List your company
ResearchFormerly WizWand

Image Denoising on SIDD (test)

45.62PSNRPAP + HB
22.781628.710834.6440.5692Aug 29, 2019Oct 4, 2020Nov 10, 2021Dec 17, 2022Jan 23, 2024Feb 28, 2025Apr 7, 2026
Updated 6mo ago

Evaluation results

MethodLinks
PAP + HBNoise Model Category=Real-noise-based, Backbone=U-shape net (SID), Synthesis method=pattern-aligned patch sampling with high-bit refinement2021.1045.620.978
ELDNoise Model Category=Physics-based, Backbone=U-shape net (SID)2021.1045.390.977
Calibrated P-GNoise Model Category=Physics-based, Backbone=U-shape net (SID)2021.1045.380.976
Pixel-wiseNoise Model Category=Real-noise-based, Backbone=U-shape net (SID)2021.1045.330.976
Camera-Aware GANNoise Model Category=DNN-based, Backbone=U-shape net (SID)2021.10450.975
Noise FlowNoise Model Category=DNN-based, Backbone=U-shape net (SID)2021.1044.680.973
Paired dataNoise Model Category=Real-noise-based, Backbone=U-shape net (SID)2021.1044.350.971
NAFNetMAC (G)=63.6, Supervised=true, Trained with=Real (SIDD)2023.0940.310.9667
RestormerMAC (G)=140.99, Supervised=true, Trained with=Real (SIDD)2023.0940.060.9601
MAXIM-3S2022.0139.960.96
MIRNet2022.0139.720.959
MPRNet2022.0139.710.958
CycleISPAdditional training data=true2022.0139.520.957
CycleISP2023.0539.520.957
DRANet2023.0539.50.957
SADNetAdditional training data=true2022.0139.460.957
SADNet2023.0539.460.957
DANet+2023.0539.430.956
DeamNet2023.0539.350.955
InvDNBlind Denoising=true, MultiScale=true, Noise Generation=true, Invertible=true, Num of Parameters=2.64, Infer Time=47.802021.0439.280.955
VDN2022.0139.280.956
VDN2023.0539.280.956
VDNBlind Denoising=true, Num of Parameters=7.81, Infer Time=99.002021.0439.260.955
VDIR2023.0539.260.955
VDN2019.0839.260.955
DANetBlind Denoising=true, MultiScale=true, Noise Generation=true, Num of Parameters=63.01, Infer Time=65.622021.0439.250.955
AINDNet + RTBlind/Non-blind=Blind, Training Env.=All, Geometric self-ensemble=true2020.0239.150.955
AINDNet + TFBlind/Non-blind=Blind, Training Env.=All, Geometric self-ensemble=true2020.0239.080.953
AINDNetBlind Denoising=true, Num of Parameters=13.762021.0439.080.953
AINDNet + RTBlind/Non-blind=Blind, Training Env.=All, Geometric self-ensemble=false2020.0239.040.955
AINDNet + TFBlind/Non-blind=Blind, Training Env.=All, Geometric self-ensemble=false2020.0238.950.952
AINDNetAdditional training data=true2022.0138.950.952
AINDNet2023.0538.950.952
LANBackbone=Restormer, Iteration=20, Loss=ZS-N2N2024.1238.860.965
AINDNet(R)Blind/Non-blind=Blind, Training Env.=Real, Geometric self-ensemble=true2020.0238.840.951
AINDNet(R)Blind/Non-blind=Blind, Training Env.=Real, Geometric self-ensemble=false2020.0238.730.95
RIDNetAdditional training data=true2022.0138.710.951
RIDNet2023.0538.710.954
LANBackbone=Restormer, Iteration=20, Loss=Nbr2Nbr2024.1238.380.958
LANBackbone=Uformer, Iteration=20, Loss=ZS-N2N2024.1238.360.964
GradNetBlind Denoising=true, Num of Parameters=1.602021.0438.340.946
pretrainedBackbone=Restormer, Iteration=0, Adaptation=None2024.1238.030.951
LANBackbone=Uformer, Iteration=20, Loss=Nbr2Nbr2024.1238.020.956
