gradient computation problems

[mokby]

I follow the instruction as described in readme.md, and when I running python3 train.py‘, errors occured, I googled and try to modify inplace=True' into 'inplace=False', but that does not help.

RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [4, 128, 64, 64]], which is output 0 of ReluBackward0, is at version 1; expected version 0 instead. Hint: the backtrace further above shows the operation that failed to compute its gradient. The variable in question was changed in there or anywhere later. Good luck!

[KaibingHu]

You can try to edit the 101 row of /models/FPN/FPN_Head.py.Chage += into =

[mokby]

Thanks ,that works

[gs-max]

您可以尝试将 /models/FPN/FPN_Head.py.Chage += 的 101 行编辑为 =
您好，我不太清楚您说的是哪里，我的101行内容是这样的：laterals[i - 1] += F.interpolate(laterals[i], size=prev_shape, mode='nearest')