junyanz
commented
4 years ago Do you mean an auto-encoder? You can still use torch.cat() if you don't want to change the code. But an unconditional discriminator D(G(x)) should also work (rather than using conditional discriminator D(G(x), x)). You need to change the input_nc for the discriminator.
in some case, The image that needs to be converted also can be also a condition image,but now this image should be consider once or twice at the process of torch.cat() at the generator ?