fangzhimu / rcd Goto Github PK

The official pytorch implementation of the paper [Real-time Controllable Denoising for Image and Video]

| Github | Page | Paper | Arxiv |

This implementation based on [BasicSR]

Basic requirements:

python 3.9.12
pytorch 1.12.1
cuda 11.8

Other requirements:

pip install -r requirements.txt
python setup.py develop --no_cuda_ext

Evaluate on Nam Dataset:

Data Preparation:

1. Download Nam dataset
2. Crop the gt and input images into 512*512 patches and save as gt.lmdb and input.lmdb, respectively. (or download from Google Drive GT and Input)
3. Edit the dataroot_lq and dataroot_gt in NAFNet-RCD-tiny.yml to the corresponding paths: /your_path/gt.lmdb and /your_path/input.lmdb

Test Nam real image noise dataset with NAFNet-RCD-tiny model, which is trained on SIDD training dataset

python basicsr/test.py --opt options/test/NAFNet-RCD-tiny.yml

Evaluate on SIDD Dataset(noise level $\sigma$ = 0-12):

Data Preparation:

1. Download the evaluation data (in lmdb format) SIDD_0_12
2. Edit the dataroot_lq and dataroot_gt in NAFNet-RCD-tiny.yml to the corresponding paths: /your_path/gt.lmdb and /your_path/input.lmdb

Test SIDD real image noise dataset with NAFNet-RCD-tiny model

python basicsr/test.py --opt options/test/NAFNet-RCD-tiny.yml

Result structure:

Groundtruth

  imageName_gt.png 

Denoise levels

  imageName_level_0.png imageName_level_1.png imageName_level_2.png imageName_level_3.png imageName_level_4.png 

AutoTune results

  imageName_res.png

Our codebase is based on the NAFNet

If you find our paper useful for your research, please consider citing our work 😊 :

@InProceedings{Zhang_2023_CVPR,
    author    = {Zhang, Zhaoyang and Jiang, Yitong and Shao, Wenqi and Wang, Xiaogang and Luo, Ping and Lin, Kaimo and Gu, Jinwei},
    title     = {Real-Time Controllable Denoising for Image and Video},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2023},
    pages     = {14028-14038}
}