Image Classification on TinyImageNet (test)
90.23AccuracyEns.
Evaluation results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Ens.Model=ViT-B/162025.12 | 90.23 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DPSModel=ViT-B/162025.12 | 89.65 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PS-KDModel=ViT-B/162025.12 | 89.32 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DLBModel=ViT-B/162025.12 | 89.29 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| KDModel=ViT-B/162025.12 | 89.16 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TEModel=ViT-B/162025.12 | 89.02 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BaselineModel=ViT-B/162025.12 | 88.99 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| KNOWNPred. (xn)=×9, Pre-training=ImageNet2025.08 | 77.58 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| KNOWNPred. (xn)=×3, Pre-training=ImageNet2025.08 | 77.53 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TaskVectorPred. (xn)=×3, Pre-training=ImageNet2025.08 | 77.49 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LogFitPred. (xn)=×3, Pre-training=ImageNet2025.08 | 77.32 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LogFitPred. (xn)=×9, Pre-training=ImageNet2025.08 | 77.29 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TaskVectorPred. (xn)=×9, Pre-training=ImageNet2025.08 | 77.29 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TSVPred. (xn)=×9, Pre-training=ImageNet2025.08 | 76.18 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Naïve Transfer (Baseline)Pred. (xn)=×1, Pre-training=ImageNet2025.08 | 76.17 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TSVPred. (xn)=×3, Pre-training=ImageNet2025.08 | 76.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MagMaxPred. (xn)=×9, Pre-training=ImageNet2025.08 | 75.98 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Ens.Model=ResNet-1012025.12 | 69.78 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Ens.Model=DenseNet-2012025.12 | 69.44 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TempBalance + SNRModel=ResNet34, Initial learning rate η0=0.1, λr=0.001, Scaling ratio (s1, s2)=(0.6, 1.4)2023.12 | 69.27 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TempBalanceModel=ResNet34, Initial learning rate η0=0.1, Scaling ratio (s1, s2)=(0.6, 1.4)2023.12 | 69.12 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SNRModel=ResNet34, Initial learning rate η0=0.1, λr=0.001, 0.005, 0.01, 0.0152023.12 | 68.69 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CALModel=ResNet34, Initial learning rate η0=0.05, 0.1, 0.152023.12 | 68.19 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DPSModel=DenseNet-2012025.12 | 67.74 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DPSModel=ResNet-1012025.12 | 67.41 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Manifold MixupBackbone=RN50, Batch Size=128, Optimizer=SGD with momentum 0.92025.09 | 67.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RCADBackbone=ResNet-182022.06 | 67.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ERM + DReSBackbone=RN50, Batch Size=128, Optimizer=SGD with momentum 0.92025.09 | 67.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| KDModel=DenseNet-2012025.12 | 66.94 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ME-ADABackbone=ResNet-182022.06 | 66.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TempBalance + SNRModel=ResNet18, Initial learning rate η0=0.1, λr=0.001, Scaling ratio (s1, s2)=(0.6, 1.4)2023.12 | 66.86 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TempBalance + SNRModel=WRN28-6, Initial learning rate η0=0.1, λr=0.0001, Scaling ratio (s1, s2)=(0.6, 1.4)2023.12 | 66.79 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TempBalanceModel=ResNet18, Initial learning rate η0=0.1, Scaling ratio (s1, s2)=(0.6, 1.4)2023.12 | 66.77 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DLBModel=DenseNet-2012025.12 | 66.66 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TempBalanceModel=WRN28-6, Initial learning rate η0=0.1, Scaling ratio (s1, s2)=(0.6, 1.4)2023.12 | 66.58 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LCDModel=ResNet-502026.05 | 66.57 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ADABackbone=ResNet-182022.06 | 66.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DSALModel=ResNet-502026.05 | 66.46 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Teacher (Scratch)Architecture=ResNet34/ResNet182022.09 | 66.44 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ERMBackbone=ResNet-182022.06 | 66.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HCDModel=ResNet-502026.05 | 66.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| KDModel=ResNet-1012025.12 | 66.36 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LCHCModel=ResNet-502026.05 | 66.31 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CALModel=ResNet18, Initial learning rate η0=0.05, 0.1, 0.152023.12 | 66.25 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SNRModel=ResNet18, Initial learning rate η0=0.1, λr=0.001, 0.005, 0.01, 0.0152023.12 | 66.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PS-KDModel=DenseNet-2012025.12 | 66.11 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SNRModel=WRN28-6, Initial learning rate η0=0.1, λr=0.00005, 0.0001, 0.0012023.12 | 66.09 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CALModel=WRN28-6, Initial learning rate η0=0.05, 0.1, 0.152023.12 | 65.88 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Label SmoothingBackbone=ResNet-1012021.11 | 65.87 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DLBModel=ResNet-1012025.12 | 65.83 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MbLSBackbone=ResNet-1012021.11 | 65.81 | — | — | — | -0.06 | — | — | — | — | — | — | — | — | — | |
| Label SmoothingBackbone=ResNet-502021.11 | 65.78 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MbLS (m=0)Backbone=ResNet-101, Margin=02021.11 | 65.72 | — | — | — | -0.15 | — | — | — | — | — | — | — | — | — | |
| Entropy Capacity PenaltyBackbone=ResNet-1012021.11 | 65.69 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PS-KDModel=ResNet-1012025.12 | 65.65 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Cross-EntropyBackbone=ResNet-1012021.11 | 65.62 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MixupBackbone=RN50, Batch Size=128, Optimizer=SGD with momentum 0.92025.09 | 65.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MbLS (m=0)Backbone=ResNet-50, Margin=02021.11 | 65.15 | — | — | — | -0.63 | — | — | — | — | — | — | — | — | — | |
| Cross-EntropyBackbone=ResNet-502021.11 | 65.02 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Entropy Capacity PenaltyBackbone=ResNet-502021.11 | 64.98 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SGD-ERArchitecture=ResNet-50, Decay Type=exponential2026.03 | 64.93 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Student (Scratch)Architecture=ResNet34/ResNet182022.09 | 64.87 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LCDModel=ResNet-182026.05 | 64.87 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MbLSBackbone=ResNet-502021.11 | 64.74 | — | — | — | -1.04 | — | — | — | — | — | — | — | — | — | |
| TEModel=DenseNet-2012025.12 | 64.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BaselineModel=DenseNet-2012025.12 | 64.53 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SGD-ERArchitecture=ResNet-34, Decay Type=exponential2026.03 | 64.53 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| WSDSArchitecture=ResNet-50, Decay Type=exponential2026.03 | 64.52 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TEModel=ResNet-1012025.12 | 64.45 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BaselineModel=ResNet-1012025.12 | 64.22 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LCHCModel=ResNet-182026.05 | 64.18 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DSALModel=ResNet-182026.05 | 64.15 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Focal Loss with Sample-Dependent smoothingBackbone=ResNet-502021.11 | 64.09 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TempBalanceModel=WRN16-8, Initial learning rate η0=0.1, Scaling ratio (s1, s2)=(0.6, 1.4)2023.12 | 64.09 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TempBalance + SNRModel=WRN16-8, Initial learning rate η0=0.1, λr=0.0001, Scaling ratio (s1, s2)=(0.6, 1.4)2023.12 | 64.08 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CMIArchitecture=ResNet34/ResNet182022.09 | 64.01 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DQASModel=ResNet-502026.05 | 64 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SNRModel=WRN16-8, Initial learning rate η0=0.1, λr=0.00005, 0.0001, 0.0012023.12 | 63.98 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| WRN-40-202021.03 | 63.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DFQArchitecture=ResNet34/ResNet182022.09 | 63.73 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CALModel=WRN16-8, Initial learning rate η0=0.05, 0.1, 0.152023.12 | 63.67 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NUIAModel=ResNet-182026.05 | 63.62 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FairGrad*alpha=0.1, Number of clients=1002025.08 | 63.46 | — | — | — | — | — | — | — | — | — | — | — | — | 0.56 | |
| HCDModel=ResNet-182026.05 | 63.44 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Focal LossBackbone=ResNet-502021.11 | 63.09 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Focal LossBackbone=ResNet-1012021.11 | 62.97 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Focal Loss with Sample-Dependent smoothingBackbone=ResNet-1012021.11 | 62.96 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ERMBackbone=RN50, Batch Size=128, Optimizer=SGD with momentum 0.92025.09 | 62.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FairGrad*alpha=0.5, Number of clients=1002025.08 | 62.84 | — | — | — | — | — | — | — | — | — | — | — | — | 0.36 | |
| Tensorion+SGDBackbone=ResNet-502026.06 | 62.7 | — | — | — | — | — | 1.65 | — | — | — | — | — | — | — | |
| CosAArchitecture=ResNet-50, Decay Type=exponential2026.03 | 62.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FGSMBackbone=ResNet-182022.06 | 62.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MADArchitecture=ResNet34/ResNet182022.09 | 62.32 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| WSDSArchitecture=ResNet-34, Decay Type=exponential2026.03 | 62.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FairGradalpha=0.1, Number of clients=1002025.08 | 62.15 | — | — | — | — | — | — | — | — | — | — | — | — | 0.56 | |
| FairGradalpha=0.05, Number of clients=1002025.08 | 62.06 | — | — | — | — | — | — | — | — | — | — | — | — | 0.52 | |
| OracleModel Class=ConvNet, Model=WRN-16-4, Number of Trials=49, Augmentation=++2024.03 | 61.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SelVSModel Class=ConvNet, Model=WRN-16-4, Number of Trials=49, Augmentation=++2024.03 | 61.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ZebraModel=ResNet-182026.05 | 61.46 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CosAArchitecture=ResNet-34, Decay Type=exponential2026.03 | 61.12 | — | — | — | — | — | — | — | — | — | — | — | — | — |