A method to to underwater image enhancement.
CycleGan is used to generate paired images to train UResnet.
[paper]
here is a video clip to show the results of our model
https://www.bilibili.com/video/av75799712/
the original videos we used are from Youtube
https://www.youtube.com/watch?v=VTmLIzz7jUU
https://www.youtube.com/watch?v=E4ZxR1C0xmc
- Linux (code have not been tested on Windows or OSX)
- NVIDIA GPU + CUDA CuDNN
- pytorch >= 0.4.1
use python train.py --input_images_path xxx/path/to/input/ --label_images_path xxx/path/to/label --snapshots_folder xxx/path/to/save/snapshot
use python train.py -h to choose more usable options
The training set should be organized like
Input image: xxx/path/input/abcd.jpg Label image: xxx/path/label/abcd_label.jpg