Out-of-Distribution Detection on CIFAR100 (test)
95.8AUROCOpenMatch
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| OpenMatchlabeled samples per class=100, supervision level=partial2021.05 | 95.8 | — | — | |
| TV-OODArchitecture=DenseNet121, Auxiliary OOD dataset=Required2026.01 | 93.87 | 24.93 | 98.42 | |
| VIMArchitecture=DenseNet121, Auxiliary OOD dataset=None2026.01 | 92.45 | 28.7 | 98.08 | |
| SADA-JEMScore Function=log p_theta(x), Training Dataset=CIFAR10, K=202022.09 | 92 | — | — | |
| KNNArchitecture=DenseNet121, Auxiliary OOD dataset=None2026.01 | 91.56 | 36.57 | 97.88 | |
| BEOEArchitecture=DenseNet121, Auxiliary OOD dataset=Required2026.01 | 91.35 | 47.58 | 98.01 | |
| SADA-JEMScore Function=max_y p_theta(y|x), Training Dataset=CIFAR10, K=202022.09 | 91 | — | — | |
| Supervisedsupervision level=full2021.05 | 90.4 | — | — | |
| SADA-JEMScore Function=log p_theta(x), Training Dataset=CIFAR10, K=52022.09 | 90 | — | — | |
| SADA-JEMScore Function=log p_theta(x), Training Dataset=CIFAR10, K=102022.09 | 90 | — | — | |
| MTClabeled samples per class=100, supervision level=partial2021.05 | 90 | — | — | |
| SADA-JEMScore Function=max_y p_theta(y|x), Training Dataset=CIFAR10, K=102022.09 | 89 | — | — | |
| TV-OODArchitecture=DenseNet121, Auxiliary OOD dataset=None2026.01 | 88.34 | 46.01 | 97.05 | |
| MahalanobisArchitecture=DenseNet121, Auxiliary OOD dataset=None2026.01 | 88.24 | 34.16 | 96.76 | |
| JEM++Score Function=max_y p_theta(y|x), Training Dataset=CIFAR10, M=202022.09 | 88 | — | — | |
| SADA-JEMScore Function=max_y p_theta(y|x), Training Dataset=CIFAR10, K=52022.09 | 88 | — | — | |
| DICEpost-hoc=true2021.11 | 87.23 | 49.72 | — | |
| WideResNetScore Function=log p_theta(x), Training Dataset=CIFAR102022.09 | 87 | — | — | |
| JEMScore Function=max_y p_theta(y|x), Training Dataset=CIFAR10, K=202022.09 | 87 | — | — | |
| NNGuideBackbone=ResNet-18, Pre-training=From scratch, ID Accuracy=75.66%2023.09 | 86.39 | 64.56 | 86.96 | |
| EFArchitecture=DenseNet121, Auxiliary OOD dataset=Required2026.01 | 86.14 | 59.94 | 96.68 | |
| WOODSArchitecture=DenseNet121, Auxiliary OOD dataset=Required2026.01 | 86.02 | 54.99 | 94.35 | |
| OEArchitecture=DenseNet121, Auxiliary OOD dataset=Required2026.01 | 85.71 | 64.93 | 96.68 | |
| KLBackbone=ResNet-18, Pre-training=From scratch, ID Accuracy=75.66%2023.09 | 85.47 | 66.27 | 85.84 | |
| EnergyBackbone=ResNet-18, Pre-training=From scratch, ID Accuracy=75.66%2023.09 | 85.47 | 66.27 | 85.84 | |
| G-ODINpost-hoc=false, requires model retraining=true2021.11 | 85.24 | 52.87 | — | |
| MaxLogitBackbone=ResNet-18, Pre-training=From scratch, ID Accuracy=75.66%2023.09 | 85.14 | 67.92 | 85.61 | |
| WideResNetScore Function=max_y p_theta(y|x), Training Dataset=CIFAR102022.09 | 85 | — | — | |
| ODINpost-hoc=true2021.11 | 84.49 | 58.14 | — | |
| ReActpost-hoc=true2021.11 | 84.47 | 62.27 | — | |
| KNNBackbone=ResNet-18, Pre-training=From scratch, ID Accuracy=75.66%2023.09 | 84.17 | 70.07 | 84.23 | |
