search

heykeetae / Self-Attention-GAN

Pytorch implementation of Self-Attention Generative Adversarial Networks (SAGAN)
2.51k stars 470 forks source link

model.py is working only for imsize=64 #49

Open franknb opened 4 years ago

franknb commented 4 years ago

Hi, I have been reading the code and trying to fix the issue, and to make it work for more image sizes at the same time. I believe the minimal imsize could be around 32 if you keep the kernel size, with minor change in the code (you need to add the same if imsize == 64 in forward part of both models and change the generator's init's structure a little bit). If you go lower than 32, you might be facing error of input size smaller than kernel size. If you go up to 256 you definitely need more layers. So there might not be a One Fit for All solution here. Any suggestions?

AlexTS1980 commented 4 years ago

Do you have a hacking (code doesn't work) or a modeling (bad results) problem?