259, 0.8675, 0.5698, 0.3666, 0.4369, 0.5717, 0.5125], [ 0.5645, 0.3908, 0.8689, 0.5002, 0.3998, 0.4359, 0.5041, 0.5314], [ 0.6088, 0.4146, 0.7864, 0.1810, 0.4185, 0.2331, 0.6403, 0.5081], [ 0.5629, 0.3920, 0.7945, 0.2791, 0.3713, 0.2506, 0.5538, 0.5200], [ 0.6072, 0.3994, 0.8853, 0.4779, 0.3450, 0.4178, 0.5725, 0.5193], [ 0.5482, 0.3731, 0.8415, 0.2738, 0.3922, 0.2913, 0.6542, 0.5502], [ 0.5835, 0.4134, 0.8025, 0.3365, 0.3402, 0.3861, 0.5850, 0.6057], [-0.0904, -0.0457, 0.7108, 0.2879, 0.3720, 0.2219, 0.5418, 0.5538]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200], [0.6104, 0.4029, 0.8737, 0.4900, 0.4087, 0.4533, 0.5070, 0.5510], [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947], [0.6204, 0.4049, 0.7975, 0.2700, 0.3938, 0.2567, 0.5700, 0.5183], [0.6201, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6286, 0.4086, 0.8408, 0.2801, 0.4162, 0.2800, 0.6725, 0.5393], [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6037, 0.6167], [0.0000, 0.0000, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.015572562260786071 step: 34 running loss: 0.00045801653708194326 Train Steps: 34/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6120, 0.4014, 0.6863, 0.2817, 0.3700, 0.2783, 0.5513, 0.5667], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050], [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467], [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800], [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250], [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.5648, 0.3873, 0.7104, 0.2988, 0.3594, 0.2853, 0.5460, 0.5725], [ 0.5962, 0.4024, 0.8461, 0.3272, 0.3354, 0.3844, 0.6203, 0.5423], [ 0.5860, 0.4008, 0.7996, 0.3109, 0.3559, 0.3106, 0.5874, 0.5184], [ 0.5606, 0.4088, 0.6864, 0.2589, 0.3830, 0.2525, 0.5607, 0.5531], [ 0.5616, 0.4096, 0.7585, 0.3136, 0.3572, 0.2629, 0.5283, 0.5779], [ 0.5532, 0.3874, 0.9154, 0.4124, 0.3580, 0.3366, 0.6117, 0.5302], [ 0.5849, 0.4117, 0.9025, 0.4614, 0.3938, 0.5506, 0.6109, 0.5616], [-0.0348, -0.0089, 0.7443, 0.2201, 0.3803, 0.2465, 0.5251, 0.5509]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6120, 0.4013, 0.6862, 0.2817, 0.3700, 0.2783, 0.5512, 0.5667], [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383], [0.6135, 0.3994, 0.7912, 0.3050, 0.3625, 0.3050, 0.5838, 0.5050], [0.6162, 0.4134, 0.6700, 0.2467, 0.3963, 0.2533, 0.5738, 0.5467], [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5462, 0.5800], [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250], [0.6241, 0.4142, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.0000, 0.0000, 0.7425, 0.2117, 0.3938, 0.2433, 0.5437, 0.5567]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.015958790871081874 step: 35 running loss: 0.0004559654534594821 Train Steps: 35/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050], [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933], [0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166], [0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5715, 0.3983, 0.7859, 0.2862, 0.3385, 0.2973, 0.5770, 0.5222], [0.5618, 0.4052, 0.8710, 0.4820, 0.4305, 0.4784, 0.5196, 0.5981], [0.6019, 0.4226, 0.8774, 0.4308, 0.4082, 0.5659, 0.6071, 0.5512], [0.5632, 0.4007, 0.8402, 0.5524, 0.4494, 0.4242, 0.5551, 0.6019], [0.5532, 0.4036, 0.8765, 0.4356, 0.3600, 0.5315, 0.5501, 0.5206], [0.5350, 0.3746, 0.8441, 0.2485, 0.4228, 0.2560, 0.6971, 0.5770], [0.5611, 0.3938, 0.8491, 0.5134, 0.3984, 0.5053, 0.6425, 0.5280], [0.5484, 0.3925, 0.8883, 0.4308, 0.3827, 0.2655, 0.6115, 0.5072]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6135, 0.3994, 0.7912, 0.3050, 0.3625, 0.3050, 0.5838, 0.5050], [0.6140, 0.4070, 0.8700, 0.5000, 0.4613, 0.4900, 0.5260, 0.5852], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933], [0.6186, 0.4129, 0.8900, 0.4567, 0.3938, 0.5417, 0.5734, 0.5110], [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586], [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166], [0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.016739433369366452 step: 36 running loss: 0.00046498426026017923 Train Steps: 36/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263], [0.6058, 0.3986, 0.8324, 0.4626, 0.3838, 0.4983, 0.5147, 0.5466], [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6312, 0.5250], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6122, 0.3993, 0.8738, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5706, 0.3897, 0.8698, 0.3285, 0.3751, 0.2562, 0.5224, 0.5354], [0.5487, 0.3692, 0.7964, 0.2755, 0.3574, 0.3253, 0.5934, 0.5550], [0.5685, 0.3815, 0.7108, 0.2668, 0.4292, 0.2223, 0.5417, 0.6050], [0.6088, 0.4080, 0.7604, 0.2438, 0.4274, 0.1796, 0.5675, 0.5478], [0.5814, 0.3926, 0.8298, 0.4636, 0.3864, 0.4965, 0.5183, 0.5379], [0.5431, 0.3703, 0.8917, 0.3262, 0.3925, 0.3486, 0.6546, 0.5392], [0.5391, 0.3666, 0.8971, 0.5011, 0.3574, 0.3801, 0.6286, 0.4892], [0.5493, 0.3695, 0.8689, 0.4610, 0.4409, 0.4958, 0.5160, 0.4922]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6186, 0.4080, 0.8625, 0.3483, 0.3787, 0.2650, 0.5320, 0.5272], [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367], [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263], [0.6058, 0.3986, 0.8324, 0.4626, 0.3837, 0.4983, 0.5147, 0.5466], [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6313, 0.5250], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6122, 0.3993, 0.8737, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.017385684681357816 step: 37 running loss: 0.0004698833697664275 Train Steps: 37/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683], [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717], [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6115, 0.4005, 0.8838, 0.3867, 0.3763, 0.4700, 0.5800, 0.5550], [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6185, 0.4146, 0.8208, 0.3592, 0.3667, 0.5381, 0.5866, 0.5155], [0.5871, 0.3943, 0.8878, 0.5392, 0.3985, 0.3881, 0.5449, 0.5594], [0.6026, 0.4076, 0.6866, 0.2655, 0.3988, 0.2523, 0.5521, 0.5697], [0.6158, 0.4022, 0.7173, 0.2858, 0.4572, 0.2319, 0.5530, 0.5862], [0.6032, 0.4139, 0.8797, 0.3877, 0.3929, 0.4822, 0.5769, 0.5445], [0.6120, 0.4118, 0.9116, 0.4091, 0.3926, 0.3357, 0.6214, 0.5160], [0.6625, 0.4409, 0.7361, 0.2061, 0.4297, 0.2643, 0.6194, 0.5430], [0.5817, 0.3856, 0.8910, 0.3248, 0.3996, 0.3122, 0.5976, 0.5089]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6132, 0.4118, 0.8200, 0.3633, 0.3562, 0.5400, 0.5787, 0.5136], [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5412, 0.5683], [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717], [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6115, 0.4005, 0.8838, 0.3867, 0.3762, 0.4700, 0.5800, 0.5550], [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117]]], device='cuda:0') loss_train_step before backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train: 0.017598614853341132 step: 38 running loss: 0.00046312144350897714 Train Steps: 38/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5988, 0.5283], [0.6131, 0.4064, 0.8638, 0.5200, 0.4788, 0.4783, 0.5258, 0.5867], [ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6273, 0.4100, 0.7137, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6304, 0.4024, 0.8925, 0.4800, 0.3937, 0.4817, 0.7485, 0.5297]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6718, 0.4344, 0.6978, 0.2278, 0.3973, 0.2618, 0.5353, 0.5026], [0.6814, 0.4551, 0.8863, 0.4626, 0.3983, 0.4597, 0.5437, 0.5441], [0.6454, 0.4267, 0.8699, 0.5458, 0.3998, 0.5120, 0.5762, 0.5138], [0.6757, 0.4400, 0.8705, 0.5272, 0.4855, 0.4925, 0.5310, 0.5669], [0.0959, 0.0477, 0.7053, 0.2470, 0.4459, 0.1847, 0.5273, 0.5702], [0.6634, 0.4301, 0.7144, 0.2299, 0.4122, 0.2717, 0.6030, 0.5523], [0.7147, 0.4760, 0.8374, 0.5446, 0.4280, 0.5935, 0.7018, 0.5490], [0.6632, 0.4300, 0.8874, 0.5051, 0.4079, 0.4964, 0.7220, 0.5106]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6043, 0.4022, 0.6888, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5987, 0.5283], [0.6132, 0.4063, 0.8637, 0.5200, 0.4787, 0.4783, 0.5258, 0.5867], [0.0000, 0.0000, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729], [0.6273, 0.4099, 0.7138, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6304, 0.4024, 0.8925, 0.4800, 0.3938, 0.4817, 0.7485, 0.5297]]], device='cuda:0') loss_train_step before backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0010, device='cuda:0', grad_fn=) loss_train: 0.018550915468949825 step: 39 running loss: 0.00047566449920384167 Train Steps: 39/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6339, 0.4118, 0.7988, 0.5800, 0.3912, 0.4583, 0.7343, 0.5760], [0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110], [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767], [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507], [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700], [0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153], [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6169, 0.3839, 0.6956, 0.2080, 0.4152, 0.2001, 0.5619, 0.5252], [0.7022, 0.4459, 0.7833, 0.5770, 0.3878, 0.4553, 0.7285, 0.5615], [0.7060, 0.4618, 0.8837, 0.4516, 0.4031, 0.5463, 0.5737, 0.5116], [0.6692, 0.4372, 0.8901, 0.4432, 0.3769, 0.3541, 0.5819, 0.5860], [0.6585, 0.4263, 0.7016, 0.2302, 0.4021, 0.2140, 0.5644, 0.5644], [0.1082, 0.0597, 0.6819, 0.2193, 0.4518, 0.2031, 0.5519, 0.5844], [0.6855, 0.4326, 0.8580, 0.5093, 0.4684, 0.5414, 0.5962, 0.5084], [0.6360, 0.4014, 0.8779, 0.4440, 0.4092, 0.4681, 0.5291, 0.5039]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6054, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168], [0.6339, 0.4118, 0.7987, 0.5800, 0.3913, 0.4583, 0.7343, 0.5760], [0.6186, 0.4129, 0.8900, 0.4567, 0.3938, 0.5417, 0.5734, 0.5110], [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767], [0.6072, 0.4029, 0.7038, 0.2150, 0.3913, 0.2267, 0.5516, 0.5507], [0.0000, 0.0000, 0.6413, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700], [0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153], [0.6185, 0.4067, 0.8838, 0.4450, 0.4038, 0.4733, 0.5213, 0.5142]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.01935047988081351 step: 40 running loss: 0.00048376199702033774 Train Steps: 40/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510], [0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433], [0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6198, 0.4115, 0.7762, 0.2717, 0.3713, 0.3200, 0.5837, 0.5683], [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6200, 0.3993, 0.8519, 0.4923, 0.3962, 0.4717, 0.6013, 0.5433]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6516, 0.4067, 0.7639, 0.2336, 0.3768, 0.3196, 0.6163, 0.5172], [0.5634, 0.3469, 0.7085, 0.2859, 0.3648, 0.3074, 0.5218, 0.5476], [0.6775, 0.4278, 0.8840, 0.4648, 0.4622, 0.5190, 0.6203, 0.5268], [0.6685, 0.4191, 0.8493, 0.2953, 0.4208, 0.2197, 0.5933, 0.5217], [0.6808, 0.4332, 0.8404, 0.5141, 0.4195, 0.5606, 0.7106, 0.5578], [0.6479, 0.4090, 0.7783, 0.2634, 0.3727, 0.3091, 0.5903, 0.5485], [0.6854, 0.4254, 0.8617, 0.4614, 0.4806, 0.5480, 0.6056, 0.5317], [0.6326, 0.3945, 0.8462, 0.4798, 0.3954, 0.4619, 0.6060, 0.5428]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510], [0.6250, 0.4054, 0.8770, 0.4723, 0.4663, 0.5367, 0.6162, 0.5433], [0.6199, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391], [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679], [0.6198, 0.4115, 0.7763, 0.2717, 0.3713, 0.3200, 0.5838, 0.5683], [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6200, 0.3993, 0.8519, 0.4923, 0.3963, 0.4717, 0.6012, 0.5433]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.019759502029046416 step: 41 running loss: 0.0004819390738791809 Train Steps: 41/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947], [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467], [0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183], [0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917], [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583], [0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150], [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6332, 0.3977, 0.6971, 0.2190, 0.3922, 0.2525, 0.5969, 0.5592], [0.6500, 0.3900, 0.7688, 0.1750, 0.4375, 0.2283, 0.6539, 0.5060], [0.5769, 0.3600, 0.8426, 0.3437, 0.3481, 0.4058, 0.5943, 0.5408], [0.6526, 0.3892, 0.7577, 0.2210, 0.4365, 0.1818, 0.5951, 0.5200], [0.7137, 0.4460, 0.8387, 0.6048, 0.4453, 0.4621, 0.5559, 0.5927], [0.6569, 0.4075, 0.6810, 0.2255, 0.3898, 0.2605, 0.5880, 0.5565], [0.6181, 0.3658, 0.7769, 0.2475, 0.4378, 0.2099, 0.5868, 0.5283], [0.6375, 0.4019, 0.9070, 0.3575, 0.4146, 0.3607, 0.7427, 0.5510]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6187, 0.4103, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683], [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947], [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467], [0.6213, 0.4001, 0.7713, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183], [0.6248, 0.4185, 0.8500, 0.5767, 0.4462, 0.4550, 0.5612, 0.5917], [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6087, 0.5583], [0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150], [0.6332, 0.4165, 0.9100, 0.3350, 0.4187, 0.3683, 0.7438, 0.5528]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.02012144192121923 step: 42 running loss: 0.00047908195050521974 Train Steps: 42/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8850, 0.2817, 0.5112, 0.2183, 0.7184, 0.5436], [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290], [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683], [0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901], [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033], [0.6245, 0.4115, 0.8700, 0.4883, 0.4625, 0.5517, 0.6100, 0.5217]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.0542, -0.0027, 0.8669, 0.2556, 0.5190, 0.1917, 0.7471, 0.5670], [ 0.6373, 0.3760, 0.7465, 0.2069, 0.4317, 0.2301, 0.6611, 0.5499], [ 0.6509, 0.4078, 0.8469, 0.4730, 0.4775, 0.4878, 0.5511, 0.5893], [ 0.6428, 0.3937, 0.8409, 0.5153, 0.4562, 0.5230, 0.5367, 0.5416], [ 0.6606, 0.4011, 0.8424, 0.3529, 0.3611, 0.4268, 0.5778, 0.5345], [ 0.6471, 0.4007, 0.7889, 0.2801, 0.3805, 0.2498, 0.5351, 0.5312], [ 0.6499, 0.3934, 0.8639, 0.4930, 0.4056, 0.5281, 0.6079, 0.5312], [ 0.6847, 0.4187, 0.8458, 0.4770, 0.4485, 0.5440, 0.6391, 0.5610]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8850, 0.2817, 0.5113, 0.2183, 0.7184, 0.5436], [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290], [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683], [0.6075, 0.4000, 0.8512, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280], [0.6213, 0.4131, 0.8438, 0.3550, 0.3512, 0.4400, 0.5716, 0.5123], [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901], [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033], [0.6245, 0.4115, 0.8700, 0.4883, 0.4625, 0.5517, 0.6100, 0.5217]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.020595025533111766 step: 43 running loss: 0.0004789540821653899 Train Steps: 43/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131], [0.6127, 0.4118, 0.8650, 0.5083, 0.4088, 0.5367, 0.5300, 0.5456], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285], [0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882], [0.6259, 0.4133, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5879, 0.3822, 0.8430, 0.3862, 0.3465, 0.4548, 0.5848, 0.5375], [0.6455, 0.4225, 0.8797, 0.4847, 0.4201, 0.5532, 0.5451, 0.5790], [0.5972, 0.3725, 0.7776, 0.4046, 0.3612, 0.4352, 0.5180, 0.5511], [0.5742, 0.3705, 0.8650, 0.5016, 0.3802, 0.4013, 0.5332, 0.5245], [0.6719, 0.4335, 0.8126, 0.2123, 0.5088, 0.1532, 0.6522, 0.5176], [0.6547, 0.4174, 0.8908, 0.4526, 0.4380, 0.5870, 0.6292, 0.5493], [0.6425, 0.4106, 0.8989, 0.4601, 0.4023, 0.5418, 0.7542, 0.5687], [0.6464, 0.4193, 0.7281, 0.1871, 0.4320, 0.2155, 0.6112, 0.5420]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131], [0.6127, 0.4118, 0.8650, 0.5083, 0.4087, 0.5367, 0.5300, 0.5456], [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285], [0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882], [0.6259, 0.4132, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947], [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150], [0.6339, 0.4102, 0.9087, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.02098468688200228 step: 44 running loss: 0.00047692470186368814 Train Steps: 44/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6250, 0.4106, 0.8700, 0.3717, 0.3588, 0.4967, 0.6038, 0.5167], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6284, 0.4127, 0.8538, 0.5867, 0.4363, 0.5083, 0.6038, 0.5433], [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500], [0.6252, 0.4158, 0.8988, 0.4083, 0.3788, 0.4783, 0.6225, 0.5633], [0.6199, 0.4065, 0.7598, 0.2385, 0.4317, 0.1981, 0.5933, 0.5221], [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540], [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5912, 0.3878, 0.8807, 0.3672, 0.3595, 0.4962, 0.6225, 0.5206], [0.5541, 0.3758, 0.7213, 0.2091, 0.3880, 0.2429, 0.5415, 0.5559], [0.5959, 0.3878, 0.8543, 0.5744, 0.4372, 0.5065, 0.5773, 0.5287], [0.5881, 0.3708, 0.8818, 0.4274, 0.4238, 0.5098, 0.5767, 0.5491], [0.6098, 0.4083, 0.8911, 0.3873, 0.3684, 0.4666, 0.6218, 0.5463], [0.6152, 0.3969, 0.7550, 0.2310, 0.4111, 0.2159, 0.5823, 0.5025], [0.5669, 0.3730, 0.9082, 0.4236, 0.4116, 0.3081, 0.6940, 0.5416], [0.6773, 0.4370, 0.8223, 0.2095, 0.4636, 0.1752, 0.6199, 0.4743]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6250, 0.4105, 0.8700, 0.3717, 0.3587, 0.4967, 0.6037, 0.5167], [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663], [0.6284, 0.4127, 0.8537, 0.5867, 0.4363, 0.5083, 0.6037, 0.5433], [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5838, 0.5500], [0.6252, 0.4158, 0.8988, 0.4083, 0.3787, 0.4783, 0.6225, 0.5633], [0.6199, 0.4065, 0.7598, 0.2385, 0.4317, 0.1981, 0.5933, 0.5221], [0.6300, 0.4102, 0.9087, 0.4433, 0.4087, 0.3067, 0.6820, 0.5540], [0.6276, 0.4095, 0.8238, 0.2250, 0.4663, 0.1783, 0.6171, 0.4869]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.021359838166972622 step: 45 running loss: 0.0004746630703771694 Train Steps: 45/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019], [0.6218, 0.4137, 0.7263, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783], [0.6145, 0.4007, 0.8775, 0.4533, 0.4562, 0.5533, 0.6088, 0.5533], [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633], [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552], [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767], [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400], [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5748, 0.3760, 0.9063, 0.4830, 0.3700, 0.4068, 0.6425, 0.4730], [0.6236, 0.4329, 0.7296, 0.2235, 0.3907, 0.2747, 0.6246, 0.5548], [0.5967, 0.4017, 0.8822, 0.4524, 0.4647, 0.5588, 0.6056, 0.5269], [0.6249, 0.4424, 0.8794, 0.4925, 0.3538, 0.3837, 0.5687, 0.5527], [0.0318, 0.0366, 0.8491, 0.2514, 0.5056, 0.2510, 0.7441, 0.5336], [0.6129, 0.4187, 0.8801, 0.5066, 0.4024, 0.4690, 0.5349, 0.5547], [0.5513, 0.3737, 0.8868, 0.4114, 0.3565, 0.4728, 0.6038, 0.5205], [0.6103, 0.4197, 0.8110, 0.4017, 0.3478, 0.3106, 0.5409, 0.5645]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019], [0.6218, 0.4137, 0.7262, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783], [0.6145, 0.4007, 0.8775, 0.4533, 0.4563, 0.5533, 0.6087, 0.5533], [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5638, 0.5633], [0.0000, 0.0000, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552], [0.6163, 0.4001, 0.8788, 0.5033, 0.4013, 0.4633, 0.5337, 0.5767], [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6137, 0.5400], [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.021688233740860596 step: 46 running loss: 0.00047148334219262165 Train Steps: 46/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383], [ nan, nan, 0.8488, 0.2300, 0.5563, 0.2100, 0.7390, 0.5679], [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690], [0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.0505, 0.0629, 0.7221, 0.2172, 0.3990, 0.1907, 0.4836, 0.5072], [0.0711, 0.0606, 0.8532, 0.2203, 0.5276, 0.2427, 0.7032, 0.5434], [0.6273, 0.4385, 0.9236, 0.3451, 0.4138, 0.3583, 0.7116, 0.5344], [0.6543, 0.4615, 0.8308, 0.5804, 0.3941, 0.4709, 0.5442, 0.5924], [0.6330, 0.4136, 0.8432, 0.5356, 0.4166, 0.5505, 0.7111, 0.5536], [0.6060, 0.4188, 0.8725, 0.4506, 0.4011, 0.5457, 0.5526, 0.5252], [0.6341, 0.4384, 0.9139, 0.5257, 0.3681, 0.3829, 0.6115, 0.4767], [0.6443, 0.4431, 0.8885, 0.5408, 0.3817, 0.3726, 0.5642, 0.5151]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.7268, 0.2333, 0.4125, 0.1933, 0.5113, 0.5383], [0.0000, 0.0000, 0.8487, 0.2300, 0.5562, 0.2100, 0.7390, 0.5679], [0.6332, 0.4165, 0.9100, 0.3350, 0.4187, 0.3683, 0.7438, 0.5528], [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118], [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690], [0.6064, 0.4019, 0.8650, 0.4517, 0.4038, 0.5367, 0.5703, 0.5609], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.022245831292821094 step: 47 running loss: 0.0004733155594217254 Train Steps: 47/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950], [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617], [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6198, 0.4115, 0.7762, 0.2717, 0.3713, 0.3200, 0.5837, 0.5683]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5713, 0.3957, 0.8536, 0.5315, 0.3781, 0.5025, 0.6913, 0.5320], [0.5430, 0.3901, 0.7948, 0.3414, 0.3455, 0.3305, 0.5656, 0.5079], [0.5416, 0.3661, 0.8745, 0.5353, 0.4398, 0.4958, 0.5801, 0.4915], [0.6007, 0.4279, 0.9035, 0.5517, 0.3764, 0.4135, 0.5504, 0.5678], [0.6196, 0.4281, 0.8863, 0.5627, 0.3712, 0.4381, 0.6375, 0.5271], [0.6077, 0.4280, 0.9010, 0.4555, 0.3897, 0.4254, 0.5202, 0.5663], [0.5874, 0.4125, 0.7613, 0.2292, 0.3627, 0.3174, 0.5922, 0.5303], [0.5564, 0.3966, 0.7779, 0.2761, 0.3650, 0.3021, 0.5742, 0.5564]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950], [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617], [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297], [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617], [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334], [0.6198, 0.4115, 0.7763, 0.2717, 0.3713, 0.3200, 0.5838, 0.5683]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.02273709830478765 step: 48 running loss: 0.0004736895480164094 Train Steps: 48/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393], [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6138, 0.5450], [0.6246, 0.4008, 0.8757, 0.5088, 0.4101, 0.5392, 0.6644, 0.5133], [0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626], [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233], [0.6127, 0.4115, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495], [0.6206, 0.4001, 0.8900, 0.3933, 0.3588, 0.3567, 0.5837, 0.5083], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5733, 0.3989, 0.8587, 0.2864, 0.4182, 0.2777, 0.6666, 0.5564], [0.5894, 0.4048, 0.8967, 0.4735, 0.4157, 0.5211, 0.6128, 0.5577], [0.5565, 0.3758, 0.9014, 0.5329, 0.4087, 0.5500, 0.6531, 0.5286], [0.5870, 0.3902, 0.8603, 0.6248, 0.4068, 0.4734, 0.6039, 0.4739], [0.5898, 0.4098, 0.9070, 0.4832, 0.3771, 0.5000, 0.6268, 0.5352], [0.5793, 0.4194, 0.7333, 0.2957, 0.3599, 0.2952, 0.5088, 0.5769], [0.5803, 0.4017, 0.9111, 0.4044, 0.3606, 0.3576, 0.5664, 0.5195], [0.5550, 0.3854, 0.8797, 0.2995, 0.4887, 0.2015, 0.6591, 0.5446]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6286, 0.4086, 0.8408, 0.2801, 0.4162, 0.2800, 0.6725, 0.5393], [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6137, 0.5450], [0.6246, 0.4008, 0.8757, 0.5088, 0.4101, 0.5392, 0.6644, 0.5133], [0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626], [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233], [0.6127, 0.4114, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495], [0.6206, 0.4001, 0.8900, 0.3933, 0.3587, 0.3567, 0.5838, 0.5083], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.023219668451929465 step: 49 running loss: 0.0004738707847332544 Train Steps: 49/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6200, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6200, 0.4049, 0.8638, 0.5617, 0.4125, 0.5100, 0.6013, 0.5317], [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385], [0.6226, 0.4103, 0.8575, 0.3450, 0.4388, 0.2067, 0.5787, 0.5383], [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5143, 0.3352, 0.7564, 0.1950, 0.4566, 0.1685, 0.6002, 0.4866], [0.6141, 0.3955, 0.8951, 0.4780, 0.3630, 0.4330, 0.5919, 0.5264], [0.6263, 0.4158, 0.8824, 0.5108, 0.3786, 0.3296, 0.6314, 0.5002], [0.6027, 0.3875, 0.8526, 0.3889, 0.3475, 0.4036, 0.6088, 0.5667], [0.5613, 0.3617, 0.8650, 0.5623, 0.4191, 0.5066, 0.6014, 0.5359], [0.5758, 0.3717, 0.8225, 0.2557, 0.4555, 0.2425, 0.6712, 0.5450], [0.5658, 0.3698, 0.8539, 0.3462, 0.4375, 0.2139, 0.5617, 0.5394], [0.5803, 0.3703, 0.8256, 0.5701, 0.4086, 0.4908, 0.5831, 0.6090]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895], [0.6201, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148], [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967], [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583], [0.6199, 0.4049, 0.8637, 0.5617, 0.4125, 0.5100, 0.6012, 0.5317], [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385], [0.6226, 0.4103, 0.8575, 0.3450, 0.4387, 0.2067, 0.5788, 0.5383], [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.023799304181011394 step: 50 running loss: 0.0004759860836202279 Train Steps: 50/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283], [0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917], [0.6336, 0.4086, 0.8900, 0.3950, 0.3900, 0.2950, 0.6504, 0.5066], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833], [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367], [0.6201, 0.4064, 0.8688, 0.5050, 0.4225, 0.5100, 0.6138, 0.5500]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6089, 0.3839, 0.8779, 0.4035, 0.3542, 0.4681, 0.6114, 0.5448], [0.6044, 0.3894, 0.8943, 0.3586, 0.4072, 0.2674, 0.6360, 0.5405], [0.5996, 0.3847, 0.8351, 0.5734, 0.4571, 0.4574, 0.5717, 0.5991], [0.6905, 0.4306, 0.9103, 0.3863, 0.3879, 0.2945, 0.6742, 0.5114], [0.6187, 0.3879, 0.8670, 0.5505, 0.3616, 0.3690, 0.5881, 0.5461], [0.4900, 0.3153, 0.6603, 0.3017, 0.3806, 0.3047, 0.5667, 0.5835], [0.6272, 0.4046, 0.7958, 0.2784, 0.4054, 0.2550, 0.6150, 0.5347], [0.5972, 0.3814, 0.8602, 0.5074, 0.4322, 0.5232, 0.6212, 0.5594]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6137, 0.5400], [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6187, 0.5283], [0.6248, 0.4185, 0.8500, 0.5767, 0.4462, 0.4550, 0.5612, 0.5917], [0.6336, 0.4086, 0.8900, 0.3950, 0.3900, 0.2950, 0.6504, 0.5066], [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833], [0.6204, 0.4110, 0.7912, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367], [0.6201, 0.4064, 0.8687, 0.5050, 0.4225, 0.5100, 0.6137, 0.5500]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.024365292192669585 step: 51 running loss: 0.0004777508273072468 Train Steps: 51/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249], [0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297], [0.6102, 0.4020, 0.8638, 0.3717, 0.3625, 0.5017, 0.6038, 0.5500], [0.6263, 0.4030, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771], [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133], [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118], [0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921], [0.6277, 0.4083, 0.8350, 0.2717, 0.4562, 0.1800, 0.5918, 0.4878]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6439, 0.4036, 0.8803, 0.5353, 0.3712, 0.4872, 0.6681, 0.5360], [0.6116, 0.3821, 0.8909, 0.4335, 0.4142, 0.3066, 0.6441, 0.5334], [0.5969, 0.3855, 0.8576, 0.3654, 0.3585, 0.4943, 0.6255, 0.5415], [0.6176, 0.3881, 0.8856, 0.4795, 0.3741, 0.5054, 0.6522, 0.4894], [0.6183, 0.3785, 0.8550, 0.5274, 0.4576, 0.5067, 0.5971, 0.5188], [0.5650, 0.3853, 0.7067, 0.3604, 0.4218, 0.2029, 0.5690, 0.6241], [0.6409, 0.4029, 0.8242, 0.2356, 0.3934, 0.2916, 0.6437, 0.5114], [0.5672, 0.3577, 0.8080, 0.2659, 0.4671, 0.1860, 0.5837, 0.4921]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249], [0.6364, 0.4165, 0.9087, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297], [0.6102, 0.4020, 0.8637, 0.3717, 0.3625, 0.5017, 0.6037, 0.5500], [0.6263, 0.4029, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771], [0.6202, 0.4053, 0.8637, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133], [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118], [0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921], [0.6277, 0.4083, 0.8350, 0.2717, 0.4563, 0.1800, 0.5918, 0.4878]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.02468289455282502 step: 52 running loss: 0.00047467104909278883 Train Steps: 52/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767], [ nan, nan, 0.7225, 0.2167, 0.3987, 0.2283, 0.5427, 0.5181], [0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748], [0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6200, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6829, 0.4331, 0.8681, 0.4453, 0.3672, 0.3511, 0.5824, 0.5813], [0.0488, 0.0061, 0.7005, 0.2047, 0.4013, 0.2191, 0.5391, 0.5093], [0.6749, 0.4260, 0.8833, 0.5126, 0.4011, 0.3471, 0.7377, 0.5690], [0.6430, 0.4035, 0.8453, 0.2920, 0.4334, 0.2236, 0.6163, 0.5345], [0.6359, 0.3727, 0.8428, 0.5143, 0.4045, 0.5073, 0.6094, 0.4836], [0.6329, 0.4247, 0.7100, 0.2810, 0.4323, 0.2096, 0.5824, 0.5666], [0.6682, 0.4187, 0.8638, 0.4755, 0.4442, 0.5742, 0.5900, 0.5133], [0.6326, 0.3849, 0.8501, 0.2994, 0.4474, 0.2098, 0.6189, 0.5257]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767], [0.0000, 0.0000, 0.7225, 0.2167, 0.3988, 0.2283, 0.5427, 0.5181], [0.6336, 0.4191, 0.8938, 0.5167, 0.3938, 0.3517, 0.7343, 0.5748], [0.6250, 0.4131, 0.8687, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6199, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913], [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773], [0.6155, 0.4124, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084], [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.024981963797472417 step: 53 running loss: 0.00047135780749947955 Train Steps: 53/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433], [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6127, 0.4115, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495], [ nan, nan, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6164, 0.4119, 0.7913, 0.2650, 0.3538, 0.3500, 0.5614, 0.5038], [0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6558, 0.4323, 0.7786, 0.2756, 0.3678, 0.3484, 0.6330, 0.5481], [ 0.6911, 0.4423, 0.8657, 0.4683, 0.4272, 0.5781, 0.5810, 0.5236], [ 0.6710, 0.4255, 0.8808, 0.4174, 0.3541, 0.3507, 0.6189, 0.5552], [ 0.6651, 0.4252, 0.8548, 0.4724, 0.3746, 0.5103, 0.6280, 0.5095], [ 0.6851, 0.4552, 0.7107, 0.2866, 0.3654, 0.2804, 0.5467, 0.5620], [-0.0314, -0.0217, 0.8840, 0.3413, 0.5100, 0.1859, 0.7191, 0.5681], [ 0.6652, 0.4315, 0.7833, 0.2691, 0.3491, 0.3422, 0.5961, 0.5102], [ 0.6593, 0.4110, 0.8902, 0.4746, 0.3892, 0.4822, 0.6407, 0.4775]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6175, 0.4091, 0.7862, 0.2800, 0.3638, 0.3583, 0.6187, 0.5433], [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098], [0.6127, 0.4114, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495], [0.0000, 0.0000, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609], [0.6164, 0.4119, 0.7912, 0.2650, 0.3537, 0.3500, 0.5614, 0.5038], [0.6264, 0.4049, 0.8988, 0.4633, 0.3812, 0.4983, 0.6326, 0.4843]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.025597798521630466 step: 54 running loss: 0.00047403330595611977 Train Steps: 54/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471], [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413], [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633], [ nan, nan, 0.7335, 0.2569, 0.3788, 0.2667, 0.5066, 0.5578], [0.6109, 0.4015, 0.7668, 0.3639, 0.3513, 0.3667, 0.5200, 0.5641], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6535, 0.4354, 0.8624, 0.4512, 0.3896, 0.4346, 0.5115, 0.5233], [ 0.6379, 0.4002, 0.9202, 0.4149, 0.4190, 0.3058, 0.7154, 0.5308], [ 0.6623, 0.4254, 0.8586, 0.5727, 0.4205, 0.5445, 0.6047, 0.5441], [-0.0117, 0.0061, 0.7192, 0.2681, 0.3944, 0.2608, 0.5313, 0.5637], [ 0.6767, 0.4417, 0.7500, 0.3647, 0.3468, 0.3627, 0.5505, 0.5424], [ 0.6308, 0.4081, 0.8827, 0.4676, 0.3585, 0.3575, 0.6272, 0.4758], [ 0.6622, 0.4342, 0.8869, 0.4127, 0.4088, 0.2478, 0.5990, 0.5257], [ 0.6462, 0.4288, 0.6884, 0.2136, 0.3919, 0.2280, 0.5715, 0.5308]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6127, 0.4084, 0.8700, 0.4467, 0.3988, 0.4317, 0.5013, 0.5471], [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413], [0.6192, 0.4128, 0.8512, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633], [0.0000, 0.0000, 0.7335, 0.2569, 0.3787, 0.2667, 0.5066, 0.5578], [0.6109, 0.4015, 0.7668, 0.3639, 0.3512, 0.3667, 0.5200, 0.5641], [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6226, 0.4098, 0.8913, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367], [0.6072, 0.4029, 0.7038, 0.2150, 0.3913, 0.2267, 0.5516, 0.5507]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.02595448528882116 step: 55 running loss: 0.0004718997325240211 Train Steps: 55/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374], [0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773], [0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626], [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064], [0.6289, 0.4024, 0.9088, 0.4567, 0.3937, 0.5633, 0.7058, 0.5609]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6168, 0.4003, 0.8846, 0.4705, 0.4641, 0.5626, 0.5990, 0.5645], [0.6576, 0.4380, 0.8644, 0.3584, 0.3492, 0.3211, 0.5855, 0.5366], [0.6508, 0.4180, 0.8447, 0.2560, 0.4321, 0.1896, 0.6037, 0.4935], [0.6707, 0.4402, 0.9167, 0.3831, 0.3854, 0.2592, 0.6097, 0.5180], [0.6497, 0.4374, 0.6726, 0.2335, 0.3940, 0.2255, 0.5302, 0.5871], [0.6735, 0.4251, 0.8556, 0.6107, 0.3848, 0.4843, 0.5976, 0.4758], [0.6407, 0.4308, 0.8873, 0.5028, 0.4670, 0.5341, 0.5162, 0.5112], [0.6604, 0.4275, 0.9052, 0.4598, 0.3730, 0.5711, 0.7126, 0.5736]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600], [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374], [0.6277, 0.4056, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823], [0.6277, 0.4117, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938], [0.6147, 0.4026, 0.6600, 0.2467, 0.4087, 0.2150, 0.5489, 0.5773], [0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626], [0.6102, 0.4005, 0.8687, 0.5100, 0.4812, 0.5400, 0.5404, 0.5064], [0.6289, 0.4024, 0.9087, 0.4567, 0.3938, 0.5633, 0.7058, 0.5609]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.026293839560821652 step: 56 running loss: 0.0004695328493003866 Train Steps: 56/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633], [0.6261, 0.4066, 0.8325, 0.2150, 0.4763, 0.2667, 0.7002, 0.5633], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333], [ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533], [0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100], [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6809, 0.4527, 0.8958, 0.5070, 0.3730, 0.3764, 0.5636, 0.5649], [0.6687, 0.4385, 0.8463, 0.2286, 0.4897, 0.2646, 0.7110, 0.5588], [0.6743, 0.4415, 0.8559, 0.4503, 0.3849, 0.4493, 0.5528, 0.5542], [0.6415, 0.4214, 0.9230, 0.4754, 0.3986, 0.4939, 0.6279, 0.5129], [0.0709, 0.0313, 0.8239, 0.2984, 0.3913, 0.2874, 0.5317, 0.5538], [0.6992, 0.4456, 0.8838, 0.5829, 0.3848, 0.4956, 0.5796, 0.5727], [0.6580, 0.4399, 0.7690, 0.1786, 0.3969, 0.2561, 0.5709, 0.4883], [0.7060, 0.4708, 0.7031, 0.2578, 0.4217, 0.2420, 0.5853, 0.5384]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5638, 0.5633], [0.6261, 0.4066, 0.8325, 0.2150, 0.4762, 0.2667, 0.7002, 0.5633], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6137, 0.5333], [0.0000, 0.0000, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533], [0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100], [0.6165, 0.4106, 0.7575, 0.1733, 0.3837, 0.2650, 0.5680, 0.5116], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517]]], device='cuda:0') loss_train_step before backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0008, device='cuda:0', grad_fn=) loss_train: 0.02706561255035922 step: 57 running loss: 0.0004748353079010389 Train Steps: 57/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6125, 0.4035, 0.7825, 0.3100, 0.3463, 0.4900, 0.5832, 0.5637], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550], [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.0304, 0.0195, 0.8875, 0.2379, 0.5391, 0.2203, 0.7275, 0.5593], [0.6255, 0.4086, 0.8686, 0.5094, 0.4391, 0.5338, 0.5083, 0.5156], [0.6054, 0.4096, 0.7995, 0.3114, 0.3484, 0.4842, 0.5791, 0.5237], [0.6546, 0.4381, 0.7531, 0.3264, 0.4934, 0.1947, 0.5435, 0.6094], [0.6182, 0.4169, 0.8919, 0.4782, 0.3639, 0.4761, 0.5869, 0.5366], [0.6371, 0.4349, 0.8747, 0.5479, 0.4648, 0.4981, 0.5032, 0.5233], [0.6280, 0.4171, 0.7957, 0.3231, 0.3506, 0.3399, 0.5864, 0.4858], [0.6626, 0.4187, 0.8417, 0.5371, 0.4203, 0.5553, 0.7206, 0.5507]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737], [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466], [0.6125, 0.4035, 0.7825, 0.3100, 0.3462, 0.4900, 0.5832, 0.5637], [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167], [0.6197, 0.4091, 0.8800, 0.4783, 0.3537, 0.4767, 0.5950, 0.5550], [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412], [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153], [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690]]], device='cuda:0') loss_train_step before backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train: 0.0273065606597811 step: 58 running loss: 0.00047080276999622583 Train Steps: 58/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501], [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058], [0.6201, 0.4017, 0.8871, 0.4621, 0.3517, 0.4675, 0.5999, 0.5106], [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795], [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267], [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6222, 0.4169, 0.8638, 0.5650, 0.4313, 0.4783, 0.5637, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5931, 0.4097, 0.8874, 0.4887, 0.3945, 0.4566, 0.4961, 0.5504], [0.6113, 0.4209, 0.8224, 0.2378, 0.4481, 0.2116, 0.5767, 0.5303], [0.6100, 0.4114, 0.8547, 0.4254, 0.3745, 0.4647, 0.5253, 0.5002], [0.5575, 0.3701, 0.8966, 0.4368, 0.3579, 0.4684, 0.5695, 0.5236], [0.6361, 0.4070, 0.8598, 0.5326, 0.4002, 0.5371, 0.7005, 0.5858], [0.5959, 0.3916, 0.8418, 0.5647, 0.4152, 0.4711, 0.5487, 0.5289], [0.6106, 0.4112, 0.8561, 0.2199, 0.4688, 0.2690, 0.6868, 0.5519], [0.6009, 0.4062, 0.8732, 0.5490, 0.4295, 0.4702, 0.5407, 0.5784]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524], [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501], [0.6084, 0.4005, 0.8400, 0.4317, 0.3762, 0.4750, 0.5476, 0.5058], [0.6201, 0.4017, 0.8871, 0.4621, 0.3517, 0.4675, 0.5999, 0.5106], [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795], [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267], [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633], [0.6222, 0.4169, 0.8637, 0.5650, 0.4313, 0.4783, 0.5638, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train: 0.027555422129807994 step: 59 running loss: 0.00046704105304759314 Train Steps: 59/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626], [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274], [0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167], [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822], [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385], [0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207], [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5805, 0.3695, 0.8528, 0.5969, 0.3916, 0.4936, 0.5851, 0.4726], [0.6133, 0.4126, 0.8841, 0.3074, 0.4825, 0.2126, 0.6232, 0.5347], [0.5386, 0.3626, 0.7153, 0.2838, 0.4685, 0.2080, 0.5480, 0.6325], [0.5736, 0.3836, 0.8924, 0.5182, 0.4034, 0.5241, 0.5703, 0.5236], [0.5928, 0.4034, 0.8754, 0.4717, 0.4411, 0.5010, 0.4946, 0.5865], [0.6145, 0.3952, 0.8317, 0.2403, 0.4494, 0.2615, 0.6516, 0.5482], [0.5552, 0.3814, 0.7775, 0.2858, 0.3488, 0.4323, 0.5621, 0.5333], [0.5896, 0.3951, 0.8348, 0.2885, 0.4179, 0.2225, 0.5542, 0.5421]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626], [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274], [0.6270, 0.4266, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167], [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822], [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385], [0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207], [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.028138416208093986 step: 60 running loss: 0.0004689736034682331 Train Steps: 60/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6268, 0.4061, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300], [ nan, nan, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983], [0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517], [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323], [0.6198, 0.4101, 0.8838, 0.5283, 0.3763, 0.5267, 0.5913, 0.5567], [0.6122, 0.3993, 0.8738, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927], [0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350], [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6601, 0.4249, 0.8365, 0.2621, 0.4635, 0.2275, 0.6326, 0.5491], [-0.0086, -0.0145, 0.7348, 0.2205, 0.4374, 0.1838, 0.5444, 0.5287], [ 0.5796, 0.3725, 0.8702, 0.4454, 0.4151, 0.5646, 0.6027, 0.5698], [ 0.6180, 0.3984, 0.7739, 0.2203, 0.4497, 0.2328, 0.6333, 0.5667], [ 0.5950, 0.3945, 0.8707, 0.5517, 0.3807, 0.5406, 0.5929, 0.5902], [ 0.5702, 0.3789, 0.8715, 0.4924, 0.4610, 0.5067, 0.5206, 0.5155], [ 0.5951, 0.3871, 0.7522, 0.2959, 0.4304, 0.2437, 0.5927, 0.5768], [ 0.5827, 0.3908, 0.8806, 0.5121, 0.3746, 0.4485, 0.5799, 0.5850]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6268, 0.4060, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300], [0.0000, 0.0000, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983], [0.6122, 0.4006, 0.8850, 0.4217, 0.4087, 0.5517, 0.6062, 0.5517], [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323], [0.6198, 0.4101, 0.8838, 0.5283, 0.3762, 0.5267, 0.5913, 0.5567], [0.6122, 0.3993, 0.8737, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927], [0.6229, 0.4066, 0.7613, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350], [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.028534956247312948 step: 61 running loss: 0.00046778616798873684 Train Steps: 61/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082], [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312], [0.6234, 0.4179, 0.7825, 0.3450, 0.3813, 0.2867, 0.5675, 0.5617], [ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100], [0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5669, 0.3619, 0.8556, 0.5292, 0.4412, 0.5751, 0.6005, 0.5261], [0.6083, 0.3899, 0.8528, 0.4225, 0.3767, 0.4475, 0.6131, 0.5482], [0.5707, 0.3839, 0.7689, 0.3624, 0.4002, 0.2881, 0.5756, 0.5738], [0.0472, 0.0215, 0.7341, 0.2395, 0.3982, 0.2607, 0.5287, 0.5618], [0.6124, 0.3909, 0.8916, 0.4302, 0.3773, 0.3933, 0.6496, 0.5194], [0.6038, 0.3976, 0.8544, 0.3207, 0.4519, 0.2539, 0.6068, 0.5451], [0.6097, 0.3940, 0.8519, 0.4202, 0.4003, 0.4898, 0.5916, 0.5555], [0.5586, 0.3608, 0.8751, 0.4247, 0.3675, 0.3782, 0.5871, 0.5745]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082], [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312], [0.6234, 0.4179, 0.7825, 0.3450, 0.3812, 0.2867, 0.5675, 0.5617], [0.0000, 0.0000, 0.7425, 0.2117, 0.3938, 0.2433, 0.5437, 0.5567], [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100], [0.6250, 0.4131, 0.8687, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367], [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.028984240110730752 step: 62 running loss: 0.00046748774372146375 Train Steps: 62/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012], [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748], [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999], [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6362, 0.5200], [0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393], [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5621, 0.3566, 0.8446, 0.5022, 0.4237, 0.4701, 0.5275, 0.5495], [0.6145, 0.3894, 0.8666, 0.4223, 0.3447, 0.4410, 0.6383, 0.5094], [0.5641, 0.3690, 0.8509, 0.5467, 0.3611, 0.4495, 0.6190, 0.5186], [0.5908, 0.3902, 0.8255, 0.5633, 0.3920, 0.4061, 0.7199, 0.5706], [0.6059, 0.3850, 0.8353, 0.4464, 0.4018, 0.5076, 0.5215, 0.4927], [0.5895, 0.3806, 0.8615, 0.3665, 0.3606, 0.5174, 0.6403, 0.5171], [0.6403, 0.4124, 0.8280, 0.2962, 0.4103, 0.2930, 0.6851, 0.5467], [0.6094, 0.4098, 0.7718, 0.3154, 0.4044, 0.2346, 0.5285, 