Skip to content

Troivyn/HazeCLIP

Repository files navigation

HazeCLIP: Towards Language Guided Real-World Image Dehazing

License

This repository contains the implementation of the paper "HazeCLIP: Towards Language Guided Real-World Image Dehazing".

We present HazeCLIP, a language-guided adaptation framework designed to enhance the real-world performance of pre-trained dehazing networks.

teaser teaser

🛠️ Setup

Set up conda environment via

conda create -n HazeCLIP python=3.9
conda activate HazeCLIP
pip install -r requirements.txt

🚀 Usage

Please modify the corresponding yaml configuration file before running the command.

🏋️ Inference

Download checkpoint from Baidu Yun (code: haze) and put it in ./weights/ folder.

python inference.py --config configs/inference.yaml

🚀 Training

Pre-training

Download synthetic data from RIDCP and put it under ./data/ folder.

python pretrain.py --config configs/pretrain.yaml

Fine-tuning

Download fine-tuning dataset from Baidu Yun (code: haze) and put it under ./data/ folder.

python finetune.py --config configs/finetune.yaml

🎓 Citation

If you find our work helpful, please consider cite our work as

@misc{wang2024hazecliplanguageguidedrealworld,
      title={HazeCLIP: Towards Language Guided Real-World Image Dehazing}, 
      author={Ruiyi Wang and Wenhao Li and Xiaohong Liu and Chunyi Li and Zicheng Zhang and Xiongkuo Min and Guangtao Zhai},
      year={2024},
      eprint={2407.13719},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2407.13719}, 
}

🎫 Acknowledgement

Parts of the codes are adopted from RIDCP, CLIP Surgery and CLIP-LIT. Thanks for their work!

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages