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.

1. python=3.9.12
2. torch==1.10.2
3. torchvision=0.11.3
4. 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}
}