/SCANet

[CVPRW 2023] SCANet: Self-Paced Semi-Curricular Attention Network for Non-Homogeneous Image Dehazing

Primary LanguagePythonMIT LicenseMIT

[CVPRW 2023] SCANet: Self-Paced Semi-Curricular Attention Network for Non-Homogeneous Image Dehazing

Paper ArXiv Poster Video

SCANet: Self-Paced Semi-Curricular Attention Network for Non-Homogeneous Image Dehazing
Yu Guo, Yuan Gao, Ryan Wen Liu*, Yuxu Lu, Jingxiang Qu, Shengfeng He, Wenqi Ren
(* Corresponding Author)
IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops

Abstract: The presence of non-homogeneous haze can cause scene blurring, color distortion, low contrast, and other degradations that obscure texture details. Existing homogeneous dehazing methods struggle to handle the non-uniform distribution of haze in a robust manner. The crucial challenge of non-homogeneous dehazing is to effectively extract the non-uniform distribution features and reconstruct the details of hazy areas with high quality. In this paper, we propose a novel self-paced semi-curricular attention network, called SCANet, for non-homogeneous image dehazing that focuses on enhancing haze-occluded regions. Our approach consists of an attention generator network and a scene reconstruction network. We use the luminance differences of images to restrict the attention map and introduce a self-paced semi-curricular learning strategy to reduce learning ambiguity in the early stages of training. Extensive quantitative and qualitative experiments demonstrate that our SCANet outperforms many state-of-the-art methods.

Requirement

  • Python 3.7
  • Pytorch 1.9.1

Network Architecture

fig_scanet

Train

  • Place the training and test image pairs in the data folder.
  • Run data/makedataset.py to generate the NH-Haze20-21-23.h5 file.
  • Run train.py to start training.

Test

  • Place the pre-training weight in the checkpoint folder.
  • Place test hazy images in the input folder.
  • Modify the weight name in the test.py.
parser.add_argument("--model_name", type=str, default='Gmodel_40', help='model name')
  • Run test.py
  • The results are saved in output folder.

Pre-training Weight Download

  • The weight40 for the NTIRE2023 val/test datasets, i.e., the weight used in the NTIRE2023 challenge.
  • The weight105 for the NTIRE2020/2021/2023 datasets.
  • The weight120 for the NTIRE2020/2021/2023 datasets (Add the 15 tested images as the training dataset).

Citation

@inproceedings{guo2023scanet,
  title={SCANet: Self-Paced Semi-Curricular Attention Network for Non-Homogeneous Image Dehazing},
  author={Guo, Yu and Gao, Yuan and Liu, Wen and Lu, Yuxu and Qu, Jingxiang and He, Shengfeng and Ren, Wenqi},
  booktitle={IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops},
  pages={1884--1893},
  year={2023}
}

If you have any questions, please get in touch with me (guoyu65896@gmail.com).

Visitor count