Skip to content

Pytorch implementation of a Conditional WGAN with Gradient Penalty

Notifications You must be signed in to change notification settings

gcucurull/cond-wgan-gp

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Pytorch Conditional WGAN with Gradient Penalty

Pytorch implementation of a Conditional WGAN with Gradient Penalty (GP).

This implementation is adapted from the Conditional GAN and WGAN-GP implementations in this amazing repository with many different GAN model.

Usage

Just run

python main.py

It will create an images directory and save generated images every few iterations.

It can be trained with MNIST (default) or Fashion-MNIST just by adding the flag --dataset fashion.

Example of the images generated by the model, conditioned by class.

Generated samples evolution as training progresses:

About

Pytorch implementation of a Conditional WGAN with Gradient Penalty

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages