Toward Better Than Pseudo-Reference in Underwater Image Enhancement

This repository is the official PyTorch implementation.

Running Environment

1. python 3.8
2. pip install -r requirements.txt

PSPL

Test

1. rename directory pre-trained to checkpoints
2. rename directory pre-datasets to datasets
3. run command to infer(test)

    python test.py --dataroot ./datasets/unsuped --name unsuped --model unsupedmix --epoch PSPL_HRNet
   

PSPL on other UIE methods

Test

1. rename directory pre-trained to checkpoints
2. rename directory pre-datasets to datasets
3. run command to infer(test)

   python test.py --dataroot ./datasets/unsuped --name unsuped --model UWCNN --epoch PSPL_UWCNN
   

4. testing NU2Net and SemiUIR (baidu cloud disk, 提取码: gkf2) can follow the way (need add corresponding files) as UWCNN. Or you can put downloaded pre-trained model files to their's original source code projects(NU2Net, Semi-UIR) to inference.

Citation

Y. Liu, Q. Jiang, X. Li, T. Luo and W. Ren, "Toward Better Than Pseudo-Reference in Underwater Image Enhancement," 
in IEEE Transactions on Image Processing, vol. 34, pp. 6168-6179, 2025.

Acknowledgements

• https://github.com/lewis081/CCL-Net

• https://github.com/RQ-Wu/UnderwaterRanker

• https://github.com/justwj/CLUIE-Net

• https://github.com/uf-robopi/UDepth

• https://github.com/Huang-ShiRui/Semi-UIR

• https://github.com/xahidbuffon/FUnIE-GAN

• https://github.com/CXH-Research/Pytorch-UW-IQA

• https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix