kzkadc / wasserstein-gan

Wasserstein GAN(-GP) implemented with PyTorch and Chainer

Wasserstein GAN(-GP)

Arjovsky, Martin, Soumith Chintala, and Léon Bottou. "Wasserstein generative adversarial networks." International Conference on Machine Learning. 2017.
Gulrajani, Ishaan, et al. "Improved training of wasserstein gans." Advances in Neural Information Processing Systems. 2017.

Implemented with Chainer and PyTorch.

Requirements (Chainer)

Chainer, OpenCV

$ pip install chainer opencv-python

Requirements (PyTorch)

PyTorch (see the official document), PyTorch-Ignite, OpenCV, Matplotlib

$ pip install pytorch-ignite opencv-python matplotlib

How to run

$ python wgan.py [options]

You can read help with -h option.

$ python wgan.py -h
usage: wgan.py [-h] [-b BATCHSIZE] [-e EPOCH] [--alpha ALPHA] [--beta1 BETA1]
               [--beta2 BETA2] [--n_cri N_CRI N_CRI] [--gp_lam GP_LAM] [-g G]
               [--result_dir RESULT_DIR]

WGAN(-GP)

optional arguments:
  -h, --help            show this help message and exit
  -b BATCHSIZE, --batchsize BATCHSIZE
  -e EPOCH, --epoch EPOCH
  --alpha ALPHA
  --beta1 BETA1
  --beta2 BETA2
  --n_cri N_CRI N_CRI
  --gp_lam GP_LAM
  -g G                  GPU ID (negative value indicates CPU mode)
  --result_dir RESULT_DIR