RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead.

[Lss6666]

各位好友们：
我想用HICO数据集重新训练ResNet50 - dcn网络（人体关键点136个）。
但是当我运行python /data/open_mmlab/frames/AlphaPose/scripts/train.py --cfg /data/open_mmlab/frames/AlphaPose/configs/halpe_136/resnet/256x192_res50_lr1e-3_2x-dcn-combined.yaml时，出现以下问题：
Traceback (most recent call last):
File "/data/open_mmlab/frames/AlphaPose/scripts/train.py", line 345, in
main()
File "/data/open_mmlab/frames/AlphaPose/scripts/train.py", line 292, in main
loss, miou = train(opt, train_loader, m, criterion, optimizer, writer)
File "/data/open_mmlab/frames/AlphaPose/scripts/train.py", line 96, in train
loss.backward()
File "/workspace/miniconda3/envs/alps/lib/python3.7/site-packages/torch/tensor.py", line 221, in backward
torch.autograd.backward(self, gradient, retain_graph, create_graph)
File "/workspace/miniconda3/envs/alps/lib/python3.7/site-packages/torch/autograd/init.py", line 132, in backward
allow_unreachable=True) # allow_unreachable flag
File "/workspace/miniconda3/envs/alps/lib/python3.7/site-packages/torch/autograd/function.py", line 89, in apply
return self._forward_cls.backward(self, *args) # type: ignore
File "/workspace/miniconda3/envs/alps/lib/python3.7/site-packages/torch/autograd/function.py", line 210, in wrapper
outputs = fn(ctx, *args)
File "/data/open_mmlab/frames/AlphaPose/alphapose/models/layers/dcn/deform_conv.py", line 110, in backward
cur_im2col_step)
RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead.
有一种解决方式是根据cezannec/capsule_net_pytorch#4
找出view()这个函数，在前面用上. reshape or a call to .contiguous()
但是deform_conv_cuda.cpp中的view()函数很多，不知道从何下手。
希望能和大家一起讨论交流，谢谢各位的解答！

[tonyskypc]

我将这个函数的所有view替换成reshape，重新编译安装，就可以了，你尝试一下。

[CosmosView]

我将这个函数的所有view替换成reshape，重新编译安装，就可以了，你尝试一下。

Nice work! Thanks a lot!