Code for reproducing experiments in Wasserstein Divergence for GANs.
Our code is built upon WGAN-GP.
For Tensorflow and its dependencies please check https://github.com/igul222/improved_wgan_training
For the training data such as Cifar10, CelebA, LSUN etc download them on the official website accordingly
For FID evaluation download the Inception model from http://download.tensorflow.org/models/image/imagenet/inception-2015-12-05.tgz
For the precalculated statistics download them from http://bioinf.jku.at/research/ttur/
Specify DATA_DIR, INCEPTION_DIR and STAT_FILE in wgan-div.py, then run
python wgan-div.py
The default training image size is 64*64.