0.5751]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6126, 0.3954, 0.8537, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350], [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927], [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012], [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748], [0.6068, 0.3963, 0.8650, 0.4317, 0.4038, 0.5083, 0.5253, 0.4999], [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6363, 0.5200], [0.6286, 0.4086, 0.8408, 0.2801, 0.4162, 0.2800, 0.6725, 0.5393], [0.6182, 0.4099, 0.7812, 0.3000, 0.3938, 0.2367, 0.5325, 0.5750]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.029316391359316185 step: 63 running loss: 0.0004653395453859712 Train Steps: 63/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6147, 0.4081, 0.8538, 0.3400, 0.3663, 0.3150, 0.5142, 0.4875], [0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350], [0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5887, 0.5017], [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719], [0.6261, 0.4029, 0.8720, 0.3364, 0.3665, 0.3753, 0.6531, 0.5183], [0.6097, 0.4024, 0.8488, 0.3717, 0.3875, 0.5517, 0.5836, 0.5591], [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6310, 0.4094, 0.8501, 0.3549, 0.3542, 0.2914, 0.5601, 0.4993], [0.6054, 0.3872, 0.7358, 0.2707, 0.4196, 0.2048, 0.6096, 0.5439], [0.6054, 0.3931, 0.8595, 0.4581, 0.3593, 0.4813, 0.5988, 0.5172], [0.5903, 0.3818, 0.8267, 0.5510, 0.3844, 0.4066, 0.5926, 0.5354], [0.6094, 0.3951, 0.8160, 0.3809, 0.3438, 0.4054, 0.5520, 0.5542], [0.6233, 0.3981, 0.8425, 0.3481, 0.3685, 0.3532, 0.6469, 0.5067], [0.5660, 0.3539, 0.8270, 0.3669, 0.3908, 0.5373, 0.6171, 0.5336], [0.5670, 0.3656, 0.8518, 0.3878, 0.3705, 0.3865, 0.5457, 0.4974]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6147, 0.4081, 0.8537, 0.3400, 0.3663, 0.3150, 0.5142, 0.4875], [0.6229, 0.4066, 0.7613, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350], [0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5888, 0.5017], [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283], [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719], [0.6261, 0.4029, 0.8720, 0.3364, 0.3665, 0.3753, 0.6531, 0.5183], [0.6097, 0.4024, 0.8487, 0.3717, 0.3875, 0.5517, 0.5836, 0.5591], [0.6082, 0.4024, 0.8737, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.029691978357732296 step: 64 running loss: 0.0004639371618395671 Train Steps: 64/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748], [0.6075, 0.4007, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280], [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633], [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133], [0.6339, 0.4102, 0.8588, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089], [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426], [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6060, 0.4027, 0.9025, 0.4582, 0.3996, 0.2778, 0.7325, 0.5489], [0.6204, 0.4067, 0.8110, 0.4773, 0.3780, 0.4878, 0.5226, 0.5209], [0.5794, 0.3736, 0.8835, 0.4338, 0.3948, 0.5220, 0.6071, 0.5618], [0.6142, 0.4049, 0.8165, 0.2311, 0.4544, 0.1651, 0.6256, 0.5015], [0.6073, 0.3961, 0.8408, 0.3007, 0.4270, 0.2088, 0.6565, 0.4918], [0.5828, 0.3903, 0.8484, 0.5090, 0.3845, 0.5092, 0.5891, 0.4947], [0.6020, 0.3937, 0.8769, 0.4152, 0.3334, 0.3141, 0.5947, 0.5279], [0.5650, 0.3774, 0.8745, 0.4475, 0.3991, 0.5974, 0.5709, 0.5031]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6329, 0.4196, 0.9237, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748], [0.6075, 0.4006, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280], [0.6189, 0.4049, 0.8888, 0.4417, 0.4212, 0.5200, 0.5987, 0.5633], [0.6266, 0.4065, 0.8313, 0.2467, 0.4787, 0.1733, 0.6313, 0.5133], [0.6339, 0.4102, 0.8587, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089], [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134], [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426], [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.030056098359636962 step: 65 running loss: 0.000462401513225184 Train Steps: 65/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6201, 0.4017, 0.8871, 0.4621, 0.3517, 0.4675, 0.5999, 0.5106], [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400], [0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683], [0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567], [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320], [0.6198, 0.4115, 0.7762, 0.2717, 0.3713, 0.3200, 0.5837, 0.5683], [0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954], [0.6228, 0.4119, 0.7938, 0.2233, 0.4674, 0.1773, 0.6188, 0.5433]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6325, 0.4188, 0.8790, 0.4454, 0.3357, 0.4470, 0.5937, 0.5070], [0.6276, 0.4159, 0.8831, 0.3919, 0.3389, 0.4420, 0.6121, 0.5288], [0.5823, 0.4039, 0.8797, 0.3453, 0.3316, 0.3823, 0.6009, 0.5428], [0.6027, 0.3982, 0.8725, 0.4607, 0.4668, 0.4718, 0.5631, 0.5381], [0.5903, 0.3948, 0.8057, 0.2344, 0.4171, 0.2533, 0.6727, 0.5183], [0.6415, 0.4335, 0.7790, 0.2666, 0.3548, 0.2899, 0.5766, 0.5454], [0.6410, 0.4316, 0.8549, 0.4109, 0.3794, 0.2458, 0.5019, 0.4821], [0.6050, 0.4117, 0.7940, 0.2302, 0.4581, 0.1521, 0.6242, 0.5132]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6201, 0.4017, 0.8871, 0.4621, 0.3517, 0.4675, 0.5999, 0.5106], [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6137, 0.5400], [0.6219, 0.4097, 0.8737, 0.3400, 0.3562, 0.4117, 0.5975, 0.5683], [0.6112, 0.4029, 0.8637, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567], [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320], [0.6198, 0.4115, 0.7763, 0.2717, 0.3713, 0.3200, 0.5838, 0.5683], [0.6137, 0.4038, 0.8562, 0.4050, 0.3812, 0.2550, 0.5106, 0.4954], [0.6228, 0.4119, 0.7937, 0.2233, 0.4674, 0.1773, 0.6187, 0.5433]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.030345589388161898 step: 66 running loss: 0.0004597816573963924 Train Steps: 66/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.6173, 0.4114, 0.7325, 0.2500, 0.4213, 0.1917, 0.5338, 0.5700], [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461], [0.6182, 0.3930, 0.8841, 0.3892, 0.3556, 0.4967, 0.6222, 0.5279], [ nan, nan, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737], [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5787, 0.5600], [ nan, nan, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6730, 0.4373, 0.9066, 0.5274, 0.3590, 0.4573, 0.6407, 0.4906], [0.6628, 0.4471, 0.7531, 0.2596, 0.4204, 0.1885, 0.5333, 0.5555], [0.6626, 0.4496, 0.8338, 0.3639, 0.3408, 0.4037, 0.5647, 0.5396], [0.6586, 0.4142, 0.8963, 0.4003, 0.3388, 0.4788, 0.6314, 0.5172], [0.0469, 0.0414, 0.8846, 0.2554, 0.5381, 0.2105, 0.7217, 0.5614], [0.6127, 0.3974, 0.8840, 0.2309, 0.5523, 0.2475, 0.7283, 0.5359], [0.6800, 0.4512, 0.8846, 0.4860, 0.4886, 0.5012, 0.5734, 0.5537], [0.0506, 0.0346, 0.7934, 0.2526, 0.3742, 0.2824, 0.5256, 0.5156]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.6173, 0.4114, 0.7325, 0.2500, 0.4212, 0.1917, 0.5337, 0.5700], [0.6058, 0.3978, 0.8288, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461], [0.6182, 0.3930, 0.8841, 0.3892, 0.3556, 0.4967, 0.6222, 0.5279], [0.0000, 0.0000, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737], [0.6300, 0.4133, 0.8537, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413], [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5788, 0.5600], [0.0000, 0.0000, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.03091353503987193 step: 67 running loss: 0.0004613960453712229 Train Steps: 67/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467], [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804], [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6063, 0.5617], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6072, 0.4156, 0.6845, 0.2488, 0.3885, 0.2646, 0.5822, 0.5506], [0.6824, 0.4430, 0.9033, 0.3734, 0.3597, 0.3630, 0.6155, 0.4861], [0.6485, 0.4382, 0.9266, 0.4681, 0.3643, 0.4627, 0.5905, 0.5646], [0.6429, 0.4136, 0.9213, 0.3295, 0.4761, 0.3300, 0.7252, 0.5254], [0.6812, 0.4458, 0.8530, 0.3079, 0.3447, 0.4176, 0.6092, 0.5309], [0.6150, 0.4112, 0.8791, 0.5605, 0.3931, 0.4561, 0.5007, 0.4900], [0.6644, 0.4421, 0.9267, 0.4800, 0.3993, 0.5388, 0.7369, 0.5406], [0.6674, 0.4306, 0.8511, 0.5840, 0.4571, 0.4663, 0.5463, 0.5880]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6162, 0.4134, 0.6700, 0.2467, 0.3963, 0.2533, 0.5738, 0.5467], [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804], [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6062, 0.5617], [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355], [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290], [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869], [0.6339, 0.4102, 0.9087, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.03130897891242057 step: 68 running loss: 0.0004604261604767731 Train Steps: 68/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583], [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933], [0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5737, 0.5700], [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600], [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.6183, 0.4076, 0.8838, 0.4517, 0.3813, 0.4483, 0.5775, 0.5633]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6401, 0.4268, 0.7226, 0.2012, 0.4088, 0.2647, 0.6165, 0.5383], [0.6213, 0.4168, 0.9257, 0.4448, 0.4271, 0.5342, 0.6339, 0.5451], [0.6416, 0.4361, 0.8674, 0.5556, 0.4946, 0.4437, 0.5999, 0.5797], [0.6356, 0.4275, 0.9069, 0.5226, 0.3884, 0.4892, 0.5855, 0.5514], [0.6988, 0.4739, 0.9074, 0.5321, 0.4143, 0.4907, 0.5857, 0.5554], [0.6401, 0.4200, 0.8829, 0.5230, 0.4142, 0.4624, 0.6204, 0.5285], [0.5978, 0.3988, 0.7610, 0.2069, 0.4505, 0.2119, 0.6215, 0.5091], [0.6463, 0.4257, 0.9085, 0.4362, 0.4016, 0.4561, 0.5870, 0.5596]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6087, 0.5583], [0.6241, 0.4142, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550], [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933], [0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5738, 0.5700], [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600], [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377], [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333], [0.6183, 0.4076, 0.8838, 0.4517, 0.3812, 0.4483, 0.5775, 0.5633]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.031762625847477466 step: 69 running loss: 0.00046032791083300674 Train Steps: 69/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317], [0.6151, 0.4058, 0.7068, 0.2680, 0.3400, 0.4083, 0.5775, 0.5733], [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166], [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268], [0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6460, 0.4203, 0.9000, 0.5420, 0.3821, 0.3687, 0.5508, 0.5364], [0.6151, 0.4056, 0.8694, 0.2938, 0.4477, 0.2250, 0.5831, 0.5344], [0.5969, 0.3859, 0.7443, 0.1821, 0.4409, 0.2535, 0.6052, 0.5541], [0.5619, 0.3696, 0.7780, 0.2227, 0.4654, 0.2409, 0.6000, 0.5386], [0.6365, 0.4150, 0.7459, 0.2567, 0.3699, 0.4079, 0.5746, 0.5771], [0.6522, 0.4290, 0.8946, 0.5280, 0.4635, 0.5333, 0.6704, 0.5245], [0.6457, 0.4102, 0.9029, 0.4656, 0.4647, 0.5256, 0.6216, 0.5263], [0.6640, 0.4348, 0.7400, 0.3104, 0.3679, 