Introduction

This is the implementation of Wasserstein Auto-Encoders paper in PyTorch.

For simplicity, I just use MNIST data with MLP architecture instead of DC-GAN for the encoder/decoder/discriminator, but you can replace them easily.

Requirement

• python 3
• PyTorch >= 0.3
• torchvision
• numpy

Train

• To train an adversarial autoencoder:

python aae.py

• To train a WAE-GAN:

python wae_gan.py

• To train a WAE-MMD:

python wae_mmd.py