Detail-recovery Image Deraining via Context Aggregation Networks (CVPR'2020)
@inproceedings{deng2020detail,
title={Detail-recovery image deraining via context aggregation networks},
author={Deng, Sen and Wei, Mingqiang and Wang, Jun and Feng, Yidan and Liang, Luming and Xie, Haoran and Wang, Fu Lee and Wang, Meng},
booktitle={Proceedings of the IEEE/CVF conference on computer vision and pattern recognition},
pages={14560--14569},
year={2020}
}
Quantitative Result
The metrics are PSNR/SSIM. Both are evaluated on RGB channels.
Note:
- The test result is much lower to the one reported in the paper. A similar issue is reported in Dengsgithub/DRD-Net#8.
- A major difference between the official code and the code in this repo is that: the official code uses sum of rain image
$O$ and rain estimation$\hat{R}$ as the input of background recovery subnet, while this repo uses a single rain image$O$ as the input, which is consistent with the paper. Experiments show that the latter one has a better performance.
| Method | Rain200L | Rain200H | Rain800 | Rain1200 | Rain1400 |
|---|---|---|---|---|---|
| DRD-Net | 32.96/0.963 | 23.01/0.726 | 23.33/0.788 | 26.25/0.814 | 26.66/0.841 |
Pretrained models can be downloaded from here
Network Complexity
| Input shape | Flops | Params |
|---|---|---|
| (3, 256, 256) | 522.54GFlops | 7.98M |