Skip to content

a-kori/vae_vq-vae

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

100 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

✨ VAE & VQ-VAE

This repository contains the implementation code for two deep learning models: VAE (Variational Autoencoder) and VQ-VAE (Vector Quantized Variational Autoencoder), as well as two project presentations. VAE & VQ-VAE Presentation explains the theory behind the models, and the Project Summary describes the implementation process and the results of our work.

The project was implemented in the context of the Undergraduate Project "Machine Learning" at TU Dortmund and supervised by Marc Höftmann and Sebastian Konietzny. It is based on the papers:

  • “Auto-Encoding Variational Bayes” by Diederik P. Kingma and MaxWelling,
  • “Neural Discrete Representation Learning” by Aaron van den Oord, Oriol Vinyals and Koray Kavukcuoglu

Running the code

To train the models, you can execute the train.ipnyb notebook found in VAE / VQ-VAE directories. Before running the code, make sure that the other Python files are included in the same directory, as they contain dependencies. You can also find the logs and the results of an example run on the CIFAR-10 dataset for each model in train.ipnyb and in the output directory.

Feel free to adjust the dataset and the number of epochs as you like. You can train the VAE on MNIST and CIFRAR-10 datasets, VQ-VAE can also be trained on CelebA. We recommend training the models on a GPU instead of a CPU for faster execution, e.g. on Google Colab.

Authors

Anastasiia Korzhylova, Ivan Shishkin, Ramneek Agnihotri, Rodi Mehi

About

This repository contains the implementation code for the VAE and VQ-VAE machine learning models. The project was implemented in the context of the Undergraduate Project "Machine Learning" at TU Dortmund.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors