CERL: A Unified Optimization Framework forLight Enhancement with Realistic Noise

Code used for CERL: A Unified Optimization Framework for Light Enhancement with Realistic Noise.

Representitive Results

Overal Architecture

Training

Download LOL dataset provided by this github respository. Put them in ./data
Download pre-trained DnCNN parameters from this github respository. Put them in ./original_model

python train.py

The parameters of fine-tuned denoising network (DnCNN) will be saved in ./checkpoints for testing.

Testing

Download pre-trained model parameters from Google Drive. Put them in ./checkpoints
Use pre-trained EnlightenGAN model to generate enhanced noisy images. Put them in ./test_imgs

python test.py

Realistic Low-light Mobile Photography (RLMP)

RLMP is a new benchmark of the real-world image low-light enhancement task. Images are captured by different types of smartphones. Compared with previous low-light datasets, images in RLMP typically display much more noticeable ISO noise, which complements the existing benchmarks and significantly challenges current enhancement methods. You can get RLMP from this link.

Citation

if you find this repo is helpful, please cite

@article{chen2021cerl,
  title={CERL: A Unified Optimization Framework for Light Enhancement with Realistic Noise},
  author={Zeyuan Chen and Yifan Jiang and Dong Liu and Zhangyang Wang},
  journal={arXiv preprint arXiv:2108.00478},
  year={2021}
}