/GANs

Generative adversarial networks

Primary LanguagePython

GAN

  • dcgan_train: several GANs (loss functions, gradient penalty, spectral normalization, network architecture)
  • evaluation: FID and inception scores
  • some helper function: to visualize training progress and generated images and prepare data
  • post filtering: remove isolated pixels

Training progress

WGAN + spetral normalization + gradient penalty 10

loss

score

evolution

Result

read_fake

no_filter