Image Denoising on CBSD68 (test)
48.1PSNRMeD
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| MeDTest Noise=Poisson Noise, Training Supervision=Invariant Feature2023.09 | 48.1 | 0.9916 | — | — | |
| MeDTest Noise=S&P, ε ∈ [0.3, 0.5], Training Supervision=Invariant Feature2023.09 | 42.33 | 0.9695 | — | — | |
| MeDTest Noise=Local Var Gaussian (LVG), Training Supervision=Invariant Feature2023.09 | 37.99 | 0.9568 | — | — | |
| MeDTest Noise=Speckle, v ∈ [25, 50], Training Supervision=Invariant Feature2023.09 | 37.21 | 0.9715 | — | — | |
| DIDNsigma=10, FLOPs=1121G, Resolution=256x2562022.03 | 36.48 | 0.957 | — | — | |
| RDUNetsigma=10, FLOPs=807G, Resolution=256x2562022.03 | 36.48 | 0.951 | — | — | |
| RDNsigma=10, FLOPs=1490G, Resolution=256x2562022.03 | 36.47 | — | — | — | |
| SRMNetsigma=10, FLOPs=285G, Resolution=256x2562022.03 | 36.46 | 0.961 | — | — | |
| RNANsigma=10, Resolution=256x2562022.03 | 36.43 | — | — | — | |
| FFDNetsigma=10, FLOPs=18G, Resolution=256x2562022.03 | 36.14 | 0.954 | — | — | |
| DnCNNsigma=10, FLOPs=36G, Resolution=256x2562022.03 | 36.12 | 0.951 | — | — | |
| IrCNNsigma=10, FLOPs=27G, Resolution=256x2562022.03 | 36.06 | 0.953 | — | — | |
| DHDNsigma=10, FLOPs=1019G, Resolution=256x2562022.03 | 36.05 | 0.953 | — | — | |
| MeDTest Noise=Average, Training Supervision=Invariant Feature2023.09 | 36.02 | 0.9145 | — | — | |
| BM3Dsigma=10, Resolution=256x2562022.03 | 35.89 | 0.951 | — | — | |
| RestormerNoise Type=Gaussian, Noise Level=sigma=152024.03 | 34.42 | 0.936 | — | — | |
| MeDTest Noise=LVG + Poisson, Training Supervision=Invariant Feature2023.09 | 34.29 | 0.9325 | — | — | |
| N2CTest Noise=Average, Training Supervision=Noisy/Clean2023.09 | 34.08 | 0.8803 | — | — | |
| N2NTest Noise=Average, Training Supervision=Noisy/Noisy2023.09 | 34.05 | 0.8626 | — | — | |
| CLIPDenoisingNoise Type=Gaussian, Noise Level=sigma=15, Training Requirement=one noise level2024.03 | 33.97 | 0.93 | — | — | |
| DnCNNNoise Type=Gaussian, Noise Level=sigma=152024.03 | 33.78 | 0.931 | — | — | |
| DBDTest Noise=Average, Training Supervision=Noisy/Clean2023.09 | 33.28 | 0.8634 | — | — | |
| HATNoise Type=Gaussian, Noise Level=sigma=15, Training Requirement=multiple noise levels2024.03 | 33.22 | 0.912 | — | — | |
| N2STest Noise=Average, Training Supervision=Noisy/Noisy2023.09 | 33.18 | 0.8607 | — | — | |
| MeDTest Noise=Poisson + Speckle v = 25, Training Supervision=Invariant Feature2023.09 | 33.04 | 0.9258 | — | — | |
| DILNoise Type=Gaussian, Noise Level=sigma=15, Training Requirement=multiple noise levels2024.03 | 32.5 | 0.906 | — | — | |
| R2RTest Noise=Average, Training Supervision=Noisy/Noisy2023.09 | 32.23 | 0.8199 | — | — | |
| CLIPDenoisingNoise Type=Salt&Pepper, Noise Level=alpha=0.012, Training Requirement=one noise level2024.03 | 31.95 | 0.89 | — | — | |
| MeDTest Noise=Gaussian σ = 25 + Speckle v = 25, Training Supervision=Invariant Feature2023.09 | 31.92 | 0.8726 | — | — | |
| CLIPDenoisingNoise Type=Speckle, Noise Level=sigma^2=0.02, Training Requirement=one noise level2024.03 | 31.82 | 0.904 | — | — | |
| IR-FlowNFE=1, sigma=252026.04 | 31.7 | 0.8927 | 0.096 | — | |
| IR-FlowNFE=2, sigma=252026.04 | 31.63 | 0.8938 | 0.078 | — | |
| IDRApproach Type=Specialized all-in-one approach, Noise level=252025.05 | 31.6 | 0.887 | — | — | |
| IR-FlowNFE=3, sigma=252026.04 | 31.6 | 0.893 | 0.071 | — | |
| RestormerApproach Type=General image restorer, Noise level=252025.05 | 31.49 | 0.884 | — | — | |
| IR-FlowNFE=4, sigma=252026.04 | 31.44 | 0.8922 | 0.068 | — | |
| InstructIRApproach Type=Specialized all-in-one approach, Noise level=252025.05 | 31.4 | 0.887 | — | — | |
| LIRTest Noise=Average, Training Supervision=Invariant Feature2023.09 | 31.29 | 0.8044 | — | — | |
| DFPIRApproach Type=Specialized all-in-one approach, Noise level=252025.05 | 31.29 | 0.889 | — | — | |
| DnCNNTVNoise-level sigma=252022.04 | 31.26 | — | — | — | |
| DnCNNNoise-level sigma=252022.04 | 31.24 | — | — | — | |
| DNCNNnoise level=252023.01 | 31.24 | 0.883 | 0.109 | 43.51 | |
| DnCNNNFE=1, sigma=252026.04 | 31.24 | 0.883 | 0.109 | — | |
| FFDNETnoise level=252023.01 | 31.22 | 0.8821 | 0.121 | 49.64 | |
| FFDNetNFE=1, sigma=252026.04 | 31.22 | 0.8821 | 0.121 | — | |
| DENOISING-ODEnoise level=25, total steps=222023.01 | 31.14 | 0.8777 | 0.074 | 28.71 | |
| IR-SDENFE=22, sigma=252026.04 | 31.14 | 0.8777 | 0.074 | — | |
| DGUNetApproach Type=General image restorer, Noise level=252025.05 | 31.1 | 0.883 | — | — | |
| CLIPDenoisingNoise Type=Gaussian, Noise Level=sigma=25, Training Requirement=one noise level2024.03 | 31.02 | 0.878 | — | — | |
| NAFNetApproach Type=General image restorer, Noise level=252025.05 | 31.02 | 0.883 | — | — | |
| IR-SDE (Denoising-ODE)Crop Size=256x2562023.01 | 31.01 | 0.8746 | 0.069 | 29.72 | |
| MaskDenoisingNoise Type=Gaussian, Noise Level=sigma=152024.03 | 30.99 | 0.888 | — | — | |
| AirNetApproach Type=Specialized all-in-one approach, Noise level=252025.05 | 30.91 | 0.882 | — | — | |
| CLIPDenoisingNoise Type=Salt&Pepper, Noise Level=alpha=0.016, Training Requirement=one noise level2024.03 | 30.85 | 0.857 | — | — | |
| CNN-BASELINEnoise level=252023.01 | 30.74 | 0.8661 | 0.162 | 56.64 | |
| BaselineNFE=1, sigma=252026.04 | 30.74 | 0.8661 | 0.162 | — | |
| RDUNetsigma=30, FLOPs=807G, Resolution=256x2562022.03 | 30.72 | 0.872 | — | — | |
| SRMNetsigma=30, FLOPs=285G, Resolution=256x2562022.03 | 30.72 | 0.878 | — | — | |
| DIDNsigma=30, FLOPs=1121G, Resolution=256x2562022.03 | 30.71 | 0.87 | — | — | |
| RDNsigma=30, FLOPs=1490G, Resolution=256x2562022.03 | 30.67 | — | — | — | |
| RNANsigma=30, Resolution=256x2562022.03 | 30.63 | — | — | — | |
| SwinIRApproach Type=General image restorer, Noise level=252025.05 | 30.59 | 0.868 | — | — | |
| CLIPDenoisingNoise Type=Speckle, Noise Level=sigma^2=0.03, Training Requirement=one noise level2024.03 | 30.48 | 0.886 | — | — | |
| DnCNNsigma=30, FLOPs=36G, Resolution=256x2562022.03 | 30.32 | 0.861 | — | — | |
| FFDNetsigma=30, FLOPs=18G, Resolution=256x2562022.03 | 30.31 | 0.86 | — | — | |
| IrCNNsigma=30, FLOPs=27G, Resolution=256x2562022.03 | 30.22 | 0.861 | — | — | |
| TAPEApproach Type=Specialized all-in-one approach, Noise level=252025.05 | 30.18 | 0.855 | — | — | |
| DHDNsigma=30, FLOPs=1019G, Resolution=256x2562022.03 | 30.12 | 0.858 | — | — | |
| CLIPDenoisingNoise Type=Poisson, Noise Level=alpha=2.5, Training Requirement=one noise level2024.03 | 29.94 | 0.874 | — | — | |
| BM3Dsigma=30, Resolution=256x2562022.03 | 29.71 | 0.843 | — | — | |
| CNN-BASELINECrop Size=256x2562023.01 | 29.69 | 0.8529 | 0.17 | 59.99 | |
| MeDTest Noise=Gaussian σ = 50 + Speckle v = 25, Training Supervision=Invariant Feature2023.09 | 29.68 | 0.7928 | — | — | |
| MeDTest Noise=Gaussian, sigma ∈ [15, 75], Training Supervision=Invariant Feature2023.09 | 29.61 | 0.8178 | — | — | |
| DDRMCrop Size=256x2562023.01 | 29.57 | 0.8484 | 0.135 | 58.99 | |
| CLIPDenoisingNoise Type=Speckle, Noise Level=sigma^2=0.04, Training Requirement=one noise level2024.03 | 29.49 | 0.87 | — | — | |
| CLIPDenoisingNoise Type=Spatial Gaussian, Noise Level=sigma=45, Training Requirement=one noise level2024.03 | 29.43 | 0.849 | — | — | |
| N2CTest Noise=Gaussian, sigma ∈ [15, 75], Training Supervision=Noisy/Clean2023.09 | 29.24 | 0.7754 | — | — | |
| N2NTest Noise=Gaussian, sigma ∈ [15, 75], Training Supervision=Noisy/Noisy2023.09 | 29.23 | 0.7634 | — | — | |
| DBDTest Noise=Gaussian, sigma ∈ [15, 75], Training Supervision=Noisy/Clean2023.09 | 29.05 | 0.7616 | — | — | |
| TransweatherApproach Type=Specialized all-in-one approach, Noise level=252025.05 | 29 | 0.841 | — | — | |
| CLIPDenoisingNoise Type=Poisson, Noise Level=alpha=3.0, Training Requirement=one noise level2024.03 | 28.71 | 0.846 | — | — | |
| N2STest Noise=Gaussian, sigma ∈ [15, 75], Training Supervision=Noisy/Noisy2023.09 | 28.58 | 0.7589 | — | — | |
| CLIPDenoisingNoise Type=Spatial Gaussian, Noise Level=sigma=50, Training Requirement=one noise level2024.03 | 28.51 | 0.825 | — | — | |
| RDUNetsigma=50, FLOPs=807G, Resolution=256x2562022.03 | 28.38 | 0.807 | — | — | |
| SRMNetsigma=50, FLOPs=285G, Resolution=256x2562022.03 | 28.38 | 0.814 | — | — | |
| DIDNsigma=50, FLOPs=1121G, Resolution=256x2562022.03 | 28.35 | 0.804 | — | — | |
| RNANsigma=50, Resolution=256x2562022.03 | 28.31 | — | — | — | |
| RDNsigma=50, FLOPs=1490G, Resolution=256x2562022.03 | 28.31 | — | — | — | |
| IR-SDEnoise level=25, total steps=1002023.01 | 28.09 | 0.7866 | 0.101 | 36.49 | |
| DnCNNTVNoise-level sigma=502022.04 | 28.07 | — | — | — | |
| FFDNetsigma=50, FLOPs=18G, Resolution=256x2562022.03 | 27.96 | 0.788 | — | — | |
| DnCNNNoise-level sigma=502022.04 | 27.95 | — | — | — | |
| DnCNNsigma=50, FLOPs=36G, Resolution=256x2562022.03 | 27.92 | 0.788 | — | — | |
| IrCNNsigma=50, FLOPs=27G, Resolution=256x2562022.03 | 27.86 | 0.789 | — | — | |
| DHDNsigma=50, FLOPs=1019G, Resolution=256x2562022.03 | 27.71 | 0.787 | — | — | |
| DENOISING-SDEnoise level=25, total steps=222023.01 | 27.65 | 0.7457 | 0.131 | 39.25 | |
| BM3Dsigma=50, Resolution=256x2562022.03 | 27.36 | 0.763 | — | — | |
| LIRTest Noise=Gaussian, sigma ∈ [15, 75], Training Supervision=Invariant Feature2023.09 | 26.67 | 0.6866 | — | — | |
| R2RTest Noise=Gaussian, sigma ∈ [15, 75], Training Supervision=Noisy/Noisy2023.09 | 26.43 | 0.6639 | — | — | |
| DLApproach Type=Specialized all-in-one approach, Noise level=252025.05 | 23.09 | 0.745 | — | — |