Image Denoising on SIDD Benchmark
40.02PSNRRestormer
Evaluation results
| Method | Links | ||
|---|---|---|---|
| RestormerTraining Supervision=Supervised (Real pairs)2023.03 | 40.02 | 96 | |
| RestormerLearning Paradigm=Supervised2026.04 | 40.02 | 96 | |
| DeamNetType=Supervised, Train Data=Paired Noisy-Clean, #Param=2.23M2025.12 | 39.58 | 91.18 | |
| VDNetType=Supervised, Train Data=Paired Noisy-Clean, #Param=7.82M2025.12 | 39.49 | 91.17 | |
| VDNTraining Supervision=Supervised (Real pairs)2023.03 | 39.26 | 95.5 | |
| VDNLearning Paradigm=Supervised2026.04 | 39.26 | 95.5 | |
| Baseline, N2CTraining Supervision=Supervised (Real pairs)2023.03 | 38.92 | 95.3 | |
| TM-BSN (D)Learning Paradigm=Self-Supervised, Evaluation Platform=Kaggle, Model Variant=D2026.04 | 38.31 | 90 | |
| APR (RD)Learning Paradigm=Self-Supervised, Evaluation Platform=Kaggle, Model Variant=(RD)2026.04 | 38.26 | 89.5 | |
| AT-BSN (D)Learning Paradigm=Self-Supervised, Evaluation Platform=Kaggle, Model Variant=(D)2026.04 | 38.14 | 89.1 | |
| TBSNLearning Paradigm=Self-Supervised, Evaluation Platform=Kaggle2026.04 | 38.02 | 88.6 | |
| PUCALearning Paradigm=Self-Supervised, Evaluation Platform=Kaggle2026.04 | 37.93 | 88.2 | |
| SASLLearning Paradigm=Self-Supervised, Evaluation Platform=Kaggle2026.04 | 37.85 | 87.9 | |
| TM-BSN (s=4)Learning Paradigm=Self-Supervised, Evaluation Platform=Kaggle, Model Variant=s=42026.04 | 37.71 | 88.1 | |
| SelfFormerCategory=Self-supervised2025.09 | 37.69 | 93.7 | |
| LG-BPNLearning Paradigm=Self-Supervised, Evaluation Platform=Kaggle2026.04 | 37.67 | 88.7 | |
| QWD-GANCategory=Self-supervised2025.09 | 37.65 | 94.2 | |
| SDAPLearning Paradigm=Self-Supervised, Evaluation Platform=Kaggle2026.04 | 37.64 | 88.2 | |
| DnCNNTraining Supervision=Supervised (Real pairs)2023.03 | 37.61 | 94.1 | |
| DnCNNLearning Paradigm=Supervised2026.04 | 37.61 | 94.1 | |
| PUCACategory=Self-supervised2025.09 | 37.54 | 93.6 | |
| NSP(TBSN)Type=Self-Supervised, Train Data=Only Noisy, #Param=12.77M2025.12 | 37.46 | 87.44 | |
| NSP(DBSN)Type=Self-Supervised, Train Data=Only Noisy, #Param=3.75M2025.12 | 37.42 | 87.48 | |
| BNN (LAN)Training Supervision=Self-Supervised2023.03 | 37.41 | 93.4 | |
| TBSNType=Self-Supervised, Train Data=Only Noisy, #Param=12.74M2025.12 | 37.08 | 85.19 | |
| AP-BSN+R3Training Supervision=Self-Supervised2023.03 | 36.91 | 93.1 | |
| C-BSNLearning Paradigm=Self-Supervised, Evaluation Platform=Kaggle2026.04 | 36.85 | 88.2 | |
| SDAPType=Self-Supervised, Train Data=Only Noisy, #Param=3.66M2025.12 | 36.8 | 85.29 | |
| APBSNCategory=Self-supervised2025.09 | 36.69 | 87.4 | |
| DnCNNTraining Noise Model=Real Noise2022.06 | 36.51 | 92.2 | |
| AP-BSN + R3Learning Paradigm=Self-Supervised, Evaluation Platform=Kaggle2026.04 | 36.51 | 87.2 | |
| AP-BSNType=Self-Supervised, Train Data=Only Noisy, #Param=3.66M2025.12 | 36.09 | 83.1 | |
| C2NType=Pseudo-Supervised, Train Data=Unpaired Noisy-Clean, #Param=217.26M2025.12 | 36.06 | 88.25 | |
| CLIPDenoisingCategory=Supervised2024.12 | 35.82 | 85.9 | |
| Positive2NegativeCategory=Self-Supervised (Single Image Based Methods)2024.12 | 35.8 | 87.5 | |
| R2RType=Pseudo-Supervised, Train Data=Paired Noisy-Noisy, #Param=0.56M2025.12 | 35.5 | 85.5 | |
| LUD-VAECategory=Self-Supervised (Noisy Dataset Based Methods)2024.12 | 35.49 | 88.3 | |
| ScoreDVICategory=Self-Supervised (Single Image Based Methods)2024.12 | 35.39 | 85.9 | |
| C2NTraining Supervision=Unpaired2023.03 | 35.35 | 93.7 | |
| C2NLearning Paradigm=Unpaired2026.04 | 35.35 | 93.7 | |
| PDDType=Supervised, Train Data=Paired Noisy-Clean, #Param=0.71M2025.12 | 35.22 | 82.21 | |
| CVF-SIDCategory=Self-Supervised (Noisy Dataset Based Methods)2024.12 | 35.05 | 85.6 | |
| blind-spot denoisingCategory=Zero-shot2025.09 | 35.05 | 92.2 | |
| CVF-SIDType=Self-Supervised, Train Data=Only Noisy, #Param=1.19M2025.12 | 35.03 | 85.61 | |
