Have any one meet this problem when "python train.py"? about deblurganv2

[17666107783]

I solved this problem! actually, this is error of pytorch, it should be : RuntimeError: expected type torch.FloatTensor but got torch.cuda.FloatTensor, so we can change the code models/loss.py line50-51 from
fakeIm[0, :, :, :] = self.transform(fakeIm[0, :, :, :])
realIm[0, :, :, :] = self.transform(realIm[0, :, :, :])
to
fakeIm[0, :, :, :] = self.transform(fakeIm[0, :, :, :].data.cpu()) realIm[0, :, :, :] = self.transform(realIm[0, :, :, :].data.cpu())
it will be ok!

[lamia482]

the root cause is that the torchvision version is incompatible, just upgrade torchvision to 0.4.0 or above and pytorch correspondingly. In the new version, torchvision transformer(functional.py:215) has fix the bug

[wakanawakana]

File "train.py", line 192, in

There is no line 192 now

[17666107783]

I meet this problems too!

[17666107783]

I solved this problem! actually, this is error of pytorch, it should be : RuntimeError: expected type torch.FloatTensor but got torch.cuda.FloatTensor, so we can change the code models/loss.py line50-51 from
fakeIm[0, :, :, :] = self.transform(fakeIm[0, :, :, :])
realIm[0, :, :, :] = self.transform(realIm[0, :, :, :])
to
fakeIm[0, :, :, :] = self.transform(fakeIm[0, :, :, :].data.cpu()) realIm[0, :, :, :] = self.transform(realIm[0, :, :, :].data.cpu())
it will be ok!

[t-martyniuk]

looks like the issue is not relevant anymore, closing