pretrainedBackbone=Uformer, Iteration=0, Adaptation=None2024.1237.550.95
MeDMAC (G)=26.18, Supervised=false, Trained with=NP + SIDD2023.0937.520.9434
MM-BSNMAC (G)=339.46, Supervised=false, Trained with=Real (SIDD)2023.0937.370.9362
Baseline (Unpruned)sparsity=0%, model_scope=Full Model2026.0437.26320.9327
LANBackbone=DnCNN, Iteration=20, Loss=Nbr2Nbr2024.1236.80.935
LANBackbone=DnCNN, Iteration=20, Loss=ZS-N2N2024.1236.780.938
pretrainedBackbone=DnCNN, Iteration=0, Adaptation=None2024.1236.60.93
DnCNN2025.0336.450.9
AINDNet(S)Blind/Non-blind=Blind, Training Env.=Synthetic, Geometric self-ensemble=true2020.0235.890.905
MeDMAC (G)=26.18, Supervised=false, Trained with=Synthetic (NP)2023.0935.810.8278
MeDTraining Data=Noise Pool (DIV2K)2023.0935.810.8278
AINDNet(S)Blind/Non-blind=Blind, Training Env.=Synthetic, Geometric self-ensemble=false2020.0235.660.903
BM3D2022.0135.650.685
DnLUT2025.0335.440.875
TWSC2023.0535.330.933
Hybrid QUBOsparsity=36%, model_scope=Full Model2026.0435.07150.914
TT Refinementsparsity=36%, model_scope=Full Model2026.0435.07150.914
SPF-LUT2025.0334.910.865
Taylor (Greedy)sparsity=36%, model_scope=Full Model2026.0434.80820.8974
CVF-SIDMAC (G)=77.86, Supervised=false, Trained with=Real (SIDD)2023.0934.710.9179
In-Camera P-GNoise Model Category=Physics-based, Backbone=U-shape net (SID)2021.1034.640.72
RC-LUT2025.0333.880.839
MCWNNM2023.0533.40.879
CBDNetBlind/Non-blind=Blind, Training Env.=All, Geometric self-ensemble=false2020.0233.280.868
CBDNetBlind Denoising=true, Noise Generation=true, Num of Parameters=4.34, Infer Time=80.762021.0433.280.868
CBDNet2019.0833.280.868
MuLUT2025.0333.240.83
N2NMAC (G)=26.18, Supervised=false, Trained with=Real (SIDD)2023.0932.820.7297
N2NTraining Data=Noise Pool (DIV2K)2023.0932.820.7297
N2SMAC (G)=26.18, Supervised=false, Trained with=Real (SIDD)2023.0930.980.6018
N2STraining Data=Noise Pool (DIV2K)2023.0930.980.6018
CBDNetAdditional training data=true2022.0130.780.801
CBDNet2023.0530.780.951
N2CTraining Data=Noise Pool (DIV2K)2023.0930.370.6028
DBD4Training Data=Noise Pool (DIV2K)2023.0930.230.6173
CBM3D2025.0330.140.702
R2RTraining Data=Noise Pool (DIV2K)2023.0929.640.5708
MC-WNNM2025.0329.450.682
SR-LUT2025.0329.380.634
LIRTraining Data=Noise Pool (DIV2K)2023.0928.760.5296
EPLL2021.0427.110.87
KSVDBlind/Non-blind=Non-blind, Geometric self-ensemble=false2020.0226.880.842
KSVD2021.0426.880.842
WNNMBlind/Non-blind=Non-blind, Geometric self-ensemble=false2020.0225.780.809
WNNM2019.0825.780.809
BM3DBlind/Non-blind=Non-blind, Geometric self-ensemble=false2020.0225.650.685
BM3D2021.0425.650.685
CBM3D2019.0825.650.685
Classic QUBO (L1)sparsity=36%, model_scope=Full Model2026.0424.79630.6346
TNRDBlind/Non-blind=Non-blind, Training Env.=Synthetic, Geometric self-ensemble=false2020.0224.730.643
TNRD2021.0424.730.643
MLPBlind/Non-blind=Non-blind, Training Env.=Synthetic, Geometric self-ensemble=false2020.0224.710.641
MLP2022.0124.710.641
MLP2019.0824.710.641
CDnCNN-BBlind/Non-blind=Blind, Training Env.=Synthetic, Geometric self-ensemble=false2020.0223.660.583
DnCNNBlind Denoising=true, Num of Parameters=0.562021.0423.660.583
DnCNN2022.0123.660.583