Sobolev Wasserstein GAN

This repo contains a reference implementation for SWGAN as described in the paper:

Towards Generalized Implementation of Wasserstein Distance in GANs
Minkai Xu, Zhiming Zhou, Guansong Lu, Jian Tang, Weinan Zhang, Yong Yu
AAAI Conference on Artificial Intelligence (AAAI), 2021.
Paper: https://arxiv.org/abs/2012.03420

The implementation is built upon the repo WGAN-GP, code for reproducing experiments in "Improved Training of Wasserstein GANs".

Prerequisites

• Python, NumPy, TensorFlow, SciPy, Matplotlib
• A recent NVIDIA GPU

Models

Configuration for all models is specified in a list of constants at the top of the file. Two models should work "out of the box":

• python gan_toy.py: Toy datasets (8 Gaussians, 25 Gaussians, Swiss Roll).

For the other models, edit the file to specify the path to the dataset in DATA_DIR before running. Each model's dataset is publicly available; the download URL is in the file.

• python gan_cifar_resnet.py: CIFAR-10

Citing

If you find SWGAN useful in your research, please consider citing the following two papers:

@article{xu2020towards,
  title={Towards Generalized Implementation of Wasserstein Distance in GANs},
  author={Xu, Minkai and Zhou, Zhiming and Lu, Guansong and Tang, Jian and Zhang, Weinan and Yu, Yong},
  journal={AAAI Conference on Artificial Intelligence (AAAI), 2021.},
  year={2020}
}

@article{gulrajani2017improved,
  title={Improved training of wasserstein gans},
  author={Gulrajani, Ishaan and Ahmed, Faruk and Arjovsky, Martin and Dumoulin, Vincent and Courville, Aaron},
  journal={arXiv preprint arXiv:1704.00028},
  year={2017}
}