Near-OOD Detection on CIFAR-10 (AUROC, FPR95)
94.82AUROCOE
Evaluation results
| Method | Links | ||
|---|---|---|---|
| OECategory=With outliers2025.11 | 94.82 | 19.84 | |
| OEMethod category=Training methods with outliers, Outlier availability during training=Outliers used2025.11 | 94.82 | 19.84 | |
| LogitNormCategory=Without outliers2025.11 | 92.33 | 29.34 | |
| LogitNormMethod category=Training methods without outliers, Outlier availability during training=None2025.11 | 92.33 | 29.34 | |
| CRAFTCategory=Without outliers2025.11 | 91.11 | 31.94 | |
| CRAFTMethod category=Training methods without outliers, Outlier availability during training=None2025.11 | 91.11 | 31.94 | |
| MCDCategory=With outliers2025.11 | 91.03 | 30.17 | |
| MCDMethod category=Training methods with outliers, Outlier availability during training=Outliers used2025.11 | 91.03 | 30.17 | |
| AggBBAS-EDArchitecture=ResNet2026.03 | 90.76 | 34.03 | |
| AggBBAS-ECArchitecture=ResNet2026.03 | 90.74 | 34 | |
| CIDERCategory=Without outliers2025.11 | 90.71 | 32.11 | |
| CIDERMethod category=Training methods without outliers, Outlier availability during training=None2025.11 | 90.71 | 32.11 | |
| BBAS-EDArchitecture=ResNet2026.03 | 90.7 | 34.11 | |
| BBAS-ECArchitecture=ResNet2026.03 | 90.67 | 34.2 | |
| KNNArchitecture=ResNet2026.03 | 90.64 | 34.01 | |
| KNNCategory=Post-hoc inference2025.11 | 90.64 | 34.01 | |
| KNNMethod category=Post-hoc inference methods, Outlier availability during training=None2025.11 | 90.64 | 34.01 | |
| RankOODCategory=Without outliers2025.11 | 90.21 | 31.72 | |
| RankOODMethod category=Training methods without outliers, Outlier availability during training=None2025.11 | 90.21 | 31.72 | |
| UDGCategory=With outliers2025.11 | 89.91 | 35.34 | |
| UDGMethod category=Training methods with outliers, Outlier availability during training=Outliers used2025.11 | 89.91 | 35.34 | |
| ConfBranchCategory=Without outliers2025.11 | 89.84 | 31.28 | |
| ConfBranchMethod category=Training methods without outliers, Outlier availability during training=None2025.11 | 89.84 | 31.28 | |
| RMDSArchitecture=ResNet2026.03 | 89.8 | 38.89 | |
| RMDSCategory=Post-hoc inference2025.11 | 89.8 | 38.89 | |
| RMDSMethod category=Post-hoc inference methods, Outlier availability during training=None2025.11 | 89.8 | 38.89 | |
| CSICategory=Without outliers2025.11 | 89.51 | 33.66 | |
| G-ODINCategory=Without outliers2025.11 | 89.12 | 45.54 | |
| G-ODINMethod category=Training methods without outliers, Outlier availability during training=None2025.11 | 89.12 | 45.54 | |
| MixOECategory=With outliers2025.11 | 88.73 | 51.45 | |
| MixOEMethod category=Training methods with outliers, Outlier availability during training=Outliers used2025.11 | 88.73 | 51.45 | |
| VIMArchitecture=ResNet2026.03 | 88.68 | 44.84 | |
| VIMCategory=Post-hoc inference2025.11 | 88.68 | 44.84 | |
| VIMMethod category=Post-hoc inference methods, Outlier availability during training=None2025.11 | 88.68 | 44.84 | |
| GENArchitecture=ResNet2026.03 | 88.2 | 53.67 | |
| GENCategory=Post-hoc inference2025.11 | 88.2 | 53.67 | |
| GENMethod category=Post-hoc inference methods, Outlier availability during training=None2025.11 | 88.2 | 53.67 | |
| TempScaleCategory=Post-hoc inference2025.11 | 88.09 | 50.96 | |
| TempScaleMethod category=Post-hoc inference methods, Outlier availability during training=None2025.11 | 88.09 | 50.96 | |
| MSPArchitecture=ResNet2026.03 | 88.03 | 48.17 | |
| MSPCategory=Post-hoc inference2025.11 | 88.03 | 48.17 | |
| MSPMethod category=Post-hoc inference methods, Outlier availability during training=None2025.11 | 88.03 | 48.17 | |
| OpenMaxCategory=Post-hoc inference2025.11 | 87.62 | 43.62 | |
| EBOCategory=Post-hoc inference2025.11 | 87.58 | 61.34 | |
| EBOMethod category=Post-hoc inference methods, Outlier availability during training=None2025.11 | 87.58 | 61.34 | |
| MLSCategory=Post-hoc inference2025.11 | 87.52 | 61.32 | |
| MLSMethod category=Post-hoc inference methods, Outlier availability during training=None2025.11 | 87.52 | 61.32 | |
| ARPLCategory=Without outliers2025.11 | 87.44 | 40.33 | |
| ReActCategory=Post-hoc inference2025.11 | 87.11 | 63.56 | |
| ReActMethod category=Post-hoc inference methods, Outlier availability during training=None2025.11 | 87.11 | 63.56 | |
| ExCeLCategory=Post-hoc inference2025.11 | 86.89 | 66.55 | |
| ExCeLMethod category=Post-hoc inference methods, Outlier availability during training=None2025.11 | 86.89 | 66.55 | |
| SCALEArchitecture=ResNet2026.03 | 84.84 | — | |
| MDSArchitecture=ResNet2026.03 | 84.2 | 49.9 | |
| MDSCategory=Post-hoc inference2025.11 | 84.2 | 49.9 | |
| ODINArchitecture=ResNet2026.03 | 82.87 | 76.19 | |
| ODINCategory=Post-hoc inference2025.11 | 82.87 | 76.19 | |
| SHEArchitecture=ResNet2026.03 | 81.54 | 79.65 | |
| SHECategory=Post-hoc inference2025.11 | 81.54 | 79.65 | |
| RankFeatCategory=Post-hoc inference2025.11 | 79.46 | 60.88 | |
| KLMCategory=Post-hoc inference2025.11 | 79.19 | 87.86 | |
| DICECategory=Post-hoc inference2025.11 | 78.34 | 70.04 | |
| ASHArchitecture=ResNet2026.03 | 75.27 | 86.78 | |
| ASHCategory=Post-hoc inference2025.11 | 75.27 | 86.78 | |
| ASHMethod category=Post-hoc inference methods, Outlier availability during training=None2025.11 | 75.27 | 86.78 | |
| MOSCategory=Without outliers2025.11 | 71.45 | 78.72 | |
| MDSEnsArchitecture=ResNet2026.03 | 60.43 | 92.96 | |
| MDSEnsCategory=Post-hoc inference2025.11 | 60.43 | 92.26 | |
| GramCategory=Post-hoc inference2025.11 | 58.66 | 90.87 | |
| GradNormCategory=Post-hoc inference2025.11 | 54.9 | 94.72 | |
| OpenGANCategory=Post-hoc inference2025.11 | 53.71 | 94.48 |