Low-light Image Enhancement via CLIP-Fourier Guided Wavelet Diffusion [Paper].
In this repo, we propose a novel and robust low-light image enhancement method, named CFWD. Extensive experiments on publicly available real-world benchmarks show that our method outperforms existing SOTA methods quantitatively and visually, maximizing image restoration similar to normal images. For more details, please refer to our paper link
Clone Repo
git clone https://github.com/He-Jinhong/CFWD.git
cd CFWD
Create Conda Environment and Install Dependencies:
pip install -r requirements.txt
You can downlaod our pre-training prompts and pre-training models from [Google Drive]
Before performing the following steps, please download our pretrained model first.
You need to modify test.py and datasets.py
according to your environment and then
python test.py
Note that you will need to download our pre-training prompts The directory structure will be arranged as:
pretrain_models
|-prompt_pretrain
|- pre_prompt_pair.pth
|-pretrain_model.pth.tar
Then, You need to modify train_clip.py and datasets.py
slightly for your environment, and then
python train_clip.py
@article{xue2024low,
title={Low-light Image Enhancement via CLIP-Fourier Guided Wavelet Diffusion},
author={Xue, Minglong and He, Jinhong and Wang, Wenhai and Zhou, Mingliang},
journal={arXiv preprint arXiv:2401.03788},
year={2024}
}
Our underlying network comes from previous works: WCDM. We thanks the authors for their contributions.