This is an official pytorch implement of SCEIR:
"Atmospheric Scattering Model Induced Statistical Characteristics Estimation for Underwater Image Restoration" Paper
We implement all the experiments in the following environment.
- python=3.9.12
- torch==1.10.2
- torchvision=0.11.3
- PIL, tqdm
For stable running, torch>=1.8.1 and torchvision>=0.4.0 are recommended.
Put the test images into "./test_input" or change the option "--test_input" to your image dir
python test.py --save_extra --gpu YOUR_DEVICE --test_input YOUR_IMAGE_DIR
Put the pair training images into "./dataset", and changes the option "--train_raw", "--train_ref", "--eval_raw" and "--eval_ref" to the relevant path.
python train.py --gpu YOUR_DEVICE
If you find SCEIR is useful in your research, please cite our paper:
@article{gao2023atmospheric,
title={Atmospheric Scattering Model Induced Statistical Characteristics Estimation for Underwater Image Restoration},
author={Gao, Shuaibo and Wu, Wenhui and Li, Hua and Zhu, Linwei and Wang, Xu},
journal={IEEE Signal Processing Letters},
year={2023},
publisher={IEEE}
}