Personal implementation of denoisers by PyTorch.
Pretrained model and dataset: [OneDrive]
- DnCNN: [Code][Paper]
- UNet: [Code][Paper]
- REDNet: [Code][Paper]
- MWCNN: [Code][Paper]
- RDN+: [Code][Paper]
- N3Net: [Code][Paper]
- CBDNet: [Code][Paper]
| Method | PSNR | SSIM | FLOPs / G | Params / M | FPS |
|---|---|---|---|---|---|
| DnCNN | 37.81 | 0.898 | 146.2 | 0.6 | 24.2 |
| UNet | 37.79 | 0.901 | 239.2 | 31.0 | 16.6 |
| REDNet | 37.95 | 0.900 | 184.9 | 0.7 | 14.2 |
| MWCNN | 38.17 | 0.901 | 259.0 | 16.2 | 11.1 |
| RDN+ | 38.19 | 0.903 | 1687.8 | 5.5 | 1.4 |
| N3Net | 38.17 | 0.902 | N/A | 0.5 | N/A |
| CBDNet | 38.15 | 0.900 | 172.0 | 4.3 | 19.7 |
| Method | PSNR | SSIM | FLOPs / G | Params / M | FPS |
|---|---|---|---|---|---|
| DuRB-P | 31.77 | 0.691 | 220.7 | 0.8 | 7.4 |
| DuRB-P-ReLu | 33.27 | 0.749 | 220.9 | 0.8 | 7.0 |
| MemNet-3 | 33.70 | 0.775 | 193.8 | 0.7 | 9.0 |
| MemNet-6 | 32.56 | 0.718 | 768.8 | 2.9 | 2.2 |
All the models have been move to the /@addtional/ folder.
| Method | PSNR | SSIM | FLOPs / G | Params / M | FPS |
|---|---|---|---|---|---|
| RDN-4 | 38.02 | 0.901 | 334.1 | 1.1 | 7.3 |
| RDN+-4 | 38.02 | 0.901 | 355.5 | 1.2 | 6.9 |
| N3Net-2-17 | 38.05 | 0.900 | 294.5 | 1.1 | 0.1 |
| CARN | 37.73 | 0.898 | 272.0 | 0.8 | 7.6 |
| CARNM | 37.71 | 0.898 | 149.5 | 0.4 | 6.6 |
| ResNet | 37.87 | 0.899 | 271.4 | 1.0 | 11.4 |
| UNet-S | 36.62 | 0.901 | 60.0 | 7.8 | 46.6 |
| SSnbt-UNet-S | 35.58 | 0.884 | 35.8 | 9.7 | 37.4 |
| SE-UNet-S | 36.50 | 0.900 | 60.0 | 7.8 | 16.6 |
| GC-UNet-S | 36.38 | 0.899 | 60.1 | 8.6 | 30.8 |
| Mobile-UNet-S | 29.33 | 0.588 | 3.5 | 0.4 | 54.9 |
| SN-UNet-S | 35.22 | 0.863 | 46.3 | 3.5 | 38.9 |