You start to see either coloration diverge a bit or random primary color spots right around 160px. I'm trying lowering the learning rate for now again but I don't think that alone is going to really do it..
Show this thread
Hey Twitter- Many of my followers here know things, so I'll ask: Know of any good GAN stabilization techniques I should try on DeOldify? I just added spectral norm to the resnet34 backbone which helped quite a bit. But I'm still hitting issues at larger image sizes...
-
-
-
For context, this is a Self-Attention GAN adapted to a U-Net generator, and the training is done on progressive sizes (64,96,128,160,192....). I know GAN stability is an ongoing line of research but I thought it'd be worth trying!https://github.com/jantic/DeOldify/blob/master/README.md …
Show this thread
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.