| MahalanobisBackbone=WideResNet-402021.10 | 84.1 | — | — | |
| ASH-SArchitecture=DenseNet121, Auxiliary OOD dataset=None2026.01 | 83.79 | 62.19 | 96.13 | |
| ImCDScore Function=log p_theta(x), Training Dataset=CIFAR102022.09 | 83 | — | — | |
| Mahalanobispost-hoc=true2021.11 | 82.73 | 55.37 | — | |
| MSPBackbone=ResNet-18, Pre-training=From scratch, ID Accuracy=75.66%2023.09 | 82.34 | 72.53 | 83.05 | |
| LogitNormArchitecture=DenseNet121, Auxiliary OOD dataset=None2026.01 | 81.71 | 72.42 | 94.2 | |
| Energypost-hoc=true2021.11 | 81.19 | 68.45 | — | |
| MLSBackbone=WideResNet-402021.10 | 80.8 | — | — | |
| Labeled Onlylabeled samples per class=100, supervision level=partial2021.05 | 80.4 | — | — | |
| EnergyArchitecture=DenseNet121, Auxiliary OOD dataset=None2026.01 | 80.18 | 74.37 | 94.91 | |
| ODINArchitecture=DenseNet121, Auxiliary OOD dataset=None2026.01 | 80.14 | 74.06 | 94.93 | |
| Energy ScoreBackbone=WideResNet-402021.10 | 79.6 | — | — | |
| MahalanobisBackbone=ResNet-18, Pre-training=From scratch, ID Accuracy=75.66%2023.09 | 78.88 | 77.54 | 79.18 | |
| MSPArchitecture=DenseNet121, Auxiliary OOD dataset=None2026.01 | 78.34 | 78.55 | 94.45 | |
| ViMBackbone=ResNet-18, Pre-training=From scratch, ID Accuracy=75.66%2023.09 | 77.7 | 80.07 | 78.63 | |
| ODINBackbone=WideResNet-402021.10 | 77.4 | — | — | |
| GradNormBackbone=ResNet-18, Pre-training=From scratch, ID Accuracy=75.66%2023.09 | 76.71 | 66.9 | 72.14 | |
| MSPBackbone=WideResNet-402021.10 | 75.5 | — | — | |
| MSPpost-hoc=true2021.11 | 74.36 | 80.13 | — | |
| VERAScore Function=log p_theta(x), Training Dataset=CIFAR102022.09 | 73 | — | — | |
| FixMatchlabeled samples per class=100, supervision level=partial2021.05 | 71.3 | — | — | |
| SSDBackbone=ResNet-18, Pre-training=From scratch, ID Accuracy=75.66%2023.09 | 69.05 | 83.21 | 65.52 | |
| JEM++Score Function=log p_theta(x), Training Dataset=CIFAR10, M=202022.09 | 68 | — | — | |
| JEMScore Function=log p_theta(x), Training Dataset=CIFAR10, K=202022.09 | 67 | — | — | |
| ROSS-fDBDAttack radius (ϵ)=2/255, Attack type=PGD-Max2026.05 | 65.42 | 72.1 | — | |
| ROSS-fDBDAttack radius (ϵ)=2/255, Attack type=PGD-Min2026.05 | 64.47 | 73.67 | — | |
| ROSS-GENAttack radius (ϵ)=2/255, Attack type=PGD-Max2026.05 | 64.19 | 75.55 | — | |
| ROSS-EBOAttack radius (ϵ)=2/255, Attack type=PGD-Max2026.05 | 63.96 | 75.76 | — | |
| ROSS-GENAttack radius (ϵ)=2/255, Attack type=PGD-Min2026.05 | 63.24 | 76.99 | — | |
| ROSS-EBOAttack radius (ϵ)=2/255, Attack type=PGD-Min2026.05 | 62.96 | 77.2 | — | |
| ROSS-MSPAttack radius (ϵ)=2/255, Attack type=PGD-Max2026.05 | 61.12 | 77.61 | — | |
| ROSS-MSPAttack radius (ϵ)=2/255, Attack type=PGD-Min2026.05 | 60.73 | 79.08 | — | |
| PRO-fDBDAttack radius (ϵ)=2/255, Attack type=PGD-Min2026.05 | 58.93 | 81.35 | — | |
| ROSS-fDBDAttack radius (ϵ)=4/255, Attack type=PGD-Max2026.05 | 58.48 | 77.52 | — | |
