Image Denoising on SIDD (val)
40.39PSNRCGNet
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CGNetMACs (G)=62.1, Inference Time (ms)=522024.01 | 40.39 | 96.4 | — | — | |
| NAFNetMACs (G)=63.6, Inference Time (ms)=532024.01 | 40.3 | 96.2 | — | — | |
| CATMACs (G)=135.7, Inference Time (ms)=3902024.01 | 40.05 | 96 | — | — | |
| RestormerMACs (G)=141, Inference Time (ms)=1022024.01 | 40.02 | 96 | — | — | |
| HINetMACs (G)=170.72024.01 | 39.99 | 96 | — | — | |
| PNGAN + MIRNet*Method Category=GAN-based, Evaluation Protocol=official out-of-the-box models2024.12 | 39.98 | 95.9 | — | — | |
| NAFNetGeneralization=None2024.12 | 39.97 | 96 | — | — | |
| MAXIMMACs (G)=169.52024.01 | 39.96 | 95.8 | — | — | |
| RestormerTraining Supervision=Supervised (Real pairs)2023.03 | 39.93 | 96 | 0.198 | — | |
| RestormerLearning Paradigm=Supervised2026.04 | 39.93 | 96 | — | — | |
| LANBaseline=NAFNet, reproduced=true2024.12 | 39.88 | 95.94 | — | — | |
| AFMBaseline=NAFNet, reproduced=true2024.12 | 39.83 | 95.92 | — | — | |
| CycleISPMACs (G)=189.52024.01 | 39.52 | 95.7 | — | — | |
| DANetLearning Paradigm=Supervised2023.03 | 39.47 | 91.8 | — | — | |
| DANet*Method Category=GAN-based, Evaluation Protocol=official out-of-the-box models2024.12 | 39.47 | 95.7 | — | — | |
| DeamNetType=Supervised, Train Data=Paired Noisy-Clean, #Param=2.23M2025.12 | 39.4 | 91.69 | — | — | |
| VDNTraining Supervision=Supervised (Real pairs)2023.03 | 39.29 | 95.6 | 0.208 | — | |
| VDNLearning Paradigm=Supervised2023.03 | 39.29 | 91.1 | — | — | |
| VDNetType=Supervised, Train Data=Paired Noisy-Clean, #Param=7.82M2025.12 | 39.29 | 91.09 | — | — | |
| VDNLearning Paradigm=Supervised2026.04 | 39.29 | 95.6 | — | — | |
| VDNTraining Data=Retrained on SIDD2019.08 | 39.28 | 90.9 | — | — | |
| NAFNet + NTNMethod Category=Ours2024.12 | 39.24 | 95.7 | — | — | |
| NTNBaseline=NAFNet2024.12 | 39.24 | 95.7 | — | — | |
| MPRNetMACs (G)=5882024.01 | 39.17 | 95.8 | — | — | |
| Baseline, N2CTraining Supervision=Supervised (Real pairs)2023.03 | 38.98 | 95.4 | 0.201 | — | |
| InvDNLearning Paradigm=Supervised2023.03 | 38.88 | 91.1 | — | — | |
| AINDNet(R)Learning Paradigm=Supervised2023.03 | 38.81 | 91.1 | — | — | |
| RIDNetLearning Paradigm=Supervised2023.03 | 38.76 | 91.3 | — | — | |
| RIDNetType=Supervised, Train Data=Paired Noisy-Clean, #Param=1.50M2025.12 | 38.71 | 95.11 | — | — | |
| CBDNetTraining Data=Retrained on SIDD2019.08 | 38.68 | 90.1 | — | — | |
| DnCNN-BTraining Data=Retrained on SIDD2019.08 | 38.41 | 90.9 | — | — | |
| TM-BSN (D)Learning Paradigm=Self-Supervised, Model Variant=D2026.04 | 38.08 | 95.2 | — | — | |
| APR (RD)Learning Paradigm=Self-Supervised, Model Variant=(RD)2026.04 | 38 | 94.7 | — | — | |
| Ours-Mixed (50%)Models=DnCNN, Mixing Ratio=50%2026.03 | 37.96 | 90.47 | — | — | |
