THU-luvision / ccdc

Cross-Camera Deep Colorization

tSinghua vIsual intelliGence and coMputational imAging lab ( GitHub | HomePage )

In this repository we provide code of the paper:

Cross-Camera Deep Colorization

Yaping Zhao, Haitian Zheng, Mengqi Ji, Ruqi Huang

arxiv link: https://arxiv.org/abs/2209.01211

Usage

0. For pre-requisites, run:

conda env create -f environment.yml
conda activate ccdc

1. Pretrained model is currently available at Google Drive and Baidu Netdisk (password: ql5l), download the X4_30w.pth, X8_30w.pth and put them in the pretrained folder.

• X4_30w.pth is pretrained on the Vimeo dataset under the scale gap 4X.
• X8_30w.pth is pretrained on the Vimeo dataset under the scale gap 8X.
• If you want to train your own model, please prepare your own training set.

2. For training,
   1. under the scale gap 4X, run:

   sh train_X4.sh
   

   or

   python train_ccdc.py  \
   --dataset demo   \
   --scale 4 \
   --display 100 \
   --batch_size 8  \
   --step_size 25000 \
   --gamma 0.5 \
   --loss CharbonnierLoss \
   --optim Adam \
   --lr 0.0001  \
   --checkpoints_dir ./exp_X4/ \
   --checkpoint_file ./pretrained/X4_30w.pth \
   --frame_num 2 \
   --with_GAN_loss 0 \
   --img_save_path result/colornetcp_exp4 \
   --net_type colornet1 \
   --pretrained 1 \
   --gpu_id 0 \
   --snapshot 5000
   

   2. under the scale gap 8X, run:

   sh train_X8.sh
   

   or

   python train_ccdc.py  \
   --dataset demo   \
   --scale 8 \
   --display 100 \
   --batch_size 8  \
   --step_size 25000 \
   --gamma 0.5 \
   --loss CharbonnierLoss \
   --optim Adam \
   --lr 0.0001  \
   --checkpoints_dir ./exp_X8/ \
   --checkpoint_file ./pretrained/X8_30w.pth \
   --frame_num 2 \
   --with_GAN_loss 0 \
   --img_save_path result/colornetcp_exp5 \
   --net_type colornet1 \
   --pretrained 1 \
   --gpu_id 0 \
   --snapshot 5000
   

3. For testing,
   1. under the scale gap 4X, run:

   sh test_X4.sh
   

   or

   python train_ccdc.py  \
   --mode test \
   --dataset demo   \
   --scale 4 \
   --display 100 \
   --batch_size 1  \
   --step_size 50000 \
   --gamma 0.5 \
   --loss CharbonnierLoss \
   --optim Adam \
   --lr 0.0001  \
   --checkpoints_dir ./exp_X4/ \
   --checkpoint_file ./pretrained/X4_30w.pth \
   --frame_num 2 \
   --with_GAN_loss 0 \
   --img_save_path result/ \
   --net_type colornet1 \
   --pretrained 0 \
   --gpu_id 0 \
   --snapshot 5000
   

   2. under the scale gap 8X, run:

   sh test_X8.sh
   

   or

   python train_ccdc.py  \
   --mode test \
   --dataset demo   \
   --scale 8 \
   --display 100 \
   --batch_size 1  \
   --step_size 50000 \
   --gamma 0.5 \
   --loss CharbonnierLoss \
   --optim Adam \
   --lr 0.0001  \
   --checkpoints_dir ./exp_X8/ \
   --checkpoint_file ./pretrained/X8_30w.pth \
   --frame_num 2 \
   --with_GAN_loss 0 \
   --img_save_path result/ \
   --net_type colornet1 \
   --pretrained 0 \
   --gpu_id 0 \
   --snapshot 5000
   

Dataset

Dataset is stored in the folder dataset/, where subfolders clean/, corrupted/, SISR/ contain ground truth HR images, corrupted LR images, upsampled LR images by interpolation (e.g., bicubic) or SISR methods. Images in SISR/ could be as same as in corrupted/, though preprocessing by advanced SISR methods (e.g., MDSR) brings a small performance boost.

testlist.txt and trainlist.txt could be modified for your experiment on other datasets.

This repo only provides a sample for demo purposes.

Citation

Cite our paper if you find it interesting!

@article{zhao2022cross,
  title={Cross-Camera Deep Colorization},
  author={Zhao, Yaping and Zheng, Haitian and Ji, Mengqi and Huang, Ruqi},
  journal={arXiv preprint arXiv:2209.01211},
  year={2022}
}