You can find more details here: https://li-chongyi.github.io/Proj_Zero-DCE.html. Have fun!

The implementation of Zero-DCE is for non-commercial use only.

Pytorch implementation of Zero-DCE

1. Python 3.7
2. Pytorch 1.0.0
3. opencv
4. torchvision 0.2.1
5. cuda100

Zero-DCE does not need special configurations. Just basic environment.

Or you can create a conda environment to run our code like this: conda create --name zerodce_env opencv pytorch==1.0.0 torchvision==0.2.1 cuda100 python=3.7 -c pytorch

Download the Zero-DCE_code first. The following shows the basic folder structure.

├── data
│   ├── test_data # testing data. You can make a new folder for your testing data, like LIME, MEF, and NPE.
│   │   ├── LIME 
│   │   └── MEF
│   │   └── NPE
│   └── train_data 
├── lowlight_test.py # testing code
├── lowlight_train.py # training code
├── model.py # Zero-DEC network
├── dataloader.py
├── snapshots
│   ├── Epoch99.pth #  A pre-trained snapshot (Epoch99.pth)

cd Zero-DCE_code

python lowlight_test.py 

The script will process the images in the sub-folders of "test_data" folder and make a new folder "result" in the "data". You can find the enhanced images in the "result" folder.

1. cd Zero-DCE_code

2. download the training data google drive or baidu cloud [password: 1234]

3. unzip and put the downloaded "train_data" folder to "data" folder

python lowlight_train.py 

@inproceedings{Zero-DCE,
 author = {Guo, Chunle Guo and Li, Chongyi and Guo, Jichang and Loy, Chen Change and Hou, Junhui and Kwong, Sam and Cong, Runmin},
 title = {Zero-reference deep curve estimation for low-light image enhancement},
 booktitle = {Proceedings of the IEEE conference on computer vision and pattern recognition (CVPR)},
 pages    = {1780-1789},
 month = {June},
 year = {2020}
}

(Full paper: http://openaccess.thecvf.com/content_CVPR_2020/papers/Guo_Zero-Reference_Deep_Curve_Estimation_for_Low-Light_Image_Enhancement_CVPR_2020_paper.pdf)

If you have any questions, please contact Chongyi Li at [email protected] or Chunle Guo at [email protected].

Thanks tuvovan ([email protected]) who re-produces our code by TF. The results of TF version look similar with our Pytorch version. But I do not have enough time to check the details. https://github.com/tuvovan/Zero_DCE_TF