AnoGAN-WGAN-pytorch

This is an UNOFFICIAL implement of Pytorch based AnoGAN combined with WGAN

Requirements

python>=3.6
torch==1.0.0
torchvision==0.2.2
albumentations>=0.4

Note

• Current version has supported training for mnist datasets.

Visualization

There is inference results below. Left side is input and right side output.

After training digit "0", in inference stage, the model can recontruct well to "0" but fail to recontruct "8" which is abnormal input.