Some bugs in train and track

[andy690]

你好，我在训练阶段发现了一个评估部分的错误，在完成训练之后，会有这样一个报错，应该是评估阶段没有输出位置？
Epoch: [5] Total time: 0:57:48 (0.4067 s / it)
Averaged stats: lr: 0.000020 grad_norm: 79.04 loss: 21.6669 (23.7960) hm: 1.6910 (2.4767) tracking: 18.9840 (20.0076) reg: 0.3937 (0.4255) wh: 0.0684 (0.1318) center_offset: 0.0087 (0.0162) boxes: 0.3018 (0.4165) giou: 0.2821 (0.3197) gnn: 0.0008 (0.0020) hm_unscaled: 1.6910 (2.4767) tracking_unscaled: 18.9840 (20.0076) reg_unscaled: 0.3937 (0.4255) wh_unscaled: 0.6835 (1.3178) center_offset_unscaled: 0.0871 (0.1624) boxes_unscaled: 0.6036 (0.8330) giou_unscaled: 0.7053 (0.7992) gnn_unscaled: 0.0789 (0.2002)
I am using Myeval
/home/ubuntu/qian/HGTMT/training/engine_RGBT_graph_track_gnnloss.py:332: FutureWarning: torch.cuda.amp.autocast(args...) is deprecated. Please use torch.amp.autocast('cuda', args...) instead.
with torch.cuda.amp.autocast(enabled=half):
Traceback (most recent call last):
File "/home/ubuntu/qian/HGTMT/training/main_RGBT-Tiny_graph_track_gnnloss.py", line 533, in
main(args)
File "/home/ubuntu/qian/HGTMT/training/main_RGBT-Tiny_graph_track_gnnloss.py", line 397, in main
test_stats, coco_evaluator = evaluate(model, criterion, postprocessors,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ubuntu/anaconda3/envs/pytorch/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/home/ubuntu/qian/HGTMT/training/engine_RGBT_graph_track_gnnloss.py", line 338, in evaluate
outputs, [edge_r, edge_i] = model(samples_r, samples_i, pre_samples_r, pre_samples_i, targets['pre_cts_r'],
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ubuntu/anaconda3/envs/pytorch/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1736, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ubuntu/anaconda3/envs/pytorch/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1747, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ubuntu/qian/HGTMT/training/deformable_detr_graph_track_gnnloss.py", line 240, in forward
merged_hs_r, merged_hs_i, [edge_r, edge_i] = self.transformer(samples_r, pre_samples_r, samples_i, pre_samples_i,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ubuntu/anaconda3/envs/pytorch/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1736, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ubuntu/anaconda3/envs/pytorch/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1747, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ubuntu/qian/HGTMT/models/deformable_transformer_lite_dual_track_graph3_gnnloss.py", line 387, in forward
tracks_feat_r = F.grid_sample(center_feats_r, (2.0 * track_sample_r - 1.0).unsqueeze(1),
^^^^^^^^^^^^^^
UnboundLocalError: cannot access local variable 'track_sample_r' where it is not associated with a value

另外在推理中，也遇到一个问题，似乎是 (SSIM) 函数时，图像尺寸小于默认的窗口大小导致的，请问您有遇到过吗
11:39:36 - mmengine - INFO - DeformConv2dPack ida_up.node_0.conv is upgraded to version 2.
06/25 11:39:36 - mmengine - INFO - DeformConv2dPack ida_up.node_1.conv is upgraded to version 2.
06/25 11:39:36 - mmengine - INFO - DeformConv2dPack ida_up.node_2.conv is upgraded to version 2.
06/25 11:39:36 - mmengine - INFO - DeformConv2dPack ida_up.node_3.conv is upgraded to version 2.
video_name DJI_0028_3
seq_info_path /home/ubuntu/qian/HGTMT/VT-Tiny-MOT/test2017/DJI_0028_3/seqinfo.ini
frameRate 15
frame offset : 1
step frame: 00000.jpg
Traceback (most recent call last):
File "/home/ubuntu/qian/HGTMT/tracking/RGBT-Tiny_private_graph_Track2_crossmodal.py", line 504, in
main(tracktor)
File "/home/ubuntu/qian/HGTMT/tracking/RGBT-Tiny_private_graph_Track2_crossmodal.py", line 438, in main
online_ids_i] = tracker.dual_step_reidV3_pre_tracking_vit(batch)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ubuntu/anaconda3/envs/pytorch/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/home/ubuntu/qian/HGTMT/tracking/tracker_rgbt_graph_track_graph_crossmodal2.py", line 620, in dual_step_reidV3_pre_tracking_vit
self.ssim.append(ssim(self.sample_r.tensors.squeeze().permute(1, 2, 0).cpu().numpy(),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ubuntu/anaconda3/envs/pytorch/lib/python3.12/site-packages/skimage/metrics/_structural_similarity.py", line 186, in structural_similarity
raise ValueError(
ValueError: win_size exceeds image extent. Either ensure that your images are at least 7x7; or pass win_size explicitly in the function call, with an odd value less than or equal to the smaller side of your images. If your images are multichannel (with color channels), set channel_axis to the axis number corresponding to the channels.
ERROR conda.cli.main_run:execute(125): conda run python /home/ubuntu/qian/HGTMT/tracking/RGBT-Tiny_private_graph_Track2_crossmodal.py failed. (See above for error)

万分感谢，期待您的回复