This repository contains an implementation of Coulomb GANs as depicted in the paper Coulomb GANs: Provably Optimal Nash Equilibria via Potential Fields by Unterthiner et al. The Coulomb GAN is the first GAN that can guarantee that it learns an optimal Nash Equilibrium, i.e., it does not suffer from mode collapse.
There are implementations for both Tensorflow and PyTorch. The TF version was used to produce the results in the paper, but the PyTorch code is easier and might be easier to understand.
Note that the code is provided "as-is" under the terms of the General Public License v2. See LICENSE.txt for the full details.