Pytorch implementation of unsupervised GANs.
For more defails about calculating Inception Score and FID Score using pytorch can be found in Pytorch-Unified-Inception-FID-Score
- DCGAN
- WGAN
- WGAN-GP
- SN-GAN
- python 3.6
- Install python packages
pip install -U pip setuptools pip install -r requirements.txt
Model | Dataset | Inception Score | FID Score |
---|---|---|---|
DCGAN | CIFAR10 | 6.111(0.088) | 41.75 |
WGAN(CNN) | CIFAR10 | 6.605(0.071) | 30.73 |
WGAN-GP(CNN) | CIFAR10 | 7.415(0.065) | 21.89 |
WGAN-GP(ResNet) | CIFAR10 | 7.829(0.076) | 15.57 |
SNGAN(CNN) | CIFAR10 | 7.521(0.111) | 20.41 |
SNGAN(ResNet) | CIFAR10 | 8.214(0.094) | 14.41 |
Download cifar10_stats.npz for calculating FID score
and put it to ./stats/cifar10_stats.npz
which is the default path
- DCGAN
python gans/dcgan.py --flagfile ./config/DCGAN_CIFAR10.txt
- WGAN(CNN)
python gans/wgan.py --flagfile ./config/WGAN_CIFAR10_CNN.txt
- WGAN-GP(CNN)
python gans/wgangp.py --flagfile ./config/WGANGP_CIFAR10_CNN.txt
- WGAN-GP(ResNet)
python gans/wgangp.py --flagfile ./config/WGANGP_CIFAR10_RES.txt
- SNGAN(CNN)
python gans/sngan.py --flagfile ./config/SNGAN_CIFAR10_CNN.txt
- SNGAN(ResNet)
python gans/sngan.py --flagfile ./config/SNGAN_CIFAR10_RES.txt
python tools/sample2gif.py --logdir path/to/logdir
e.g.
python tools/sample2gif.py --logdir ./logs/DCGAN_CIFAR10
output GIF is ./logs/DCGAN_CIFAR10/progress.gif