| CVF-SIDLearning Paradigm=Self-Supervised, Evaluation Platform=Kaggle2026.04 | 35.03 | 85.6 | |
| MASHCategory=Zero-shot2025.09 | 34.89 | 92 | |
| MASHCategory=Self-Supervised (Single Image Based Methods)2024.12 | 34.8 | 81.4 | |
| R2RTraining Supervision=Self-Supervised2023.03 | 34.78 | 89.8 | |
| R2RCategory=Self-supervised2025.09 | 34.78 | 89.8 | |
| DnCNNTraining Noise Model=sRGB noise model (Our model)2022.06 | 34.74 | 91.2 | |
| CVF-SIDTraining Supervision=Self-Supervised2023.03 | 34.71 | 91.7 | |
| CVF-SIDCategory=Self-supervised2025.09 | 34.71 | 91.7 | |
| ScoreDVICategory=Zero-shot2025.09 | 34.7 | 89 | |
| LUD-VAECategory=Self-supervised2025.09 | 34.7 | 92 | |
| CBDNetType=Supervised, Train Data=Paired Noisy-Clean, #Param=6.79M2025.12 | 34.51 | 84.02 | |
| OursAvg. Infer. time (s)=37.2, Iterations=800, EMA weight=0.992024.01 | 34.32 | 90.3 | |
| BM3DType=Non-Learning, Train Data=None, #Param=-2025.12 | 34.26 | 69.5 | |
| Zhou et al.Training Supervision=Supervised (Synthetic pairs)2023.03 | 34 | 89.8 | |
| DnCNNTraining Noise Model=Noise Flow2022.06 | 33.81 | 89.4 | |
| DnCNNTraining Noise Model=C2N2022.06 | 33.76 | 90.1 | |
| DIPAvg. Infer. time (s)=333.22024.01 | 33.67 | 86.3 | |
| PD-denoisingCategory=Zero-shot2025.09 | 33.61 | 89 | |
| Ours (faster)Avg. Infer. time (s)=7.9, Iterations=200, EMA weight=0.92024.01 | 33.6 | 89.6 | |
| FasterDIPAvg. Infer. time (s)=138.22024.01 | 33.55 | 85.9 | |
| DnCNNTraining Noise Model=Diagonal Gaussian2022.06 | 33.34 | 86.7 | |
| CBDNetTraining Supervision=Supervised (Synthetic pairs)2023.03 | 33.28 | 86.8 | |
| CBDNetLearning Paradigm=Supervised2026.04 | 33.28 | 86.8 | |
| NN+denoisingCategory=Zero-shot2025.09 | 33.1 | 89.5 | |
| DnCNNTraining Noise Model=Full Gaussian2022.06 | 32.72 | 87.3 | |
| N2SType=Self-Supervised, Train Data=Only Noisy, #Param=2.58M2025.12 | 32.57 | 67.68 | |
| DnCNNTraining Noise Model=Isotropic Gaussian2022.06 | 32.48 | 85.5 | |
| UIDNetTraining Supervision=Unpaired2023.03 | 32.48 | 89.7 | |
| DnCNNTraining Noise Model=Heteroscedastic Gaussian2022.06 | 32.24 | 84.9 | |
| Masked DenoisingCategory=Supervised2024.12 | 31.99 | 60.1 | |
| DIPCategory=Zero-shot2025.09 | 31.97 | 85.4 | |
| N2VType=Self-Supervised, Train Data=Only Noisy, #Param=2.58M2025.12 | 31.77 | 59.79 | |
| UnprocessingCategory=Supervised2024.12 | 31.64 | 53.1 | |
| APBSN-singleCategory=Zero-shot2025.09 | 30.76 | 81.5 | |
| WNNMType=Non-Learning, Train Data=None, #Param=-2025.12 | 30.52 | 44.98 | |
| DnCNNType=Supervised, Train Data=Paired Noisy-Clean, #Param=0.66M2025.12 | 30.31 | 43.71 | |
| ZS-N2NCategory=Self-Supervised (Single Image Based Methods)2024.12 | 30.19 | 42.8 | |
| Noise2SelfTraining Supervision=Self-Supervised2023.03 | 29.56 | 80.8 | |
| Self2SelfCategory=Self-Supervised (Single Image Based Methods)2024.12 | 29.51 | 65.1 | |
| Self2SelfCategory=Zero-shot2025.09 | 29.51 | 65.1 | |
| NBR2NBRCategory=Self-supervised2025.09 | 27.9 | 67.9 | |
| ATBSN-singleCategory=Self-Supervised (Single Image Based Methods)2024.12 | 27.86 | 31.2 | |
| Noise2VoidTraining Supervision=Self-Supervised2023.03 | 27.68 | 66.8 | |
| Noise2VoidCategory=Self-Supervised (Noisy Dataset Based Methods)2024.12 | 27.68 | 66.8 | |
| N2S*Avg. Infer. time (s)=114.42024.01 | 26.93 | 65.8 | |
| DnCNNTraining Supervision=Supervised (Synthetic pairs)2023.03 | 26.25 | 59.9 | |
| WNNMTraining Supervision=Non-learning based2023.03 | 25.78 | 80.9 | |
| WNNMLearning Paradigm=Non-Learning2026.04 | 25.78 | 80.9 | |
| WNNMCategory=Zero-shot2025.09 | 25.78 | 80.9 | |
| BM3DTraining Supervision=Non-learning based2023.03 | 25.65 | 68.5 | |
| BM3DLearning Paradigm=Non-Learning2026.04 | 25.65 | 68.5 | |
| ZS-N2NAvg. Infer. time (s)=15.12024.01 | 25.61 | 55.9 | |
| N2V*Avg. Infer. time (s)=98.12024.01 | 25.34 | 59.5 |