0.3367, 0.5077, 0.5379]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395], [0.6199, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317], [0.6151, 0.4058, 0.7068, 0.2680, 0.3400, 0.4083, 0.5775, 0.5733], [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166], [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268], [0.6109, 0.4041, 0.6975, 0.3167, 0.3512, 0.3383, 0.5153, 0.5319]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.03225781302899122 step: 70 running loss: 0.0004608259004141603 Train Steps: 70/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235], [0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207], [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5737, 0.5350], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167], [0.6200, 0.4049, 0.8638, 0.5617, 0.4125, 0.5100, 0.6013, 0.5317], [0.6201, 0.4102, 0.7288, 0.2417, 0.4150, 0.2383, 0.6100, 0.5500]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6148, 0.3716, 0.8605, 0.5585, 0.3871, 0.4737, 0.6259, 0.5169], [0.6468, 0.4023, 0.8991, 0.4805, 0.4321, 0.5419, 0.6411, 0.5354], [0.6433, 0.4191, 0.7664, 0.3155, 0.3697, 0.4264, 0.5722, 0.5404], [0.6585, 0.4212, 0.8732, 0.4792, 0.3988, 0.3011, 0.5767, 0.5431], [0.6219, 0.3866, 0.9016, 0.4525, 0.4625, 0.5749, 0.6231, 0.5492], [0.6424, 0.4062, 0.7986, 0.3269, 0.3891, 0.3860, 0.6075, 0.6323], [0.6266, 0.4012, 0.8681, 0.5553, 0.4246, 0.4999, 0.5951, 0.5453], [0.6402, 0.4050, 0.7469, 0.2468, 0.4276, 0.2472, 0.5946, 0.5733]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6310, 0.4017, 0.8562, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006], [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235], [0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207], [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5738, 0.5350], [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383], [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6037, 0.6167], [0.6199, 0.4049, 0.8637, 0.5617, 0.4125, 0.5100, 0.6012, 0.5317], [0.6201, 0.4102, 0.7287, 0.2417, 0.4150, 0.2383, 0.6100, 0.5500]]], device='cuda:0') loss_train_step before backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train: 0.032472519553266466 step: 71 running loss: 0.0004573594303276967 Train Steps: 71/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6277, 0.4083, 0.8350, 0.2717, 0.4562, 0.1800, 0.5918, 0.4878], [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6090, 0.4010, 0.7838, 0.3483, 0.3538, 0.3783, 0.5462, 0.5077]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5940, 0.3746, 0.8675, 0.4779, 0.3429, 0.3852, 0.6126, 0.4978], [0.6416, 0.4011, 0.7967, 0.2772, 0.4555, 0.2147, 0.5898, 0.5085], [0.6110, 0.3931, 0.8839, 0.4816, 0.3927, 0.5575, 0.7458, 0.5450], [0.5519, 0.3622, 0.6993, 0.2598, 0.4502, 0.2145, 0.5542, 0.5974], [0.6182, 0.4113, 0.8660, 0.4374, 0.4177, 0.5403, 0.5742, 0.5472], [0.6018, 0.4055, 0.8068, 0.4443, 0.4412, 0.2809, 0.5477, 0.6292], [0.5763, 0.3665, 0.7078, 0.2500, 0.4051, 0.2317, 0.5698, 0.5809], [0.5822, 0.3735, 0.7703, 0.3540, 0.3481, 0.3917, 0.5402, 0.5193]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672], [0.6277, 0.4083, 0.8350, 0.2717, 0.4563, 0.1800, 0.5918, 0.4878], [0.6339, 0.4102, 0.9087, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390], [0.6218, 0.4185, 0.7337, 0.2650, 0.4625, 0.1950, 0.5688, 0.5800], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209], [0.6193, 0.4050, 0.7312, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6090, 0.4010, 0.7837, 0.3483, 0.3537, 0.3783, 0.5462, 0.5077]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.03298366616945714 step: 72 running loss: 0.0004581064745757936 Train Steps: 72/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533], [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433], [0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319], [0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6097, 0.4024, 0.8488, 0.3717, 0.3875, 0.5517, 0.5836, 0.5591], [0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6499, 0.4078, 0.7000, 0.2581, 0.3769, 0.3111, 0.6247, 0.5697], [0.6281, 0.4074, 0.8035, 0.3519, 0.3447, 0.3648, 0.5826, 0.5726], [0.6364, 0.4119, 0.8539, 0.5459, 0.3686, 0.3947, 0.6423, 0.4841], [0.6095, 0.3974, 0.8234, 0.3717, 0.3595, 0.3201, 0.5477, 0.5440], [0.6284, 0.3995, 0.6904, 0.3434, 0.3354, 0.3422, 0.5280, 0.5376], [0.5998, 0.3851, 0.8500, 0.4988, 0.3631, 0.4099, 0.5674, 0.5175], [0.6158, 0.3904, 0.8237, 0.4010, 0.3881, 0.5619, 0.6200, 0.5523], [0.6299, 0.4122, 0.8502, 0.3714, 0.3427, 0.4069, 0.6090, 0.5711]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533], [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5412, 0.5433], [0.6109, 0.4041, 0.6975, 0.3167, 0.3512, 0.3383, 0.5153, 0.5319], [0.6108, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117], [0.6097, 0.4024, 0.8487, 0.3717, 0.3875, 0.5517, 0.5836, 0.5591], [0.6219, 0.4097, 0.8737, 0.3400, 0.3562, 0.4117, 0.5975, 0.5683]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.033311496314127 step: 73 running loss: 0.0004563218673168082 Train Steps: 73/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960], [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510], [ nan, nan, 0.7648, 0.2722, 0.3962, 0.2183, 0.5060, 0.5422], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881], [0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196], [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494], [0.6166, 0.4008, 0.8563, 0.5667, 0.4388, 0.4933, 0.5575, 0.5567]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6001, 0.3739, 0.8487, 0.4550, 0.4385, 0.5025, 0.5561, 0.5098], [0.6396, 0.3958, 0.6744, 0.2837, 0.3477, 0.3114, 0.5517, 0.5619], [0.0268, 0.0122, 0.7260, 0.2481, 0.3726, 0.2221, 0.5261, 0.5384], [0.6357, 0.3967, 0.8064, 0.5874, 0.3649, 0.5221, 0.6275, 0.4840], [0.5734, 0.3575, 0.8358, 0.5049, 0.3809, 0.4387, 0.5359, 0.5817], [0.6598, 0.4303, 0.8430, 0.4853, 0.4221, 0.2899, 0.5880, 0.6172], [0.6129, 0.3875, 0.8133, 0.5460, 0.3682, 0.4424, 0.7284, 0.5532], [0.6262, 0.3910, 0.8388, 0.5480, 0.4184, 0.5080, 0.5845, 0.5708]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960], [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510], [0.0000, 0.0000, 0.7648, 0.2722, 0.3963, 0.2183, 0.5060, 0.5422], [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649], [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881], [0.6278, 0.4253, 0.8875, 0.5017, 0.4112, 0.2750, 0.5413, 0.6196], [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494], [0.6166, 0.4008, 0.8562, 0.5667, 0.4387, 0.4933, 0.5575, 0.5567]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.03379115313873626 step: 74 running loss: 0.00045663720457751706 Train Steps: 74/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6201, 0.4151, 0.8588, 0.5467, 0.3700, 0.3950, 0.5637, 0.5933], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123], [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167], [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6307, 0.4029, 0.8650, 0.5200, 0.3763, 0.4017, 0.7311, 0.5366], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6089, 0.3922, 0.7377, 0.3172, 0.3326, 0.3313, 0.5328, 0.5695], [0.6255, 0.4055, 0.8288, 0.5502, 0.3399, 0.3870, 0.5390, 0.5697], [0.6152, 0.4073, 0.6935, 0.2694, 0.3570, 0.2763, 0.5415, 0.5174], [0.5788, 0.3750, 0.7982, 0.3405, 0.3481, 0.3064, 0.5433, 0.5249], [0.6074, 0.3984, 0.8524, 0.4886, 0.3761, 0.5244, 0.6152, 0.4800], [0.6091, 0.3986, 0.8160, 0.4387, 0.3483, 0.4726, 0.5366, 0.5699], [0.6254, 0.4090, 0.8349, 0.5263, 0.3610, 0.4049, 0.7104, 0.5167], [0.6412, 0.4223, 0.8794, 0.3954, 0.4289, 0.2125, 0.5985, 0.4951]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749], [0.6202, 0.4151, 0.8587, 0.5467, 0.3700, 0.3950, 0.5638, 0.5933], [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123], [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167], [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6307, 0.4029, 0.8650, 0.5200, 0.3762, 0.4017, 0.7311, 0.5366], [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.03405617133830674 step: 75 running loss: 0.00045408228451075655 Train Steps: 75/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689], [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631], [0.6293, 0.3982, 0.8700, 0.5300, 0.3763, 0.4717, 0.7050, 0.5297], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748], [0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071], [0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050], [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6188, 0.5433]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6041, 0.3908, 0.7807, 0.3757, 0.3441, 0.3718, 0.5173, 0.5518], [0.5819, 0.3833, 0.8373, 0.4099, 0.3521, 0.4983, 0.5535, 0.5542], [0.6308, 0.3993, 0.8683, 0.5513, 0.3575, 0.4532, 0.6896, 0.5235], [0.6291, 0.4198, 0.8922, 0.5122, 0.3483, 0.4118, 0.6168, 0.5032], [0.6506, 0.4303, 0.8431, 0.5729, 0.3834, 0.3998, 0.7119, 0.5599], [0.6258, 0.4162, 0.7446, 0.2005, 0.3655, 0.2652, 0.5402, 0.5004], [0.6046, 0.4005, 0.8637, 0.5245, 0.4355, 0.5057, 0.5201, 0.4984], [0.6070, 0.4025, 0.8941, 0.4713, 0.3727, 0.5104, 0.5979, 0.5312]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6102, 0.4001, 0.7738, 0.3583, 0.3462, 0.3800, 0.5524, 0.5689], [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631], [0.6293, 0.3982, 0.8700, 0.5300, 0.3762, 0.4717, 0.7050, 0.5297], [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150], [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748], [0.6163, 0.4114, 0.7650, 0.2017, 0.3762, 0.2867, 0.5631, 0.5071], [0.6097, 0.3988, 0.8650, 0.5250, 0.4212, 0.5200, 0.5675, 0.5050], [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6187, 0.5433]]], device='cuda:0') loss_train_step before backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train: 0.03425430311472155 step: 76 running loss: 0.00045071451466738885 Train Steps: 76/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6122, 0.3993, 0.8738, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927], [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [ nan, nan, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220], [0.6102, 0.4020, 0.8638, 0.3717, 0.3625, 0.5017, 0.6038, 0.5500], [0.6339, 0.4102, 0.8588, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.5782, 0.3960, 0.8596, 0.4026, 0.3638, 0.2947, 0.5104, 0.5605], [ 0.5868, 0.3921, 0.8928, 0.5126, 0.3761, 0.3818, 0.6223, 0.5027], [ 0.5819, 0.3889, 0.8771, 0.4751, 0.4446, 0.4962, 0.5090, 0.5008], [ 0.6055, 0.4110, 0.8534, 0.2530, 0.4695, 0.1872, 0.6282, 0.5183], [ 0.5992, 0.3971, 0.8742, 0.5657, 0.3681, 0.3804, 0.6038, 0.4837], [-0.0291, -0.0051, 0.7582, 0.2469, 0.3822, 0.2729, 0.5228, 0.5483], [ 0.5652, 0.3814, 0.8680, 0.3559, 0.3709, 0.4956, 0.6065, 0.5715], [ 0.6148, 0.4079, 0.8644, 0.3119, 0.4575, 0.2078, 0.6471, 0.5255]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6133, 0.4094, 0.8495, 0.4028, 0.3587, 0.3200, 0.5003, 0.5407], [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938], [0.6122, 0.3993, 0.8737, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927], [0.6275, 0.4111, 0.8462, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904], [0.6272, 0.4071, 0.8737, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.0000, 