| AT-BSN (D)Learning Paradigm=Self-Supervised2023.03 | 37.88 | 94.6 | — | — | |
| AT-BSN (D)Learning Paradigm=Self-Supervised, Model Variant=(D)2026.04 | 37.88 | 94.6 | — | — | |
| DnCNNTraining Supervision=Supervised (Real pairs)2023.03 | 37.73 | 94.3 | 0.245 | — | |
| DnCNNLearning Paradigm=Supervised2023.03 | 37.73 | 94.3 | — | — | |
| DnCNNLearning Paradigm=Supervised2026.04 | 37.73 | 94.3 | — | — | |
| Real (100%)Models=DnCNN, Mixing Ratio=100%2026.03 | 37.72 | 89.05 | — | — | |
| TBSNLearning Paradigm=Self-Supervised2026.04 | 37.71 | — | — | — | |
| NAFlow-Mixed (50%)Models=DnCNN, Mixing Ratio=50%2026.03 | 37.66 | 90.43 | — | — | |
| Ours (100%)Models=DnCNN, Mixing Ratio=100%2026.03 | 37.63 | 89.6 | — | — | |
| RealDenoising Network=DnCNN, Training Noise Source=Real noisy-clean pairs2026.03 | 37.63 | 93.6 | — | — | |
| PNGDenoising Network=DnCNN, Training Noise Source=Synthetic2026.03 | 37.55 | 93.7 | — | — | |
| NeCA-W-Mixed (50%)Models=DnCNN, Mixing Ratio=50%2026.03 | 37.51 | 89.52 | — | — | |
| PUCALearning Paradigm=Self-Supervised2026.04 | 37.49 | 88 | — | — | |
| BNN (LAN)Training Supervision=Self-Supervised2023.03 | 37.39 | 93.4 | 0.176 | — | |
| Spatially-Adaptive (UNet)Learning Paradigm=Self-Supervised2023.03 | 37.39 | 93.4 | — | — | |
| SASLLearning Paradigm=Self-Supervised2026.04 | 37.39 | 93.4 | — | — | |
| LG-BPNLearning Paradigm=Self-Supervised2023.03 | 37.32 | 88.6 | — | — | |
| LG-BPNLearning Paradigm=Self-Supervised2026.04 | 37.32 | 88.6 | — | — | |
| TM-BSN (s=4)Learning Paradigm=Self-Supervised, Model Variant=s=42026.04 | 37.31 | 94 | — | — | |
| SDAP (E)Learning Paradigm=Self-Supervised2023.03 | 37.3 | 89.4 | — | — | |
| SDAPLearning Paradigm=Self-Supervised2026.04 | 37.3 | 89.4 | — | — | |
| NAFlow (100%)Models=DnCNN, Mixing Ratio=100%2026.03 | 37.27 | 89.89 | — | — | |
| NAFlowDenoising Network=DnCNN, Training Noise Source=Synthetic2026.03 | 37.22 | 93.5 | — | — | |
| RealTrained on=Real noisy data, Evaluation model=DnCNN2025.12 | 37.16 | 89.9 | — | — | |
| NSP(TBSN)Type=Self-Supervised, Train Data=Only Noisy, #Param=12.77M2025.12 | 37.12 | 88.53 | — | — | |
| GuidNoiseTrained on=Synthetic data, Evaluation model=DnCNN2025.12 | 37.07 | 90.1 | — | — | |
| NSP(DBSN)Type=Self-Supervised, Train Data=Only Noisy, #Param=3.75M2025.12 | 37.02 | 88.65 | — | — | |
| NAFlowTrained on=Synthetic data, Evaluation model=DnCNN2025.12 | 37 | 89.5 | — | — | |
| NeCA-WDenoising Network=DnCNN, Training Noise Source=Synthetic2026.03 | 36.82 | 93.2 | — | — | |
| AT-BSNLearning Paradigm=Self-Supervised2023.03 | 36.8 | 93.4 | — | — | |
| AP-BSN+R3Training Supervision=Self-Supervised2023.03 | 36.74 | 93.4 | 0.281 | — | |
| APBSNCategory=Self-supervised2025.09 | 36.73 | 87.8 | — | — | |