| ROSS-GENAttack radius (ϵ)=4/255, Attack type=PGD-Max2026.05 | 57.69 | 81.07 | — | |
| ROSS-EBOAttack radius (ϵ)=4/255, Attack type=PGD-Max2026.05 | 57.56 | 81.28 | — | |
| ROSS-fDBDAttack radius (ϵ)=4/255, Attack type=PGD-Min2026.05 | 57.13 | 80.43 | — | |
| ROSS-GENAttack radius (ϵ)=4/255, Attack type=PGD-Min2026.05 | 56.66 | 83.33 | — | |
| ROSS-EBOAttack radius (ϵ)=4/255, Attack type=PGD-Min2026.05 | 56.44 | 83.45 | — | |
| fDBDAttack radius (ϵ)=2/255, Attack type=PGD-Min2026.05 | 54.74 | 83.11 | — | |
| ODINAttack radius (ϵ)=2/255, Attack type=PGD-Min2026.05 | 54.01 | 88.37 | — | |
| IGEBMScore Function=max_y p_theta(y|x), Training Dataset=CIFAR102022.09 | 54 | — | — | |
| ROSS-MSPAttack radius (ϵ)=4/255, Attack type=PGD-Max2026.05 | 52.84 | 82.57 | — | |
| ODINAttack radius (ϵ)=2/255, Attack type=PGD-Max2026.05 | 52.81 | 88.98 | — | |
| ROSS-MSPAttack radius (ϵ)=4/255, Attack type=PGD-Min2026.05 | 52.73 | 85.01 | — | |
| PRO-GENAttack radius (ϵ)=2/255, Attack type=PGD-Max2026.05 | 51.45 | 87.33 | — | |
| fDBDAttack radius (ϵ)=2/255, Attack type=PGD-Max2026.05 | 51.21 | 84.29 | — | |
| PRO-fDBDAttack radius (ϵ)=2/255, Attack type=PGD-Max2026.05 | 50.76 | 83.72 | — | |
| PRO-GENAttack radius (ϵ)=2/255, Attack type=PGD-Min2026.05 | 50.33 | 84.91 | — | |
| IGEBMScore Function=log p_theta(x), Training Dataset=CIFAR102022.09 | 50 | — | — | |
| GENAttack radius (ϵ)=2/255, Attack type=PGD-Min2026.05 | 49.92 | 86.72 | — | |
| MSPAttack radius (ϵ)=2/255, Attack type=PGD-Min2026.05 | 49.66 | 86.75 | — | |
| EBOAttack radius (ϵ)=2/255, Attack type=PGD-Min2026.05 | 48.83 | 88.27 | — | |
| EBOAttack radius (ϵ)=2/255, Attack type=PGD-Max2026.05 | 47.35 | 91.77 | — | |
| GENAttack radius (ϵ)=2/255, Attack type=PGD-Max2026.05 | 47.02 | 90.58 | — | |
| ROSS-fDBDAttack radius (ϵ)=8/255, Attack type=PGD-Max2026.05 | 44.72 | 86.3 | — | |
| ROSS-GENAttack radius (ϵ)=8/255, Attack type=PGD-Max2026.05 | 43.94 | 89.19 | — | |
| ROSS-EBOAttack radius (ϵ)=8/255, Attack type=PGD-Max2026.05 | 43.92 | 89.23 | — | |
| ROSS-EBOAttack radius (ϵ)=8/255, Attack type=PGD-Min2026.05 | 43.78 | 92.29 | — | |
| ROSS-GENAttack radius (ϵ)=8/255, Attack type=PGD-Min2026.05 | 43.73 | 92.33 | — | |
| MSPAttack radius (ϵ)=2/255, Attack type=PGD-Max2026.05 | 43.06 | 90.33 | — | |
| ROSS-fDBDAttack radius (ϵ)=8/255, Attack type=PGD-Min2026.05 | 42.59 | 90.68 | — | |
| PRO-fDBDAttack radius (ϵ)=4/255, Attack type=PGD-Min2026.05 | 38.82 | 94.01 | — | |
| ROSS-MSPAttack radius (ϵ)=8/255, Attack type=PGD-Min2026.05 | 38 | 93.57 | — | |
| ROSS-MSPAttack radius (ϵ)=8/255, Attack type=PGD-Max2026.05 | 37.54 | 90.3 | — | |
| fDBDAttack radius (ϵ)=4/255, Attack type=PGD-Min2026.05 | 32.94 | 95.53 | — | |
| MSPAttack radius (ϵ)=4/255, Attack type=PGD-Min2026.05 | 32.4 | 97.3 | — | |
| ODINAttack radius (ϵ)=4/255, Attack type=PGD-Max2026.05 | 31.36 | 97.53 | — | |
| fDBDAttack radius (ϵ)=4/255, Attack type=PGD-Max2026.05 | 30.4 | 96.54 | — |