0.0000, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220], [0.6102, 0.4020, 0.8637, 0.3717, 0.3625, 0.5017, 0.6037, 0.5500], [0.6339, 0.4102, 0.8587, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089]]], device='cuda:0') loss_train_step before backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train: 0.03449908675975166 step: 77 running loss: 0.00044804008778898265 Train Steps: 77/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899], [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312], [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233], [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6063, 0.5567], [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967], [0.6346, 0.4092, 0.7712, 0.5917, 0.4037, 0.4767, 0.7343, 0.5725], [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5928, 0.4148, 0.9317, 0.4549, 0.3857, 0.4069, 0.6875, 0.5749], [0.6258, 0.4227, 0.8987, 0.4046, 0.3453, 0.4238, 0.5952, 0.5166], [0.5839, 0.4023, 0.9205, 0.4549, 0.3527, 0.4937, 0.6135, 0.5055], [0.6059, 0.4120, 0.7293, 0.2632, 0.4128, 0.1986, 0.5330, 0.5595], [0.6360, 0.4297, 0.9069, 0.4480, 0.4612, 0.5162, 0.5732, 0.5343], [0.5389, 0.3747, 0.9047, 0.5002, 0.3684, 0.3171, 0.6194, 0.4796], [0.5922, 0.4071, 0.8178, 0.5521, 0.3789, 0.4638, 0.7159, 0.5648], [0.5700, 0.3894, 0.9051, 0.4533, 0.4485, 0.4864, 0.5142, 0.4808]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6346, 0.4144, 0.9087, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899], [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312], [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233], [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783], [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6062, 0.5567], [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967], [0.6346, 0.4092, 0.7713, 0.5917, 0.4038, 0.4767, 0.7343, 0.5725], [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960]]], device='cuda:0') loss_train_step before backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0006, device='cuda:0', grad_fn=) loss_train: 0.03509918806958012 step: 78 running loss: 0.0004499895906356426 Train Steps: 78/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6202, 0.4064, 0.7879, 0.2179, 0.4567, 0.1725, 0.5955, 0.5478], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833], [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717], [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5432, 0.3973, 0.8100, 0.2312, 0.4705, 0.1608, 0.5825, 0.5160], [0.5358, 0.3683, 0.9406, 0.3140, 0.4232, 0.4218, 0.7012, 0.5485], [0.5776, 0.4162, 0.9353, 0.4523, 0.4308, 0.5172, 0.5749, 0.5158], [0.6260, 0.4445, 0.7304, 0.2638, 0.4196, 0.2368, 0.5704, 0.5261], [0.5931, 0.4030, 0.9394, 0.4244, 0.3533, 0.3649, 0.5951, 0.5182], [0.6915, 0.4722, 0.7127, 0.3191, 0.3733, 0.2816, 0.5586, 0.5610], [0.5933, 0.4062, 0.7941, 0.2518, 0.3930, 0.2678, 0.5787, 0.5510], [0.5903, 0.4067, 0.7309, 0.2871, 0.3600, 0.3128, 0.5212, 0.5276]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6202, 0.4064, 0.7879, 0.2179, 0.4567, 0.1725, 0.5955, 0.5478], [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575], [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533], [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517], [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482], [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833], [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717], [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510]]], device='cuda:0') loss_train_step before backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0009, device='cuda:0', grad_fn=) loss_train: 0.03596113619278185 step: 79 running loss: 0.00045520425560483354 Train Steps: 79/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633], [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320], [0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609], [0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5613, 0.5900], [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496], [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672], [0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050], [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[-0.0010, 0.0065, 0.7765, 0.2641, 0.4116, 0.2566, 0.5234, 0.5575], [ 0.6268, 0.4027, 0.8630, 0.4974, 0.4012, 0.5111, 0.7375, 0.5369], [ 0.6158, 0.4085, 0.8687, 0.4403, 0.4261, 0.5473, 0.5952, 0.5478], [ 0.6377, 0.4170, 0.8245, 0.4601, 0.3776, 0.3475, 0.5361, 0.5637], [ 0.6432, 0.4277, 0.8856, 0.2673, 0.4200, 0.2645, 0.6101, 0.5431], [ 0.6187, 0.3922, 0.8841, 0.5153, 0.3864, 0.4361, 0.6395, 0.4873], [ 0.6301, 0.4144, 0.8737, 0.4948, 0.4597, 0.5032, 0.5711, 0.5104], [ 0.6202, 0.4080, 0.8791, 0.5169, 0.3921, 0.3731, 0.5952, 0.5370]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633], [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320], [0.6064, 0.4019, 0.8650, 0.4517, 0.4038, 0.5367, 0.5703, 0.5609], [0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5612, 0.5900], [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496], [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672], [0.6097, 0.3988, 0.8650, 0.5250, 0.4212, 0.5200, 0.5675, 0.5050], [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.036256922205211595 step: 80 running loss: 0.0004532115275651449 Train Steps: 80/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6179, 0.3993, 0.8925, 0.4789, 0.3879, 0.4900, 0.6041, 0.5279], [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852], [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667], [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297], [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6284, 0.4093, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6646, 0.4309, 0.8815, 0.4630, 0.3912, 0.5001, 0.6282, 0.5300], [0.6050, 0.4034, 0.8733, 0.5086, 0.4680, 0.4959, 0.5481, 0.5852], [0.6617, 0.4315, 0.8732, 0.3614, 0.3578, 0.4102, 0.6455, 0.5599], [0.6401, 0.4181, 0.8733, 0.5595, 0.3670, 0.4489, 0.6568, 0.5374], [0.6406, 0.4236, 0.8144, 0.3135, 0.4189, 0.2087, 0.5796, 0.5340], [0.6158, 0.4083, 0.8343, 0.4385, 0.3714, 0.4626, 0.5720, 0.5615], [0.6516, 0.4322, 0.7191, 0.2065, 0.4086, 0.2620, 0.6223, 0.5469], [0.6441, 0.4177, 0.8827, 0.4689, 0.3794, 0.3867, 0.6357, 0.5137]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6179, 0.3993, 0.8925, 0.4789, 0.3879, 0.4900, 0.6041, 0.5279], [0.6140, 0.4070, 0.8700, 0.5000, 0.4613, 0.4900, 0.5260, 0.5852], [0.6257, 0.4167, 0.8775, 0.3433, 0.3562, 0.4133, 0.6200, 0.5667], [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297], [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411], [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656], [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483], [0.6284, 0.4092, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167]]], device='cuda:0') loss_train_step before backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train: 0.03644900632207282 step: 81 running loss: 0.00044998773237126937 Train Steps: 81/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117], [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5988, 0.5700], [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578], [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917], [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323], [0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6277, 0.3878, 0.9010, 0.5158, 0.3749, 0.4779, 0.6465, 0.5020], [0.6288, 0.4082, 0.7888, 0.2897, 0.3721, 0.3737, 0.5910, 0.5788], [0.5866, 0.3920, 0.6858, 0.2162, 0.4286, 0.2067, 0.5621, 0.5650], [0.6369, 0.4128, 0.7630, 0.4370, 0.3602, 0.3508, 0.5537, 0.5989], [0.6528, 0.4111, 0.7752, 0.2209, 0.4564, 0.2176, 0.6471, 0.5461], [0.6481, 0.4258, 0.7230, 0.2635, 0.4205, 0.2282, 0.5779, 0.5720], [0.6336, 0.3955, 0.8779, 0.5571, 0.3745, 0.4803, 0.6392, 0.4987], [0.6238, 0.4054, 0.8646, 0.5141, 0.3662, 0.4475, 0.5779, 0.6036]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117], [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5987, 0.5700], [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578], [0.6179, 0.4118, 0.7278, 0.4237, 0.3587, 0.3400, 0.5675, 0.5917], [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323], [0.6193, 0.4050, 0.7312, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656], [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973], [0.6186, 0.4060, 0.8750, 0.5050, 0.3537, 0.4367, 0.5813, 0.6083]]], device='cuda:0') loss_train_step before backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0002, device='cuda:0', grad_fn=) loss_train: 0.036643293395172805 step: 82 running loss: 0.00044686943164844883 Train Steps: 82/90 Loss: 0.0004 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862], [ nan, nan, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583], [ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550], [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773], [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598], [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717], [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6645, 0.4274, 0.8826, 0.4182, 0.3632, 0.4493, 0.6396, 0.5114], [0.1255, 0.0742, 0.7000, 0.2189, 0.4249, 0.2004, 0.5262, 0.5696], [0.1267, 0.0676, 0.7274, 0.2114, 0.3882, 0.2491, 0.5524, 0.5611], [0.6629, 0.4283, 0.6798, 0.2681, 0.3986, 0.2032, 0.5701, 0.5781], [0.6747, 0.4213, 0.8377, 0.5059, 0.4841, 0.4928, 0.5337, 0.5692], [0.6728, 0.4323, 0.8164, 0.5713, 0.4575, 0.4565, 0.5950, 0.5876], [0.6986, 0.4415, 0.7335, 0.2450, 0.3789, 0.2552, 0.6003, 0.5945], [0.7199, 0.4423, 0.8084, 0.5859, 0.3913, 0.4404, 0.6138, 0.5356]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862], [0.0000, 0.0000, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583], [0.0000, 0.0000, 0.7425, 0.2117, 0.3938, 0.2433, 0.5437, 0.5567], [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5362, 0.5550], [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773], [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598], [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717], [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267]]], device='cuda:0') loss_train_step before backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0014, device='cuda:0', grad_fn=) loss_train: 0.03802865924080834 step: 83 running loss: 0.0004581766173591366 Train Steps: 83/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[ nan, nan, 0.8363, 0.3317, 0.3563, 0.3367, 0.5329, 0.5142], [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533], [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528], [0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350], [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050], [0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.0603, 0.0229, 0.7907, 0.3199, 0.3460, 0.3630, 0.5440, 0.5371], [0.6344, 0.3954, 0.6876, 0.2187, 0.3711, 0.3368, 0.6029, 0.5666], [0.6674, 0.4167, 0.8742, 0.3484, 0.4033, 0.3879, 0.7346, 0.5647], [0.6683, 0.3995, 0.7279, 0.2874, 0.4217, 0.2310, 0.5822, 0.5594], [0.6703, 0.4092, 0.8422, 0.4734, 0.4231, 0.4850, 0.5436, 0.5723], [0.6740, 0.4162, 0.8302, 0.2882, 0.4772, 0.2208, 0.6648, 0.5577], [0.6306, 0.3858, 0.8653, 0.4550, 0.3553, 0.4258, 0.5858, 0.5191], [0.6742, 0.4291, 0.8340, 0.5555, 0.3548, 0.3969, 0.5518, 