taucontrib

Thank you very much @Mut1nyJD, i was just about to write a class conditional version by myself. I thought about implementing class conditional batch normalization and a projection discriminator as...

@Mut1nyJD thanks for the tipp. Do you add the fake class additionally to real/fake classification or do you get rid of the real_fake conv = nn.Conv2d(chan, 1, ..) and replace...

Alright got it @Mut1nyJD One last question: The standard CrossEntropyLoss doesn't work for multi label classification. This means that the generated images only belong to the fake class and not...

Same problem here! Did you found any solution in the meantime?

`fast 8/3` not `fast 3` in the second code line

Ok so this behavior is expected? Then I got the documentation wrong - closing this issue now, thanks!:)