Out-of-distribution Detection on CIFAR-10 vs SVHN (test)
99.84AUROCForte+SVM
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Forte+SVMSupervision=Unsupervised2024.10 | 99.84 | 0 | — | — | — | — | — | — | — | — | — | — | |
| Forte+GMMSupervision=Unsupervised2024.10 | 99.49 | 0 | — | — | — | — | — | — | — | — | — | — | |
| Forte+KDESupervision=Unsupervised2024.10 | 98.37 | 7.53 | — | — | — | — | — | — | — | — | — | — | |
| DOSESupervision=Unsupervised2024.10 | 97.3 | 13.16 | — | — | — | — | — | — | — | — | — | — | |
| TTSupervision=Unsupervised2024.10 | 87 | 61.28 | — | — | — | — | — | — | — | — | — | — | |
| LLRSupervision=Unsupervised2024.10 | 64.21 | 76.36 | — | — | — | — | — | — | — | — | — | — | |
| WAICSupervision=Unsupervised2024.10 | 14.3 | 98.83 | — | — | — | — | — | — | — | — | — | — | |
| q(X|theta_eta)Supervision=Unsupervised2024.10 | 6.5 | 100 | — | — | — | — | — | — | — | — | — | — | |
| AHGCBackbone=ResNet-18, Optimizer=SGD2024.12 | 0.9999 | 0 | — | 99.99 | 100 | — | — | — | 91.89 | 94.97 | 95.37 | 95.4 | |
| Reg. MahalanobisBackbone=ResNet182026.02 | 0.9967 | — | — | — | — | 99.05 | — | — | — | — | — | — | |
| OursBackbone=ResNet2019.12 | 0.995 | — | 96.7 | — | — | — | 97.6 | — | — | — | — | — | |
| Deep Abstaining ClassifierBackbone=ResNet 342021.05 | 0.9949 | 1.89 | — | — | — | — | — | — | — | — | — | — | |
| Deep Abstaining ClassifierBackbone=Wide ResNet 40x22021.05 | 0.9946 | 2 | — | — | — | — | — | — | — | — | — | — | |
| Mahalanobis distance-based scoreBackbone=ResNet, Validation Setup=OOD samples2018.07 | 0.991 | 96.4 | 95.8 | — | — | — | — | — | — | — | — | — | |
| MahalanobisBackbone=ResNet2018.07 | 0.991 | 96.4 | 95.8 | 98.3 | 99.6 | — | — | — | — | — | — | — | |
| MahalanobisBackbone=ResNet2019.12 | 0.991 | — | 95.8 | — | — | — | 87.8 | — | — | — | — | — | |
| OursBackbone=DenseNet2019.12 | 0.991 | — | 95.9 | — | — | — | 96.1 | — | — | — | — | — | |
| COMBOODBackbone=ResNet182026.02 | 0.9906 | — | — | — | — | 97.79 | — | — | — | — | — | — | |
| Auxiliary Rotation Predictionaverage_of_runs=52019.06 | 0.989 | 2.7 | — | — | — | 89.8 | — | — | — | — | — | — | |
| Outlier ExposureBackbone=Wide ResNet 40x22021.05 | 0.984 | 4.8 | — | — | — | — | — | — | — | — | — | — | |
| Mahalanobis distance-based scoreBackbone=DenseNet, Validation Setup=OOD samples2018.07 | 0.981 | 90.8 | 93.9 | — | — | — | — | — | — | — | — | — | |
| MahalanobisBackbone=DenseNet2018.07 | 0.981 | 90.8 | 93.9 | 96.6 | 99.2 | — | — | — | — | — | — | — | |
| MahalanobisBackbone=DenseNet2019.12 | 0.981 | — | 93.9 | — | — | — | 90.8 | — | — | — | — | — | |
| AENIBScore=log Dir_0.05(y) + log N(z_n; 0, I)2023.03 | 0.98 | — | — | — | — | — | — | — | — | — | — | — | |
| Mahalanobis distance-based scoreBackbone=DenseNet, Validation Setup=adversarial samples2018.07 | 0.976 | 89.6 | 92.6 | — | — | — | — | — | — | — | — | — | |
| UDGBackbone=ResNet-18, Optimizer=SGD2024.12 | 0.9749 | 13.26 | — | 95.66 | 98.69 | — | — | — | 36.64 | 56.81 | 76.77 | 89.54 | |
| SupConScore=max_y p(y|x)2023.03 | 0.97 | — | — | — | — | — | — | — | — | — | — | — | |
| VIBScore=log Dir_0.05(y)2023.03 | 0.97 | — | — | — | — | — | — | — | — | — | — | — | |
| ODINBackbone=ResNet, Validation Setup=OOD samples2018.07 | 0.967 | 86.6 | 91.1 | — | — | — | — | — | — | — | — | — | |
| ODINBackbone=ResNet2018.07 | 0.967 | 86.6 | 91.1 | 92.5 | 98.5 | — | — | — | — | — | — | — | |
| ODINBackbone=ResNet2019.12 | 0.967 | — | 91.1 | — | — | — | 70.3 | — | — | — | — | — | |
| REGradBackbone=ViT2024.04 | 0.9662 | — | — | — | — | 95.69 | — | — | — | — | — | — | |
| OEBackbone=ResNet-18, Optimizer=SGD2024.12 | 0.9643 | 20.88 | — | 93.62 | 98.32 | — | — | — | 32.72 | 47.33 | 67.2 | 86.75 | |
