Demo and Implementation of P2CNet and two-stage UIE (An free lunch for most existing UIE models without any tuning):
Our approach can successfully break through the generalization bottleneck of current deep UIE models without re-training:
Our approach can also benefit DCP-based methods by effective dark channel extraction:
Our work was accepted by IEEE Transactions on Circuits and Systems for Video Technology 2023, and can be early accessed via manuscript.
Our P2CNet estimates the probabilistic distribution of colors by multi-scale volumetric fusion of texture and color features.
The P2CNet network and model weights are provided in the models and ckpt files respectively.
For the enhancement network and the model weights of CLUIE, please refer to the CLUIE-Net
We provide the enhancement demo of using P2CNet and CLUIE-Net in demo.ipynb and test.py, which are easily to be adapted with most existing enhancement models or algorthms.
If you are interested in this work, please cite the following work:
@ARTICLE{10220126,
author={Rao, Yuan and Liu, Wenjie and Li, Kunqian and Fan, Hao and Wang, Sen and Dong, Junyu},
journal={IEEE Transactions on Circuits and Systems for Video Technology},
title={Deep Color Compensation for Generalized Underwater Image Enhancement},
year={2024},
volume={34},
number={4},
pages={2577-2590},
doi={10.1109/TCSVT.2023.3305777}}