Florian C.F. Schulz

For e.g. DCGAN or pix2pix people usually start with discriminator training with detached generated fakes and then optimize Generator. Not sure why it is the other way round here. Dont...

Not sure why you think that. At least that is the usual way to train GANs, check out https://pytorch.org/tutorials/beginner/dcgan_faces_tutorial.html

Very good job @siriusdemon and thanks for sharing! Tried to push it to GPU but speed is almost same compared to CPU in my case, anybody more successful?

I'll guess that is because he uses nn.CrossEntropyLoss() for the binary classification in the toy dataset. In my opinion, this should be a parameter you can set (see below), so...