Image Denoising on CBSD68 sigma=50 (test)
28.56PSNRSwinIR
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SwinIR#param=11.9M2023.12 | 28.56 | — | |
| PromptIRprotocol=one-by-one IR2025.03 | 28.49 | — | |
| VLU-Netprotocol=one-by-one IR2025.03 | 28.46 | — | |
| Full-ft#param=119M2023.12 | 28.39 | — | |
| LORA#param=995K2023.12 | 28.38 | — | |
| FacT#param=537K2023.12 | 28.38 | — | |
| AdaptIR#param=515K2023.12 | 28.38 | — | |
| VPT#param=884K2023.12 | 28.37 | — | |
| Adapter#param=691K2023.12 | 28.37 | — | |
| AdaptFor.#param=677K2023.12 | 28.37 | — | |
| Pretrain2023.12 | 28.36 | — | |
| RDN#param=15.6M2023.12 | 28.31 | — | |
| InstructIRprotocol=one-by-one IR2025.03 | 28.3 | — | |
| AirNetprotocol=one-by-one IR2025.03 | 28.23 | — | |
| IDRprotocol=one-by-one IR2025.03 | 28.14 | — | |
| Restormerprotocol=one-by-one IR2025.03 | 28.11 | — | |
| FFDNet#param=0.8M2023.12 | 27.96 | — | |
| FFDNetprotocol=one-by-one IR2025.03 | 27.96 | — | |
| DnCNNprotocol=one-by-one IR2025.03 | 27.95 | — | |
| DGUNetprotocol=one-by-one IR2025.03 | 27.92 | — | |
| SSF#param=373K2023.12 | 27.64 | — | |
| MeDTraining Schema=Random variance (sigma in [5, 50]), Training Paradigm=Invariant Feature2023.09 | 27.48 | 0.753 | |
| N2CTraining Schema=Random variance (sigma in [5, 50]), Training Paradigm=Noisy/Clean2023.09 | 27.41 | 0.7361 | |
| N2NTraining Schema=Random variance (sigma in [5, 50]), Training Paradigm=N2N2023.09 | 27.15 | 0.7219 | |
| DBD4Training Schema=Random variance (sigma in [5, 50]), Training Paradigm=Noisy/Clean2023.09 | 27.13 | 0.729 | |
| SwinIRprotocol=one-by-one IR2025.03 | 27.13 | — | |
| N2STraining Schema=Random variance (sigma in [5, 50]), Training Paradigm=Noisy/Noisy2023.09 | 27 | 0.7114 | |
| R2RTraining Schema=Random variance (sigma in [5, 50]), Training Paradigm=Noisy/Noisy2023.09 | 26.92 | 0.6911 | |
| CLIPDenoisingTraining=Single noise level2024.03 | 26.69 | 0.731 | |
| DILTraining=Multiple noise levels required2024.03 | 26.43 | 0.717 | |
| MeDTraining Schema=Fixed variance (sigma=25), Training Paradigm=Invariant Feature2023.09 | 25.67 | 0.6026 | |
| LIRTraining Schema=Random variance (sigma in [5, 50]), Training Paradigm=Invariant Feature2023.09 | 25.13 | 0.6191 | |
| DBD4Training Schema=Fixed variance (sigma=25), Training Paradigm=Noisy/Clean2023.09 | 25.12 | 0.5583 | |
| N2CTraining Schema=Fixed variance (sigma=25), Training Paradigm=Noisy/Clean2023.09 | 24.76 | 0.5519 | |
| N2NTraining Schema=Fixed variance (sigma=25), Training Paradigm=N2N2023.09 | 24.59 | 0.5385 | |
| R2RTraining Schema=Fixed variance (sigma=25), Training Paradigm=Noisy/Noisy2023.09 | 24.02 | 0.5133 | |
| N2STraining Schema=Fixed variance (sigma=25), Training Paradigm=Noisy/Noisy2023.09 | 22.13 | 0.3928 | |
| LIRTraining Schema=Fixed variance (sigma=25), Training Paradigm=Invariant Feature2023.09 | 21.97 | 0.3578 | |
| HATTraining=Multiple noise levels required2024.03 | 20.95 | 0.441 | |
| MaskDenoising2024.03 | 20.68 | 0.432 | |
| Restormer2024.03 | 19.92 | 0.365 | |
| DnCNN2024.03 | 19.84 | 0.363 |