discrete-embeddings

This repository contains TensorFlow 2 implementations of a few VQ-VAE papers that caught my interest. The core idea is to create a Variational Autoencoders which bottleneck into a discrete representation. Each directory contains a paper implementation and results from experiments. Currently, the following papers are implemented:

• Vanilla VAE (paper)
• VQ-VAE (paper)

I plan to add the following papers to this repository:

• VQ-VAE 2 (paper)
• Jukebox (paper)

Setup

All the implementations require:

• Python 3;
• TensorFlow 2;
• Tensorboard (to inspect training metrics);
• Matplotlib (to generate plots);

At the root of this repository, there is a requirements.txt file listing the installed packages on my development environment.