Variational Autoencoder

Example of vanilla VAE for face image generation at resolution 128x128.

Auto-Encoding Variational Bayes: https://arxiv.org/abs/1312.6114

Generation:

Original Faces vs. Reconstructed Faces:

How to Run

You need to have pytorch >= v0.4.1 and cuda/cuDNN drivers installed.

To install requirements:

pip install -r requirements.txt

To download and prepare dataset:

python prepare_celeba.py

To train:

python VAE.py