Skip to content

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

Notifications You must be signed in to change notification settings

gamut-mapping/GamutNet

Repository files navigation

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

overview_figure

Dataset:

  1. Use these links to download our dataset:
  1. 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.
  2. 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
  1. 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.
  1. 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
  1. 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
  1. 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}
}

About

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

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages