ReLLIE: Deep Reinforcement Learning for Customized Low-Light Image Enhancement

This repository contains the official implementation of the ACMMM 2021 paper ReLLIE: Deep Reinforcement Learning for Customized Low-Light Image Enhancement.

Introduction

To tackle the low-light image enhancement (LLIE) problem, we propose a novel deep reinforcement learning based method, dubbed ReLLIE, for customized low-light enhancement. Specifically, ReLLIE models LLIE as a markov decision process, i.e., estimating the pixel-wise image-specific curves sequentially and recurrently. Given the reward computed from a set of carefully crafted non-reference loss functions, a lightweight network is proposed to estimate the curves for enlightening of a low-light image input. For more details, please refer to our orginal paper.

Requirement

• Python 3.5+
• Chainer 5.0+
• Cupy 5.0+
• OpenCV 3.4+
• Torch 1.6

You can install the required libraries by the command pip install -r requirements.txt. We checked this code on cuda-10.0 and cudnn-7.3.1.

Usage

Training

If you want to train the model

1. git clone [email protected]:GuoLanqing/ReLLIE.git
2. download the training data LOL dataset or your own dataset
3. unzip and put the downloaded "ours485" and "eval15" folders to root folder

python train.py

Testing with pretrained models

If you want to test the pretrained model on noisy low-light images (enhancement with denoising)

python test.py

or on high-quality low-light images (enhancement without denoising)

python test_el.py

References

We used the publicly avaliable pretrained models of FFDNet as the denoiser module.

We obtained the LOL and LIME dataset from

• https://daooshee.github.io/BMVC2018website/
• https://drive.google.com/file/d/0BwVzAzXoqrSXb3prWUV1YzBjZzg/view?resourcekey=0-VZXvwdwr7QbH3FoX10yPXg

Our implementation is based on PixelRL. We would like to thank them.

Citation

Preprint available here.

In case of use, please cite our publication:

R. Zhang, L. Guo, S. Huang and B. Wen, "ReLLIE: Deep Reinforcement Learning for Customized Low-Light Image Enhancement," ACM MM 2021.

Bibtex:

@article{zhang2021rellie,
  title={ReLLIE: Deep Reinforcement Learning for Customized Low-Light Image Enhancement},
  author={Zhang, Rongkai and Guo, Lanqing and Huang, Siyu and Wen, Bihan},
  journal={arXiv preprint arXiv:2107.05830},
  year={2021}
}

Contact

If you have any questions, please contact [email protected] or [email protected]