AdversarialAutoencoder

Pytorch Lightning implementation of adverserial autoencoder (AAE)

• paper: https://arxiv.org/abs/1511.05644

Result of code

In paper, figure 4 shows the example latent for MNIST dataset. It can be a gaussian mixture or swill roll.

My code can generate each of latent.

• Gaussian Mixture

• Swiss roll

Requirements

• python==3.8.7

How to use

1. Install requirements

pip install -r requirements.txt

2. Run

python src/train.py 

train.py has --sample_latent argument.

2.1) Latent with swiss_roll

python src/train.py --sample_latent swiss_roll

2.2) Latent with gaussian_mixture

python src/train.py --sample_latent gaussian_mixture