Ocean 训练问题

[tm9161]

您好，请教个问题，我按照教程运行python tracking/onekey.py（单独运行train_ocean.py 错误一样）的时候遇到下面报错，不知道是什么问题?

Traceback (most recent call last): File "./tracking/train_ocean.py", line 259, in main() File "./tracking/train_ocean.py", line 250, in main model, writer_dict = ocean_train(train_loader, model, optimizer, epoch + 1, curLR, config, writer_dict, logger, device=device) File "/data/code/siam/TracKit/tracking/../lib/core/function.py", line 54, in ocean_train loss.backward() File "/home/tm/anaconda3/lib/python3.8/site-packages/torch/tensor.py", line 245, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "/home/tm/anaconda3/lib/python3.8/site-packages/torch/autograd/init.py", line 145, in backward Variable._execution_engine.run_backward( File "/home/tm/anaconda3/lib/python3.8/site-packages/torch/autograd/function.py", line 89, in apply return self._forward_cls.backward(self, *args) # type: ignore File "/home/tm/anaconda3/lib/python3.8/site-packages/torch/autograd/function.py", line 210, in wrapper outputs = fn(ctx, *args) File "/data/code/siam/TracKit/tracking/../lib/models/dcn/deform_conv.py", line 85, in backward deform_conv_cuda.deform_conv_backward_parameters_cuda( 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.

[JudasDie]

您好，请教个问题，我按照教程运行python tracking/onekey.py（单独运行train_ocean.py 错误一样）的时候遇到下面报错，不知道是什么问题?

Traceback (most recent call last): File "./tracking/train_ocean.py", line 259, in main() File "./tracking/train_ocean.py", line 250, in main model, writer_dict = ocean_train(train_loader, model, optimizer, epoch + 1, curLR, config, writer_dict, logger, device=device) File "/data/code/siam/TracKit/tracking/../lib/core/function.py", line 54, in ocean_train loss.backward() File "/home/tm/anaconda3/lib/python3.8/site-packages/torch/tensor.py", line 245, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "/home/tm/anaconda3/lib/python3.8/site-packages/torch/autograd/init.py", line 145, in backward Variable._execution_engine.run_backward( File "/home/tm/anaconda3/lib/python3.8/site-packages/torch/autograd/function.py", line 89, in apply return self._forward_cls.backward(self, *args) # type: ignore File "/home/tm/anaconda3/lib/python3.8/site-packages/torch/autograd/function.py", line 210, in wrapper outputs = fn(ctx, *args) File "/data/code/siam/TracKit/tracking/../lib/models/dcn/deform_conv.py", line 85, in backward deform_conv_cuda.deform_conv_backward_parameters_cuda( 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.

看起来是你没有成功编译deform conv. 检查下环境和install.sh是不是一样。或者去掉align训练没有align的。

[tm9161]

您好，请教个问题，我按照教程运行python tracking/onekey.py（单独运行train_ocean.py 错误一样）的时候遇到下面报错，不知道是什么问题? Traceback (most recent call last): File "./tracking/train_ocean.py", line 259, in main() File "./tracking/train_ocean.py", line 250, in main model, writer_dict = ocean_train(train_loader, model, optimizer, epoch + 1, curLR, config, writer_dict, logger, device=device) File "/data/code/siam/TracKit/tracking/../lib/core/function.py", line 54, in ocean_train loss.backward() File "/home/tm/anaconda3/lib/python3.8/site-packages/torch/tensor.py", line 245, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "/home/tm/anaconda3/lib/python3.8/site-packages/torch/autograd/init.py", line 145, in backward Variable._execution_engine.run_backward( File "/home/tm/anaconda3/lib/python3.8/site-packages/torch/autograd/function.py", line 89, in apply return self._forward_cls.backward(self, *args) # type: ignore File "/home/tm/anaconda3/lib/python3.8/site-packages/torch/autograd/function.py", line 210, in wrapper outputs = fn(ctx, *args) File "/data/code/siam/TracKit/tracking/../lib/models/dcn/deform_conv.py", line 85, in backward deform_conv_cuda.deform_conv_backward_parameters_cuda( 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.

看起来是你没有成功编译deform conv. 检查下环境和install.sh是不是一样。或者去掉align训练没有align的。

我重新编译了一下deform_conv 还是不行，看了一下install.sh里的配置，因为我用的30的卡，cuda是11.1，torch1.8.1，还有一个是mpi4py这个没安装成功，其他都按照install里安的，不知道有没有关系。

我自己再看看吧，谢谢回复。

[Jee-King]

@tm9161 您好 请问你解决这个问题了么，我也遇到同样的问题了

[Jee-King]

@JudasDie 您好，把align参数设置成False，就不会出现这个问题了。请问 这个参数会对性能有较大的影响吗？

[tm9161]

@tm9161 您好 请问你解决这个问题了么，我也遇到同样的问题了

没，我也是设置了False。

[l-sf]

@tm9161 你好，我是3080ti + cudatoolkit11.1 + torch1.8 ，在python setup.py develop这一步编译就报错了，感觉是cuda版本太高的问题，请问你遇到这个问题了吗？怎么解决的？

[JudasDie]

Please refer to the new repo. of JudasDie/SOTS. Thx.

l-sf @.***> 于2022年7月12日周二 20:37写道：

@tm9161 https://github.com/tm9161 你好，我是3080ti + cudatoolkit11.1 + torch1.8 ，在python setup.py develop这一步编译就报错了，感觉是cuda版本太高的问题，请问你遇到这个问题了吗？怎么解决的？

— Reply to this email directly, view it on GitHub https://github.com/researchmm/TracKit/issues/81#issuecomment-1181708701, or unsubscribe https://github.com/notifications/unsubscribe-auth/AF6U2PDIDVBCFSXBLHWR7VTVTVRKFANCNFSM46HFC5YA . You are receiving this because you were mentioned.Message ID: @.***>

-- From: Zhang Zhipeng Institution: National Laboratory of Pattern Recognition Address: 95 Zhongguancun East Road, 100190, BEIJING, CHINA Email: @.***

Best Wishes