information-bottleneck-gan

In progress Pytorch implementation of the paper Information bottleneck through variational glasses.

python train_ibn_dcgan.py --epochs=150 --dataset-name=CelebA --g-learning-rate=2e-4 --eg-learning-rate=2e-4 
                            --d-learning-rate=2e-4
                            --data-dir= /path/to/noncropped/CelebA/

CelebA generation after 12 epochs: