Work in progress...

• This repository contains source code for WaterGAN developed in WaterGAN: Unsupervised Generative Network to Enable Real-time Color Correction of Monocular Underwater Images.
• This code is modified from Taehoon Kim's DCGAN-tensorflow (MIT-licensed). Our modifications are MIT-licensed.

Download data:

1. MHL test tank dataset: MHL.tar.gz
2. Jamaica field dataset: Jamaica.tar.gz

Coming soon...

Train a model:

Original in-air images:

Synthetic underwater images produced by WaterGAN:

WaterGAN outputs a dataset with paired true color, depth, and (synthetic) underwater images. We can use this to train an end-to-end network for underwater image restoration. Source code and pretrained models for the end-to-end network are available here. For more details, see the paper.

Raw underwater images gathered from a survey in a pure water tank:

Corrected images using data generated with WaterGAN to train an end-to-end underwater image restoration network:

@article{li2017watergan,
    author    = {Jie Li, Katherine A. Skinner, Ryan M. Eustice and
               Matthew Johnson{-}Roberson},
  title     = {WaterGAN: Unsupervised Generative Network to Enable Real-time Color
               Correction of Monocular Underwater Images},
  journal   = {CoRR},
  volume    = {abs/1702.07392},
  year      = {2017},
  url       = {http://arxiv.org/abs/1702.07392},
  timestamp = {Wed, 01 Mar 2017 14:26:00 +0100},
  biburl    = {http://dblp.uni-trier.de/rec/bib/journals/corr/LiSEJ17},
  bibsource = {dblp computer science bibliography, http://dblp.org}
}