Skip to content

Official code for "Contrastive Semi-supervised Learning for Underwater Image Restoration via Reliable Bank" (2023 CVPR)

Notifications You must be signed in to change notification settings

Huang-ShiRui/Semi-UIR

Repository files navigation

Contrastive Semi-supervised Learning for Underwater Image Restoration via Reliable Bank (CVPR 2023)

Shirui Huang*, Keyan Wang*+, Huan Liu, Jun Chen, Yunsong Li

*Equal Contributions +Corresponding Author

Xidian University, McMaster University

Introduction

This is the official repository for our recent paper, "Contrastive Semi-supervised Learning for Underwater Image Restoration via Reliable Bank link", where more implementation details are presented.

Abstract

Despite the remarkable achievement of recent underwater image restoration techniques, the lack of labeled data has become a major hurdle for further progress. In this work, we propose a mean-teacher based Semi-supervised Underwater Image Restoration (Semi-UIR) framework to incorporate the unlabeled data into network training. However, the naive mean-teacher method suffers from two main problems: (1) The consistency loss used in training might become ineffective when the teacher's prediction is wrong. (2) Using L1 distance may cause the network to overfit wrong labels, resulting in confirmation bias. To address the above problems, we first introduce a reliable bank to store the ``best-ever" outputs as pseudo ground truth. To assess the quality of outputs, we conduct an empirical analysis based on the monotonicity property to select the most trustworthy NR-IQA method. Besides, in view of the confirmation bias problem, we incorporate contrastive regularization to prevent the overfitting on wrong labels. Experimental results on both full-reference and non-reference underwater benchmarks demonstrate that our algorithm has obvious improvement over SOTA methods quantitatively and qualitatively.

Figure 1. An overview of our approach.

Dependencies

  • Ubuntu==18.04
  • Pytorch==1.8.1
  • CUDA==11.1

Other dependencies are listed in requirements.txt

Data Preparation

Run data_split.py to randomly split your paired datasets into training, validation and testing set.

Run estimate_illumination.py to get illumination map of the corresponding image.

Finally, the structure of data are aligned as follows:

data
├── labeled
│   ├── input
│   └── GT
│   └── LA
├── unlabeled
│   ├── input
│   └── LA
│   └── candidate
└── val
    ├── input
    └── GT
    └── LA
└── test
    ├── benchmarkA
        ├── input
        └── LA

You can download the training set and test sets from benchmarks UIEB, EUVP, UWCNN, Sea-thru, RUIE.

Test

Put your test benchmark under data/test folder, run estimate_illumination.py to get its illumination map.

Run test.py and you can find results from folder result.

Train

To train the framework, run create_candiate.py to initialize reliable bank. Hyper-parameters can be modified in trainer.py.

Run train.py to start training.

Citation

If you use the code in this repo for your work, please cite the following bib entries:

@inproceedings{huang2023contrastive,
  title={Contrastive Semi-supervised Learning for Underwater Image Restoration via Reliable Bank},
  author={Huang, Shirui and Wang, Keyan and Liu, Huan and Chen, Jun and Li, Yunsong},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={18145--18155},
  year={2023}
}

Acknowledgement

The training code architecture is based on the PS-MT and DMT-Net and thanks for their work. We also thank for the following repositories: IQA-Pytorch, UWNR, MIRNetv2, 2022-CVPR-AirNet, SPSR, Non-Local-Sparse-Attention, AFF, AECR-Net, UIEB, EUVP, UWCNN, Sea-thru, RUIE, MMLE, PWRNet, Ucolor, CWR, FUnIE-GAN

About

Official code for "Contrastive Semi-supervised Learning for Underwater Image Restoration via Reliable Bank" (2023 CVPR)

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages