Out-of-Distribution Detection on CIFAR-10 ID CIFAR-100 OOD
96AUCKNN
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| KNNModel=BiT-M2025.05 | 96 | — | — | — | — | 19 | — | |
| GradPCAModel=BiT-M2025.05 | 93.68 | — | — | — | — | 28.67 | — | |
| MahalanobisModel=BiT-M2025.05 | 93.56 | — | — | — | — | 29.39 | — | |
| OEAdversarial radius (epsilon)=0.012021.06 | 91.1 | — | 94.91 | 0 | 0.9 | — | — | |
| PlainAdversarial radius (epsilon)=0.012021.06 | 90 | — | 95.01 | 0 | 0.7 | — | — | |
| ProoDAdversarial radius (epsilon)=0.01, Delta (bias shift)=32021.06 | 89.8 | — | 94.99 | 46.1 | 46.8 | — | — | |
| NNGuideBackbone=ResNet-182026.03 | 89.55 | — | — | — | — | — | 41.62 | |
| KNNModel=TIMM2025.05 | 89.36 | — | — | — | — | 54.04 | — | |
| LaRExBackbone=DenseNet-1002026.03 | 89.34 | — | — | — | — | — | 50.15 | |
| MSPBackbone=ResNet-182026.03 | 88.76 | — | — | — | — | — | 46.29 | |
| GAIA-AModel=BiT-M2025.05 | 88.72 | — | — | — | — | 39.48 | — | |
| Prototype FusionBackbone=DenseNet-1002026.03 | 88.34 | — | — | — | — | — | 45.81 | |
| LaRExBackbone=ResNet-182026.03 | 88.2 | — | — | — | — | — | 53.45 | |
| MaxLogitBackbone=ResNet-182026.03 | 88.09 | — | — | — | — | — | 53.99 | |
| Prototype FusionBackbone=ResNet-182026.03 | 87.94 | — | — | — | — | — | 50.27 | |
| ReActBackbone=DenseNet-1002026.03 | 87.33 | — | — | — | — | — | 51.97 | |
| NNGuideBackbone=DenseNet-1002026.03 | 87.2 | — | — | — | — | — | 46.6 | |
| GAIA-AModel=TIMM2025.05 | 86.92 | — | — | — | — | 55.12 | — | |
| MSPBackbone=DenseNet-1002026.03 | 86.78 | — | — | — | — | — | 47.06 | |
| ViMBackbone=DenseNet-1002026.03 | 86.65 | — | — | — | — | — | 57.02 | |
| NECOBackbone=DenseNet-1002026.03 | 86.56 | — | — | — | — | — | 58.07 | |
| ESOODBackbone=DenseNet-1002026.03 | 86.56 | — | — | — | — | — | 58.67 | |
| EnergyBackbone=DenseNet-1002026.03 | 86.44 | — | — | — | — | — | 59.12 | |
| MaxLogitBackbone=DenseNet-1002026.03 | 86.41 | — | — | — | — | — | 59.11 | |
| GradPCAModel=TIMM2025.05 | 86.32 | — | — | — | — | 53.86 | — | |
| ACETAdversarial radius (epsilon)=0.012021.06 | 86 | — | 93.43 | 0 | 4 | — | — | |
| MSPModel=BiT-M2025.05 | 85.71 | — | — | — | — | 38.84 | — | |
| ViMBackbone=ResNet-182026.03 | 85.55 | — | — | — | — | — | 53.83 | |
| MahalanobisModel=TIMM2025.05 | 85.37 | — | — | — | — | 68.19 | — | |
| EnergyBackbone=ResNet-182026.03 | 85.04 | — | — | — | — | — | 64.9 | |
| GAIA-ZModel=TIMM2025.05 | 84.63 | — | — | — | — | 51.78 | — | |
| ESOODBackbone=ResNet-182026.03 | 84.56 | — | — | — | — | — | 62.34 | |
| MSPModel=TIMM2025.05 | 83.9 | — | — | — | — | 62.71 | — | |
| MahalanobisBackbone=ResNet-182026.03 | 82.11 | — | — | — | — | — | 59.59 | |
| EnergyModel=TIMM2025.05 | 81.47 | — | — | — | — | 52.82 | — | |
| MaxLogitsModel=TIMM2025.05 | 81.43 | — | — | — | — | 53.73 | — | |
| ODINModel=TIMM2025.05 | 81.43 | — | — | — | — | 53.77 | — | |
| ODINModel=BiT-M2025.05 | 81.2 | — | — | — | — | 44.84 | — | |
| MaxLogitsModel=BiT-M2025.05 | 81.19 | — | — | — | — | 44.86 | — | |
| ReActBackbone=ResNet-182026.03 | 80.06 | — | — | — | — | — | 82.33 | |
| MahalanobisBackbone=DenseNet-1002026.03 | 78.99 | — | — | — | — | — | 67.97 | |
| EnergyModel=BiT-M2025.05 | 78.77 | — | — | — | — | 51.95 | — | |
| SGPAEModel Category=Kernel, Backbone=ViT, Data Augmentation=without2026.03 | 78.72 | 74.82 | — | — | — | — | — | |
| ATOMAdversarial radius (epsilon)=0.012021.06 | 78.3 | — | 93.63 | 0 | 21.7 | — | — | |
| DICEModel=TIMM2025.05 | 77.78 | — | — | — | — | 55.71 | — | |
| DEModel Category=Kernel, Backbone=ViT, Data Augmentation=without2026.03 | 77.57 | 74.23 | — | — | — | — | — | |
| NECOBackbone=ResNet-182026.03 | 76.86 | — | — | — | — | — | 67.72 | |
| GOOD80Adversarial radius (epsilon)=0.01, Classifier Architecture=different2021.06 | 76.7 | — | 87.39 | 47.1 | 57.1 | — | — | |
| Gaussian OOD + L_uncertaintyOutlier Synthesis=Gaussian, Loss=L_uncertainty2026.03 | 76.47 | 99.35 | — | — | — | 80 | — | |
| GCOS OOD + L_uncertaintyOutlier Synthesis=GCOS, Loss=L_uncertainty2026.03 | 76.13 | 99.33 | — | — | — | 83.5 | — | |
| ReActModel=TIMM2025.05 | 76.08 | — | — | — | — | 57.87 | — | |
| GradNormBackbone=ResNet-182026.03 | 76 | — | — | — | — | — | 77.34 | |
| SGPAModel Category=Kernel, Backbone=ViT, Data Augmentation=without2026.03 | 74.98 | 71.11 | — | — | — | — | — | |
| MCDModel Category=Kernel, Backbone=ViT, Data Augmentation=without2026.03 | 74.26 | 70.46 | — | — | — | — | — | |
| SNGPModel Category=Kernel, Backbone=ViT, Data Augmentation=without2026.03 | 73.91 | 70.23 | — | — | — | — | — | |
| GCOS OOD + L_regOutlier Synthesis=GCOS, Loss=L_reg2026.03 | 72.53 | 99.16 | — | — | — | 89.5 | — | |
| KFLLLAModel Category=Kernel, Backbone=ViT, Data Augmentation=without2026.03 | 72.28 | 68.89 | — | — | — | — | — | |
| MLEModel Category=Kernel, Backbone=ViT, Data Augmentation=without2026.03 | 72.08 | 67.74 | — | — | — | — | — | |
| MFVIModel Category=Kernel, Backbone=ViT, Data Augmentation=without2026.03 | 71.77 | 67.22 | — | — | — | — | — | |
| MLEModel Category=SDP, Backbone=ViT, Data Augmentation=without2026.03 | 68.93 | 64.33 | — | — | — | — | — | |
| GOOD100Adversarial radius (epsilon)=0.01, Classifier Architecture=different2021.06 | 67.8 | — | 86.96 | 48.1 | 49.7 | — | — | |
| ProoD-DiscAdversarial radius (epsilon)=0.012021.06 | 62.9 | — | — | 57.1 | 57.8 | — | — | |
| GradNormBackbone=DenseNet-1002026.03 | 60.24 | — | — | — | — | — | 94.3 | |
| GAIA-ZModel=BiT-M2025.05 | 53.02 | — | — | — | — | 90.02 | — | |
| DICEModel=BiT-M2025.05 | 48.24 | — | — | — | — | 95.07 | — | |
| ReActModel=BiT-M2025.05 | 44.21 | — | — | — | — | 96.84 | — | |
| DAEDL2025.10 | — | 88.19 | — | — | — | — | — | |
| DAEDL2026.05 | — | 88.19 | — | — | — | — | — | |
| Dropout2025.10 | — | 45.57 | — | — | — | — | — | |
| Dropout2026.05 | — | 85.39 | — | — | — | — | — | |
| DUQ2026.05 | — | 84.75 | — | — | — | — | — | |
| EDL2025.10 | — | 84.18 | — | — | — | — | — | |
| EDL2026.05 | — | 87.13 | — | — | — | — | — | |
| F-EDL2025.10 | — | 88.37 | — | — | — | — | — | |
| F-EDL2026.05 | — | 88.37 | — | — | — | — | — | |
| I-EDL2025.10 | — | 84.84 | — | — | — | — | — | |
| I-EDL2026.05 | — | 84.84 | — | — | — | — | — | |
| MoDEX2026.05 | — | 89.28 | — | — | — | — | — | |
| NatPN2026.05 | — | 86.98 | — | — | — | — | — | |
| PostNet2026.05 | — | 87.07 | — | — | — | — | — | |
| R-EDL2025.10 | — | 87.73 | — | — | — | — | — | |
| R-EDL2026.05 | — | 87.73 | — | — | — | — | — | |
| Re-EDL2026.05 | — | 88.31 | — | — | — | — | — | |
| RED2026.05 | — | 87.84 | — | — | — | — | — |