VAE Models

This project is about different small Variational Autoencoder models trained on CIFAR10 dataset.

What exactly are these models?

• a standard VAE (convvae)
• Variational Lossy Autoencoder (vlae)
• Vector Quantised-Variational AutoEncoder (vq-vae)

Run

1. Install all dependencies listed in requirements.txt. Note that the model has only been tested in the versions shown in the text file.
2. Choose an appropriate model
   • convvae stands for classic Variational AutoEncoder
   • vlae stands for Variational Lossy AutoEncoder
   • vq-vae stands for Vector Quantised-Variational AutoEncoder (default)

cd src && python3 main.py --name vlae

As far as the output, several plots will be saved in images directory (train plot + samples + reconstruction images)

Training stage takes about 3 hours (on GPU NVIDIA GeForce GTX 1080 Ti).