Skip to content

Conditional Wasserstein Generative Adversarial Network for image-to-image translation.

License

Notifications You must be signed in to change notification settings

JoshuaEbenezer/cwgan

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

14 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

CWGAN

Conditional Wasserstein Generative Adversarial Network for image-to-image translation.

Implementation of the paper: Single Image Haze Removal Using Conditional Wasserstein Generative Adversarial Networks.

Please cite the following work if you use this code:

Single Image Haze Removal Using Conditional Wasserstein Generative Adversarial Networks J.P. Ebenezer, B. Das, S. Mukhopadhyay 2019 27th European Signal Processing Conference (EUSIPCO), 1-5

Arxiv link: https://arxiv.org/pdf/1903.00395.pdf

Training a model

  1. Clone/download the repo
  2. Go to ./scripts/
  3. Change the database location and the other options in train_pix2pix.sh and execute it.

Testing a model

  1. After training the model, go to ./scripts/
  2. Change the database location and the other options in test_pix2pix.sh and execute it.

Acknowledgements

This code is based on https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix (for cGAN) and https://github.com/caogang/wgan-gp (for wGAN).

About

Conditional Wasserstein Generative Adversarial Network for image-to-image translation.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published