Physical Model Guided Deep Image Deraining (ICME'2020)
@inproceedings{zhu2020physical,
title={Physical model guided deep image deraining},
author={Zhu, Honghe and Wang, Cong and Zhang, Yajie and Su, Zhixun and Zhao, Guohui},
booktitle={2020 IEEE International Conference on Multimedia and Expo (ICME)},
pages={1--6},
year={2020},
organization={IEEE}
}
Quantitative Result
The metrics are PSNR/SSIM
. Both are evaluated on RGB channels.
NOTE:
- Due to limited computation resource:
- batch size is reduced from 32 to 24
- For Rain1200 and Rain1400, training epochs is reduced from 2000 to 200
Method | Rain200L | Rain200H | Rain800 | Rain1200 | Rain1400 |
---|---|---|---|---|---|
pmg_c64d5s3 | 37.84/0.983 | 28.79/0.897 | 27.37/0.859 | 33.12/0.922 | 31.38/0.919 |
Pretrained models can be downloaded from here
Network Complexity
Input shape | Flops | Params |
---|---|---|
(3, 256, 256) | 98.38GFlops | 2.77M |