GamutNet: Restoring-Wide-Gamut-Colors-for-Camera-Captured-Images

Hoang M. Le, Taehong Jeong, Abdelrahman Abdelhamed, Hyun Joon Shin and Michael S. Brown

Repo for the paper: GamutNet: Restoring Wide-Gamut Colors for Camera-Captured Images

Overview

Dataset:

1. Use these links to download our dataset:

• ProPhoto Images: prop-8bpc.zip 106.4GB
• sRGB Images: srgb-8bpc-1.rar 57.1GB srgb-8bpc-2.rar 55.9GB
• split files: split.zip 65KB

2. You then need to uncompress ProPhoto to the folder prop-8bpc and all two sRGB parts into the folder srgb-8bpc. Each of them has 5000 images.
3. You need to modify the path of the above folders in split txt files

How to start training

1. Some basic running scripts:

python trainer_main.py --split_path SPLIT_PATH 

python trainer_main.py --split_path ./split_output --model_size small --patch_size 64 64 --batch_size 64 --gpus "1" --train_num_workers 32 --val_num_workers 32 --max_patches_per_image 1000 --max_epochs 20

2. Specifying hint_mode:

... --hint_mode HINT_MODE --input_channels INPUT_CHANNELS

• --hint_mode none --input_channels 3: no hint will be provided.
• --hint_mode o2o_all --input_channels 4: one-to-one mask (white, in-gamut, and black) will be provided.
• --hint_mode o2o_rgb --input_channels 6: per-channel one-to-one mask will be provided.
• --hint_mode pw_27 --input_channels 4: the pw_mask_27 normalized by 26 will be provided.

3. Specifying `patch_size':

... --patch_size PATCH_HEIGHT PATCH_WIDTH

• e.g. --patch_size 32 32
• e.g. --patch_size 64 64
• e.g. --patch_size 128 128

4. Specifying max_patches_per_image:

... --max_patches_per_image MAX_PATCHES_PER_IMAGE

• e.g. given batch_size=32, --max_patches_per_image 32000 will yield 1000 iterations per image
• e.g. given batch_size=128, --max_patches_per_image 12800: will yield 100 iterations per image

5. Specifying num_workers for DataLoader:

• --train_num_workers: default is 2
• --val_num_workers: default is 1
• --test_num_workers: default is 1

Citation

@inproceedings{le2021gamutnet,
  title={GamutNet: Restoring Wide-Gamut Colors for Camera-Captured Images},
  author={Le, Hoang and Jeong, Taehong and Abdelhamed, Abdelrahman and Shin, Hyun Joon and Brown, Michael S},
  booktitle={Color and Imaging Conference},
  volume={2021},
  number={29},
  pages={7--12},
  year={2021},
  organization={Society for Imaging Science and Technology}
}