Skip to content

Latest commit

 

History

History
28 lines (20 loc) · 874 Bytes

README.md

File metadata and controls

28 lines (20 loc) · 874 Bytes

GAN

Pytorch implementation of the DCGAN paper and Improved Techniques for Training GANs. Currently, only feature matching is implemented. Minibatch discrimination, label smoothing and historical averaging will be implemented soon.

Demo

CelebA dataset (vanilla adversarial loss)

Alt Text

CelebA dataset (feature matching loss)

Alt Text

How to use

Put your images in a folder and specify path to that folder in config/hyperparameters.yaml. Also, other training configurations can be modified in the file. Then run python main.py.

Training progress will be logged in WandB.

TODO

  • Minibatch discrimination
  • Label smoothing
  • Historical averaging