Gaussian Denoising on BSD68 (Specific Noise Levels)
28.32PSNR (sigma=50)CEA-IR
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CEA-IRParams.=22M, Model Capacity=Heavy2026.05 | 28.32 | 34.22 | 31.58 | |
| CEA-IR-SParams.=10M, Model Capacity=Light2026.05 | 28.27 | 34.16 | 31.51 | |
| Res-HAIRParams.=29M, Model Capacity=Heavy2026.05 | 28.24 | 34.16 | 31.51 | |
| VLU-NetParams.=35M, Model Capacity=Heavy2026.05 | 28.23 | 34.13 | 31.48 | |
| LoRA-IRParams.=–, Model Capacity=Heavy2026.05 | 28.18 | 34.06 | 31.42 | |
| MoCE-IRParams.=25M, Model Capacity=Heavy2026.05 | 28.18 | 34.11 | 31.45 | |
| UniProcessorParams.=1002M, Model Capacity=Heavy2026.05 | 28.17 | 34.08 | 31.42 | |
| MoCE-IR-SParams.=11M, Model Capacity=Light2026.05 | 28.16 | 34.08 | 31.42 | |
| PromptIR+ArtParams.=33M, Model Capacity=Heavy2026.05 | 28.14 | 34.06 | 31.42 | |
| GridformerParams.=34M, Model Capacity=Heavy2026.05 | 28.11 | 33.93 | 31.37 | |
| PromptIRParams.=36M, Model Capacity=Heavy2026.05 | 28.06 | 33.98 | 31.31 | |
| AirNetParams.=9M, Model Capacity=Light2026.05 | 28 | 33.92 | 31.26 | |
| MPRNetParams.=16M, Model Capacity=Heavy2026.05 | 27.56 | 33.54 | 30.89 | |
| Restormer+Ours-cTraining protocol=Single model for all noise levels, Pre-trained CLIP model=true2024.03 | 26.91 | 32.11 | 29.78 | |
| GRL-B+Ours-cTraining protocol=Separate model per noise level, Pre-trained CLIP model=true2024.03 | 26.91 | 32.29 | 29.76 | |
| DLParams.=2M, Model Capacity=Light2026.05 | 26.9 | 33.05 | 30.41 | |
| Restormer+Ours-cTraining protocol=Separate model per noise level, Pre-trained CLIP model=true2024.03 | 26.86 | 32.24 | 29.81 | |
| DRUNet+Ours-cTraining protocol=Single model for all noise levels, Pre-trained CLIP model=true2024.03 | 26.84 | 32.2 | 29.73 | |
| KBNet_sMACs=69G2023.03 | 26.65 | 31.98 | 29.54 | |
| RestormerMACs=141G2023.03 | 26.62 | 31.96 | 29.52 | |
| RestormerTraining protocol=Single model for all noise levels2024.03 | 26.62 | 31.95 | 29.51 | |
| RestormerTraining protocol=Separate model per noise level2024.03 | 26.62 | 31.96 | 29.52 | |
| GRL-BTraining protocol=Separate model per noise level2024.03 | 26.6 | 32 | 29.54 | |
| DRUNetMACs=144G2023.03 | 26.59 | 31.91 | 29.48 | |
| DRUNetTraining protocol=Single model for all noise levels2024.03 | 26.59 | 31.91 | 29.48 | |
| SwinIRMACs=759G2023.03 | 26.58 | 31.97 | 29.5 | |
| DeamNetMACs=146G2023.03 | 26.54 | 31.91 | 29.44 | |
| MWCNN2023.03 | 26.53 | 31.86 | 29.41 | |
| DAGLMACs=256G2023.03 | 26.51 | 31.93 | 29.46 | |
| FOCNet2023.03 | 26.5 | 31.83 | 29.38 | |
| RNANMACs=496G2023.03 | 26.48 | — | — | |
| NLRN2023.03 | 26.47 | 31.88 | 29.41 | |
| FDGANParams.=–, Model Capacity=Light2026.05 | 26.43 | 30.25 | 28.81 | |
| BRDNetParams.=–, Model Capacity=Light2026.05 | 26.34 | 32.26 | 29.76 | |
| FFDNet2023.03 | 26.29 | 31.63 | 29.19 | |
| DnCNNMACs=37G2023.03 | 26.23 | 31.62 | 29.16 | |
| IRCNN2023.03 | 26.19 | 31.63 | 29.15 | |
| DA-CLIPParams.=125M, Model Capacity=Heavy2026.05 | 22.29 | 30.02 | 24.86 | |
| LPNetParams.=–, Model Capacity=Light2026.05 | 21.26 | 26.47 | 24.77 |