Skip to content

Experiments using generative methods such as VQVAE, PixelCNN, and autoregressive Transformers.

Notifications You must be signed in to change notification settings

b-hahn/generative_experiments

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

12 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Generative Experiments

This repo contains a number of approaches for generative modeling and is continuously updated as I work with new methods.

Currently, the main model is a VQVAE[0] along with a PixelCNN++ prior. I based my JAX implementation of VQVAE on this notebook and trained it on a custom dataset containing topographic map tiles of Switzerland. PixelCNN was slightly adapted from here to work with learning the latent codes from VQVAE instead of RGB images.

Image Reconstructions

Todos

  • retrain PixelCNN++ for longer for better results

  • conditional generation

  • build web demo

WIP.

References

[0] https://arxiv.org/abs/1711.00937

[1] https://arxiv.org/pdf/1701.05517.pdf

About

Experiments using generative methods such as VQVAE, PixelCNN, and autoregressive Transformers.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages