New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Self-Attention GAN (SAGAN) example #66

merged 1 commit into from Dec 13, 2018


None yet
2 participants
Copy link

TE-KazukiYoshiyama commented Dec 12, 2018

Self-Attention Generative Adversarial Networks (SAGAN) is a good GAN example since it contains many recently proposed techniques for training GANs, the following techniques are included in this SAGANs example.

  • Spectral normalization: normalization of weights with its singular value to constrain the discriminator Lipchitz norm
  • Two Time-Scale Update Rule (TTUR): unbalanced learning rate for the generator and discriminator
  • Self-Attention: Attention of the whole region of an image
  • cGANs with Projection Discriminator: projection of labels in the discriminator at the end, then use it in the adversarial loss
  • Class Conditional Batch Normalaization (CCBN): Incorporation of label information into the gamma and beta of the batch normalization.

This comment has been minimized.

Copy link

TE-StefanUhlich commented Dec 13, 2018

@TE-KazukiYoshiyama Thanks for this nice example and all the effort you put in.

@TE-StefanUhlich TE-StefanUhlich merged commit 3b1421e into master Dec 13, 2018

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment