lucidrains / lightweight-gan

Implementation of 'lightweight' GAN, proposed in ICLR 2021, in Pytorch. High resolution image generations that can be trained within a day or two
MIT License
1.62k stars 220 forks source link

DiffAugment Forward Pass #95

Open shreejalt opened 3 years ago

shreejalt commented 3 years ago

Hi @lucidrains, Thanks a lot for your work. I had one doubt regarding your implementation of DiffAugment.

When you train the discriminator, you pass in D_aug the fake image using detach=True. Can you explain why you do that?

Also, when you pass the real image in D_aug, you don't do the same?

Is it because you pass the requires_grad=True variable i.e generated image and you don't want to pass the gradients through the generator?

Thanking you in advance