| TBSNType=Self-Supervised, Train Data=Only Noisy, #Param=12.74M2025.12 | 36.59 | 85.74 | — | — | |
| SDAPType=Self-Supervised, Train Data=Only Noisy, #Param=3.66M2025.12 | 36.58 | 86.3 | — | — | |
| QWD-GANCategory=Self-supervised2025.09 | 36.31 | 95.6 | — | — | |
| NeCA-W (100%)Models=DnCNN, Mixing Ratio=100%2026.03 | 36.26 | 88.12 | — | — | |
| C-BSNLearning Paradigm=Self-Supervised2023.03 | 36.22 | 93.5 | — | — | |
| C-BSNLearning Paradigm=Self-Supervised2026.04 | 36.22 | 93.5 | — | — | |
| NeCA-WTrained on=Synthetic data, Evaluation model=DnCNN2025.12 | 36.2 | 89.7 | — | — | |
| AP-BSN + R3Learning Paradigm=Self-Supervised2023.03 | 35.76 | 91.2 | — | — | |
| AP-BSN + R3Learning Paradigm=Self-Supervised2026.04 | 35.76 | — | — | — | |
| C2NTraining Supervision=Unpaired2023.03 | 35.36 | 93.2 | 0.192 | — | |
| C2NLearning Paradigm=Unpaired2023.03 | 35.36 | 93.2 | — | — | |
| C2NType=Pseudo-Supervised, Train Data=Unpaired Noisy-Clean, #Param=217.26M2025.12 | 35.36 | 89.01 | — | — | |
| C2N + DIDN*Method Category=GAN-based, Evaluation Protocol=official out-of-the-box models2024.12 | 35.36 | 93.21 | — | — | |
| C2NLearning Paradigm=Unpaired2026.04 | 35.36 | 93.2 | — | — | |
| blind-spot denoisingCategory=Zero-shot2025.09 | 35.31 | 86.8 | — | — | |
| Positive2NegativeCategory=Self-Supervised (Single Image Based Methods)2024.12 | 35.24 | 88.9 | — | — | |
| MASHCategory=Self-Supervised (Single Image Based Methods)2024.12 | 35.06 | 85.1 | — | — | |
| MASHCategory=Zero-shot2025.09 | 35.06 | 85.1 | — | — | |
| R2RLearning Paradigm=Self-Supervised2023.03 | 35.04 | 84.4 | — | — | |
| R2RType=Pseudo-Supervised, Train Data=Paired Noisy-Noisy, #Param=0.56M2025.12 | 35.04 | 84.4 | — | — | |
| R2RCategory=Self-supervised2025.09 | 35.04 | 84.4 | — | — | |
| NeCA-STrained on=Synthetic data, Evaluation model=DnCNN2025.12 | 35.02 | 88 | — | — | |
| LUD-VAE2024.03 | 34.91 | 89.2 | — | — | |
| LUD-VAECategory=Self-Supervised (Noisy Dataset Based Methods)2024.12 | 34.91 | 89.2 | — | — | |
| LUD-VAECategory=Self-supervised2025.09 | 34.91 | 94.4 | — | — | |
| CVF-SID2024.03 | 34.81 | 94.4 | — | — | |
| CVF-SIDCategory=Self-Supervised (Noisy Dataset Based Methods)2024.12 | 34.81 | 87.2 | — | — | |
| CVF-SIDCategory=Self-supervised2025.09 | 34.81 | 94.4 | — | — | |
| CLIPDenoising2024.03 | 34.79 | 86.6 | — | — | |
| CLIPDenoisingCategory=Supervised2024.12 | 34.79 | 86.6 | — | — | |
| CLIPDenoising*Method Category=Others, Evaluation Protocol=official out-of-the-box models2024.12 | 34.79 | 89.82 | — | — | |
| ScoreDVICategory=Self-Supervised (Single Image Based Methods)2024.12 | 34.75 | 85.6 | — | — | |
| ScoreDVICategory=Zero-shot2025.09 | 34.75 | 85.6 | — | — | |
| Flow-sRGBDenoising Network=DnCNN, Training Noise Source=Synthetic2026.03 | 34.74 | 91.2 | — | — |