0.6012]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.0000, 0.0000, 0.8363, 0.3317, 0.3562, 0.3367, 0.5329, 0.5142], [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533], [0.6332, 0.4165, 0.9100, 0.3350, 0.4187, 0.3683, 0.7438, 0.5528], [0.6229, 0.4066, 0.7613, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350], [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611], [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459], [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050], [0.6260, 0.4214, 0.8537, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.03851493066758849 step: 84 running loss: 0.0004585110793760534 Train Steps: 84/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650], [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817], [0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389], [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896], [0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110], [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083], [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6436, 0.4039, 0.8519, 0.4282, 0.3554, 0.4118, 0.5926, 0.5785], [0.6478, 0.4293, 0.7909, 0.3250, 0.4391, 0.2788, 0.5641, 0.5899], [0.6030, 0.3804, 0.8897, 0.3661, 0.4324, 0.3936, 0.7197, 0.5556], [0.5850, 0.3690, 0.7338, 0.2239, 0.3816, 0.2882, 0.5983, 0.5582], [0.5742, 0.3576, 0.7863, 0.3901, 0.3561, 0.3192, 0.5261, 0.6010], [0.5740, 0.3685, 0.8627, 0.4725, 0.3765, 0.5421, 0.5540, 0.5164], [0.6090, 0.3848, 0.8427, 0.4962, 0.3518, 0.4459, 0.5653, 0.6092], [0.6362, 0.3829, 0.8636, 0.4940, 0.3785, 0.4176, 0.6509, 0.5184]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6203, 0.4096, 0.8863, 0.4267, 0.3537, 0.4117, 0.6025, 0.5650], [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817], [0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494], [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389], [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896], [0.6186, 0.4129, 0.8900, 0.4567, 0.3938, 0.5417, 0.5734, 0.5110], [0.6186, 0.4060, 0.8750, 0.5050, 0.3537, 0.4367, 0.5813, 0.6083], [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332]]], device='cuda:0') loss_train_step before backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0004, device='cuda:0', grad_fn=) loss_train: 0.03894413818488829 step: 85 running loss: 0.00045816633158692104 Train Steps: 85/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767], [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402], [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050], [0.6274, 0.4003, 0.8638, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6141, 0.3935, 0.7946, 0.2674, 0.3842, 0.2769, 0.5639, 0.5398], [0.6237, 0.3929, 0.8780, 0.4692, 0.4058, 0.5315, 0.6351, 0.5098], [0.6030, 0.3903, 0.8513, 0.2258, 0.5386, 0.2168, 0.7136, 0.5800], [0.6055, 0.3781, 0.8742, 0.4740, 0.3947, 0.4553, 0.5317, 0.6069], [0.5899, 0.3682, 0.8513, 0.5185, 0.4040, 0.4640, 0.5885, 0.5614], [0.6357, 0.3988, 0.8772, 0.5516, 0.3704, 0.3951, 0.5980, 0.4916], [0.5939, 0.3816, 0.8711, 0.5063, 0.4372, 0.5173, 0.5934, 0.5332], [0.6167, 0.3730, 0.8572, 0.5795, 0.3756, 0.4856, 0.6109, 0.4972]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6204, 0.4049, 0.7975, 0.2700, 0.3938, 0.2567, 0.5700, 0.5183], [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758], [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609], [0.6163, 0.4001, 0.8788, 0.5033, 0.4013, 0.4633, 0.5337, 0.5767], [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402], [0.6272, 0.4071, 0.8737, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742], [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050], [0.6274, 0.4003, 0.8637, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.03921863157302141 step: 86 running loss: 0.0004560305996862955 Train Steps: 86/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459], [ nan, nan, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [ nan, nan, 0.7725, 0.2611, 0.3675, 0.2733, 0.5413, 0.5167], [0.6115, 0.4081, 0.6725, 0.2433, 0.4088, 0.1933, 0.5167, 0.5544], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[ 0.6535, 0.4477, 0.7380, 0.3077, 0.4887, 0.1751, 0.5390, 0.6052], [ 0.6257, 0.4103, 0.8346, 0.5530, 0.4003, 0.4779, 0.6911, 0.5352], [-0.0232, -0.0088, 0.8852, 0.2373, 0.5437, 0.2008, 0.7311, 0.5667], [ 0.6336, 0.4117, 0.8687, 0.4738, 0.4279, 0.4695, 0.5236, 0.5331], [ 0.0291, 0.0120, 0.7872, 0.2631, 0.3804, 0.2830, 0.5299, 0.5368], [ 0.5745, 0.3887, 0.7025, 0.2337, 0.4180, 0.1965, 0.5124, 0.5520], [ 0.6495, 0.4207, 0.8767, 0.5533, 0.3920, 0.4965, 0.6463, 0.5284], [ 0.6101, 0.3976, 0.8975, 0.4903, 0.3915, 0.5094, 0.5763, 0.4975]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6270, 0.4266, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183], [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459], [0.0000, 0.0000, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737], [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383], [0.0000, 0.0000, 0.7725, 0.2611, 0.3675, 0.2733, 0.5412, 0.5167], [0.6115, 0.4081, 0.6725, 0.2433, 0.4087, 0.1933, 0.5167, 0.5544], [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285], [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.039484524459112436 step: 87 running loss: 0.00045384510872543027 Train Steps: 87/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6267, 0.4080, 0.8438, 0.2633, 0.4763, 0.1800, 0.6259, 0.5240], [0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400], [0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285], [0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378], [0.6223, 0.4028, 0.8988, 0.4200, 0.3763, 0.5733, 0.6375, 0.5167], [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850], [0.6224, 0.4097, 0.7438, 0.2267, 0.3850, 0.2850, 0.5988, 0.5250], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5603, 0.3784, 0.8676, 0.2693, 0.4874, 0.1654, 0.6212, 0.5118], [0.5558, 0.3819, 0.8536, 0.2485, 0.4586, 0.1737, 0.6244, 0.5170], [0.5744, 0.3882, 0.9210, 0.4832, 0.4121, 0.5297, 0.6812, 0.5251], [0.6119, 0.4007, 0.8821, 0.6022, 0.3731, 0.4060, 0.5844, 0.5124], [0.6405, 0.4299, 0.9178, 0.4135, 0.3962, 0.5504, 0.6143, 0.5061], [0.5793, 0.4005, 0.9085, 0.4624, 0.4908, 0.5308, 0.6161, 0.5828], [0.5847, 0.4053, 0.7593, 0.2255, 0.4041, 0.2817, 0.5970, 0.5267], [0.5840, 0.4049, 0.8577, 0.5849, 0.3925, 0.4816, 0.6681, 0.5257]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6267, 0.4080, 0.8438, 0.2633, 0.4762, 0.1800, 0.6259, 0.5240], [0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400], [0.6254, 0.3993, 0.8988, 0.4767, 0.3988, 0.5517, 0.6955, 0.5285], [0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378], [0.6223, 0.4028, 0.8988, 0.4200, 0.3762, 0.5733, 0.6375, 0.5167], [0.6184, 0.4079, 0.8838, 0.4617, 0.4837, 0.5650, 0.6175, 0.5850], [0.6224, 0.4097, 0.7437, 0.2267, 0.3850, 0.2850, 0.5987, 0.5250], [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297]]], device='cuda:0') loss_train_step before backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0005, device='cuda:0', grad_fn=) loss_train: 0.03999189910246059 step: 88 running loss: 0.00045445339889159766 Train Steps: 88/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767], [0.6117, 0.4019, 0.8538, 0.4067, 0.3513, 0.3583, 0.5663, 0.5133], [0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6289, 0.4032, 0.8419, 0.5446, 0.4075, 0.5017, 0.6312, 0.5117], [0.6131, 0.4037, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680], [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5787, 0.5550], [0.6216, 0.4099, 0.7225, 0.2033, 0.4188, 0.2217, 0.5975, 0.5283], [0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.6066, 0.3985, 0.9077, 0.5174, 0.4141, 0.4509, 0.5534, 0.5614], [0.5354, 0.3544, 0.8979, 0.4296, 0.3646, 0.3531, 0.5639, 0.4929], [0.5940, 0.4112, 0.8060, 0.2632, 0.5045, 0.1279, 0.6124, 0.5162], [0.6000, 0.3899, 0.8762, 0.5717, 0.4203, 0.5012, 0.6693, 0.4945], [0.5872, 0.3957, 0.7123, 0.2898, 0.3786, 0.2703, 0.5455, 0.5416], [0.5807, 0.3968, 0.8483, 0.2766, 0.3685, 0.3557, 0.5858, 0.5332], [0.5939, 0.4277, 0.7381, 0.2260, 0.4264, 0.2097, 0.5972, 0.5062], [0.5100, 0.3544, 0.7657, 0.2191, 0.4093, 0.1690, 0.5424, 0.4659]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6163, 0.4001, 0.8788, 0.5033, 0.4013, 0.4633, 0.5337, 0.5767], [0.6116, 0.4019, 0.8537, 0.4067, 0.3512, 0.3583, 0.5663, 0.5133], [0.6245, 0.4100, 0.7763, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417], [0.6289, 0.4031, 0.8419, 0.5446, 0.4075, 0.5017, 0.6313, 0.5117], [0.6131, 0.4036, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680], [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5788, 0.5550], [0.6216, 0.4099, 0.7225, 0.2033, 0.4187, 0.2217, 0.5975, 0.5283], [0.6055, 0.4015, 0.7425, 0.2033, 0.4112, 0.1883, 0.5217, 0.4823]]], device='cuda:0') loss_train_step before backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0007, device='cuda:0', grad_fn=) loss_train: 0.040671683149412274 step: 89 running loss: 0.0004569852039259806 Train Steps: 89/90 Loss: 0.0005 torch.Size([8, 600, 800]) torch.Size([8, 8]) tensor([[0.6110, 0.4047, 0.8700, 0.4483, 0.3713, 0.3967, 0.5088, 0.5517], [0.6138, 0.4101, 0.8800, 0.5083, 0.4637, 0.5950, 0.5587, 0.5077], [0.6272, 0.4045, 0.8538, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6224, 0.4179, 0.8700, 0.5683, 0.4037, 0.4683, 0.5650, 0.5600], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283], [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250], [0.6069, 0.3975, 0.8625, 0.5083, 0.4388, 0.5483, 0.5650, 0.4967], [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833]], device='cuda:0', dtype=torch.float64) predictions are: tensor([[0.5591, 0.3871, 0.8722, 0.4501, 0.3691, 0.3893, 0.5137, 0.5304], [0.6191, 0.4178, 0.8732, 0.5310, 0.4599, 0.5640, 0.5903, 0.5062], [0.6055, 0.4156, 0.8485, 0.5866, 0.3716, 0.4301, 0.6193, 0.4586], [0.5629, 0.3912, 0.8800, 0.5770, 0.4015, 0.4430, 0.5927, 0.5503], [0.5660, 0.3942, 0.8862, 0.4269, 0.3550, 0.3593, 0.5990, 0.5308], [0.5981, 0.4102, 0.9046, 0.4074, 0.3817, 0.3150, 0.6375, 0.5045], [0.6029, 0.4083, 0.8549, 0.5220, 0.4479, 0.5373, 0.5806, 0.5060], [0.6232, 0.4102, 0.8882, 0.4807, 0.3777, 0.4603, 0.6204, 0.4733]], device='cuda:0', grad_fn=) landmarks are: tensor([[[0.6110, 0.4047, 0.8700, 0.4483, 0.3713, 0.3967, 0.5088, 0.5517], [0.6138, 0.4101, 0.8800, 0.5083, 0.4638, 0.5950, 0.5587, 0.5077], [0.6271, 0.4045, 0.8537, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649], [0.6224, 0.4179, 0.8700, 0.5683, 0.4038, 0.4683, 0.5650, 0.5600], [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283], [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250], [0.6069, 0.3975, 0.8625, 0.5083, 0.4387, 0.5483, 0.5650, 0.4967], [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6037, 0.4833]]], device='cuda:0') loss_train_step before backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train_step after backward: tensor(0.0003, device='cuda:0', grad_fn=) loss_train: 0.04101139504928142 step: 90 running loss: 0.00045568216721423796 Valid Steps: 10/10 Loss: nan 05 -------------------------------------------------- Epoch: 10 Train Loss: 0.0005 Valid Loss: nan -------------------------------------------------- Training Complete Total Elapsed Time : 452.1295247077942 s