Another recent member in the GAN family!
Seems to be an improvement over Improved WGAN (Improved^2 GAN)?
To run an example for MNIST generation:
python cramer-gan.py
This is a modification of Improved WGAN, where the losses are changed according to the following: