GitHub - redrock303/ADF-LLIE (original) (raw)

Unveiling Advanced Frequency Disentanglement Paradigm for Low-Light Image Enhancement (ECCV 2024)

Kun Zhou, Xinyu Lin, Wenbo Li, Xiaogang Xu, Yuanhao Cai, Zhonghang Liu, Xiaoguang Han, Jiangbo Lu

News, We have release the training and inference scripts for LoLv2 and our paper is available on [ARXIV].

Visual results, along with uploaded checkpoints, can be accessed at [google drive].

We have explored ten SOTA LLIE baselines:

CNN: UNet (written by Chatgpt), [MIR-Net ECCV 2020],[MIR-Net-V2 T-PAMI 2022]

Transformer: [SNR CVPR 2022], [Retinexformer ICCV 2023 ], [Restormer CVPR 2020]

Mamba: [RetinexMamba], [MambaIR ECCV 2024],

Diffusion-based: [Diff-L Siggraph 2023] and Flow-based: [LLFlow AAAI 2022]).

Our proposed AFD-LLIE consistently and significantly enhances their performance.

Let's make UNet (by ChatGPT) great again!

Even a basic UNet can achieve impressive results and outperform Retinexformer with 24.25dB on LOL-v2.

Citation

If our work is useful for your research, please consider citing:

@inproceedings{zhou2024,
title={Unveiling Advanced Frequency Disentanglement Paradigm for Low-Light Image Enhancement},
author={Kun Zhou, Xinyu Lin, Wenbo Li, Xiaogang Xu, Yuanhao Cai, Zhonghang Liu, Xiaoguang Han, and Jiangbo Lu},
booktitle={Proceedings of the European Conference on Computer Vision},
year={2024}
}