Skip to content

[CVPR 2023] Curricular Contrastive Regularization for Physics-aware Single Image Dehazing

Notifications You must be signed in to change notification settings

YuZheng9/C2PNet

Repository files navigation

[CVPR 2023] Curricular Contrastive Regularization for Physics-aware Single Image Dehazing

PWC PWC

This is the official PyTorch codes for the paper:

Curricular Contrastive Regularization for Physics-aware Single Image Dehazing
Yu Zheng, Jiahui Zhan, Shengfeng He, Junyu Dong, Yong Du* ( * indicates corresponding author)
IEEE/CVF Conference on Computer Vision and Pattern Recognition

Network Architecture

Architecture

News

  • Apr 20, 2023: We release training code.

Getting started

Install

We test the code on PyTorch 1.10.1 + CUDA 11.4 .

  1. Create a new conda environment
conda create -n c2pnet python=3.7.11
conda activate c2pnet
  1. Install dependencies
pip install torch==1.10.0+cu111 torchvision==0.11.0+cu111 torchaudio==0.10.0 -f https://download.pytorch.org/whl/torch_stable.html
pip install -r requirements.txt

Training and Evaluation

Prepare dataset for evaluation

You can download the pretrained models and datasets on Google Drive(pretrained models and testset).

The final file path will be arranged as (please check it carefully):

|-trained_models
     |- ITS.pkl
     |- OTS.pkl
     └─ ... (model name)
|-data
     |-SOTS
        |- indoor
           |- hazy
              |- 1400_1.png 
              |- 1401_1.png 
              └─ ... (image name)
           |- clear
              |- 1400.png 
              |- 1401.png
              └─ ... (image name)
        |- outdoor
           |- hazy
              |- 0001_0.8_0.2.jpg 
              |- 0002_0.8_0.08.jpg
              └─ ... (image name)
           |- clear
              |- 0001.png 
              |- 0002.png
              └─ ... (image name)

Prepare dataset for train

Since our training dataset contains numerous negative samples, which has led to an oversized dataset, we have only uploaded the ITS training dataset on BaiduNetdisk. However, users can create additional datasets using the create_lmdb.py, where they can define the number of negative samples and the types of existing dehazers used.

The final file path will be arranged as (please check it carefully):

data
   |-ITS
      |- ITS.lmdb
   |- ...(dataset name)

Evaluation

Test C2PNet on SOTS-indoor dataset

python dehaze.py -d indoor

Test C2PNet on SOTS-outdoor dataset

python dehaze.py -d outdoor

See python dehaze.py -h for the list of optional arguments

Train

Train network on ITS dataset

python main.py --crop --crop_size=240 --blocks=19 --gps=3 --bs=2 --lr=0.0001 --trainset='its_train' --testset='its_test' --steps=1000000 --eval_step=5000 --clcrloss --clip

Train network on OTS dataset

python main.py --crop --crop_size=240 --blocks=19 --gps=3 --bs=2 --lr=0.0001 --trainset='ots_train' --testset='ots_test' --steps=1500000 --eval_step=5000 --clcrloss --clip

Citation

If you find our work useful for your research, please cite us:

@inproceedings{zheng2023curricular,
  title={Curricular Contrastive Regularization for Physics-aware Single Image Dehazing},
  author={Zheng, Yu and Zhan, Jiahui and He, Shengfeng and Dong, Junyu and Du, Yong},
  booktitle={IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2023}
}

About

[CVPR 2023] Curricular Contrastive Regularization for Physics-aware Single Image Dehazing

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages