Our code can be split up into two implementation sections -- one on Normalising Flows, and one on the PyTorch VQVAE.
We built this code on original code at https://github.com/VincentStimper/normalizing-flows with initial authors.
- Vincent Stimper
- Lukas Ryll
- David Liu
Our contributions include
- A Gumbel-Softmax target distribution class
- Code for training the flows
We built this code on original code at https://github.com/ritheshkumar95/pytorch-vqvae with initial authors.
- Rithesh Kumar
- Tristan Deleu
- Evan Racah
Our contributions include
- Changed way of doing stop-gradient
- Code for test-time optimizations
- Visualizing/sampling code
- Extension of PixelCNN prior to other datasets.
- Method for Normalizing Flows