Can we change the U-net to fit any size of the input image?

[j96w]

I tried to input the U-net with different size of the input image. It seems that sometimes when the size number is odd, it will lead to some problem when we upsample, especially when we concat different Xi(x, x1, x2, x3, x4, x5). Is there some way we can change the U-net to fit any size of the input image? Thankyou.

[milesial]

Yes I know about this issue, I mention it in the README. It seems to me that there should be no problem when upsampling thanks to the padding, that cause the two sizes to match up in order to concat. But I think my padding implementation (https://github.com/milesial/Pytorch-UNet/blob/master/unet/unet_parts.py#L66) may be wrong.

Check out the forks (https://github.com/milesial/Pytorch-UNet/network), some people have modified this part of the code, but I have not tested it.

More specifically https://github.com/HaiyongJiang/U-Net-Pytorch-Unstructured-Buggy/commit/0e854509c2cea854e247a9c615f175f76fbb2e3a

[j96w]

Ya, after deleting one of the ')' and switch diffX and diffY in HaiyongJiang's code. Everything works well. Thank you.