runtimeError: If your Function modifies inplace an input that is a view of another Tensor, your Functionannot return more than one Tensor, This is not supported by the current autograd engine, You shouldither make sure the input is not a view (using .clone() for example) or make your Function only returnone Tensor (potentially splitting it into two Functions: one doing the inplace that returns a singleTensor and a second one that does the other operations).You can ask on the forum https: //discuss.pytorch.org/ if you need help to do this change.

[zhanghaochen-817]

When I ran train.py, the program reported an error.

The error occurred on line 96 in the train.py file: outputs=model (fimg, labels,...) I think it was an error in the model, and then I debugged the model function and found that it was resnet that reported an error in outputting each layer's feature map statement: that is, outputting feat4 1, feat8 1. feat16 1. feat32_ One error was reported, and finally, I debugged again and entered the BN layer of the resnet network. I have never seen an error before.

Can you tell me if it's a data problem or something else, I feel that the resnet model is definitely correct, but the cityscapes dataset I do use is fine

[lpetflo]

I noticed the same issue when trying to use td4-psp18_cityscapes.yml for training. Could you please mention your fix for this?