USLN: A statistically guided lightweight network for underwater image enhancement via dual-statistic white balance and multi-color space stretch

Extensive experiments show that USLN significantly reduces the required network capacity (over 98%) and achieves state-of-the-art performance.

python 3.9, pytorch 1.10.1

if you want to train the model:
1, put your datasets into corresponding folders ("images_train", "labels_train", "images_val", "labels_val")
2, run train.py
3, the checkpoints will be saved in "logs"

if you want to test the model:
1, put your datasets into "images_test"
2, run test.py (load model checkpoints from "logs" first)
3, the result will be saved in "pred"

@misc{USLN,
  doi = {10.48550/ARXIV.2209.02221},
  url = {https://arxiv.org/abs/2209.02221}, 
  author = {Xiao, Ziyuan and Han, Yina and Rahardja, Susanto and Ma, Yuanliang},
  keywords = {Computer Vision and Pattern Recognition (cs.CV), FOS: Computer and information sciences, FOS:   Computer and information sciences},
  title = {USLN: A statistically guided lightweight network for underwater image enhancement via dual-statistic white balance and multi-color space stretch},
  publisher = {arXiv},
  year = {2022}, 
  copyright = {arXiv.org perpetual, non-exclusive license}
}

The code is made available for academic research purpose only. This project is open sourced under MIT license.

If you have any questions, please contact Ziyuan Xiao at [email protected].