| AdaFocalObjective Category=Baseline2026.06 | 0.9637 | — | — | — | — | — | — | — | — | — | — | — | |
| ConjNormBackbone=ResNet-18, Optimizer=SGD2024.12 | 0.9632 | 9.89 | — | 94.88 | 96.25 | — | — | — | 33.82 | 67.94 | 80.31 | 91.72 | |
| MOODFLOPs=0.79 x 10^82021.04 | 0.96 | — | — | — | — | — | — | — | — | — | — | — | |
| Cross-entropyScore=log Dir_0.05(y)2023.03 | 0.96 | — | — | — | — | — | — | — | — | — | — | — | |
| GramBackbone=ResNet182026.02 | 0.9578 | — | — | — | — | 91 | — | — | — | — | — | — | |
| ODINBackbone=DenseNet, Validation Setup=OOD samples2018.07 | 0.955 | 86.2 | 91.4 | — | — | — | — | — | — | — | — | — | |
| Mahalanobis distance-based scoreBackbone=ResNet, Validation Setup=adversarial samples2018.07 | 0.955 | 75.8 | 89.1 | — | — | — | — | — | — | — | — | — | |
| ODINBackbone=DenseNet2018.07 | 0.955 | 86.2 | 91.4 | 78.8 | 98.3 | — | — | — | — | — | — | — | |
| ODINBackbone=DenseNet2019.12 | 0.955 | — | 91.4 | — | — | — | 86.2 | — | — | — | — | — | |
| Deep Mahalanobis DetectorBackbone=ResNet 342021.05 | 0.955 | 24.2 | — | — | — | — | — | — | — | — | — | — | |
| SconeBackbone=ResNet-18, Optimizer=SGD2024.12 | 0.9548 | 11.15 | — | 93.46 | 94.58 | — | — | — | 32.18 | 58.44 | 72.38 | 90.33 | |
| Serrà et altype=likelihood ratio-based, variant=best results, FLOPs=2.78 x 10^102021.04 | 0.95 | — | — | — | — | — | — | — | — | — | — | — | |
| VIBScore=max_y p(y|x)2023.03 | 0.95 | — | — | — | — | — | — | — | — | — | — | — | |
| Cross-entropyScore=max_y p(y|x)2023.03 | 0.94 | — | — | — | — | — | — | — | — | — | — | — | |
| DFLObjective Category=Baseline2026.06 | 0.9387 | — | — | — | — | — | — | — | — | — | — | — | |
| LDExtra data=No, Intrusive=No2024.04 | 0.936 | — | — | — | — | — | — | — | — | — | — | — | |
| Deep EnsemblesExtra data=No, Intrusive=Yes, Number of models=52024.04 | 0.933 | — | — | — | — | — | — | — | — | — | — | — | |
| LL ratioExtra data=Yes, Intrusive=Yes2024.04 | 0.93 | — | — | — | — | — | — | — | — | — | — | — | |
| LL ratioExtra data=Yes, Intrusive=Yes2024.04 | 0.93 | — | — | — | — | — | — | — | — | — | — | — | |
| ODINBackbone=DenseNet, Validation Setup=adversarial samples2018.07 | 0.928 | 70.5 | 86.5 | — | — | — | — | — | — | — | — | — | |
| DUQExtra data=No, Intrusive=Yes2024.04 | 0.927 | — | — | — | — | — | — | — | — | — | — | — | |
| KNNBackbone=ResNet182026.02 | 0.9267 | — | — | — | — | 96.25 | — | — | — | — | — | — | |
| Deep EnsemblesExtra data=No, Intrusive=Yes, Number of models=32024.04 | 0.926 | — | — | — | — | — | — | — | — | — | — | — | |
| KNNExtra data=Yes, Intrusive=No2024.04 | 0.926 | — | — | — | — | — | — | — | — | — | — | — | |
| ExGradBackbone=ViT2024.04 | 0.9241 | — | — | — | — | 88.06 | — | — | — | — | — | — | |
| Perturb x2024.04 | 0.9235 | — | — | — | — | 91.57 | — | — | — | — | — | — | |
| REGrad*layer-selective=true, perturbation=true2024.04 | 0.9232 | — | — | — | — | 89.42 | — | — | — | — | — | — | |
| TSObjective Category=Baseline2026.06 | 0.922 | — | — | — | — | — | — | — | — | — | — | — | |
| EBOBackbone=ResNet-18, Optimizer=SGD2024.12 | 0.9208 | 30.56 | — | 80.95 | 96.28 | — | — | — | 1.85 | 5.74 | 21.44 | 75.81 | |
| Maximum Softmax Probability (MSP)average_of_runs=52019.06 | 0.919 | 25.7 | — | — | — | 64 | — | — | — | — | — | — | |
| LA2024.04 | 0.9186 | — | — | — | — | 88.06 | — | — | — | — | — | — | |
| Clipped HNNsScore=Energy score2021.07 | 0.9157 | 49.04 | — | — | — | 98.12 | — | — | — | — | — | — | |
| FeDLaS-MbLSObjective Category=MbLS-based2026.06 | 0.914 | — | — | — | — | — | — | — | — | — | — | — | |
| CEObjective Category=Baseline2026.06 | 0.9137 | — | — | — | — | — | — | — | — | — | — | — | |
| Clipped HNNScoring function=softmax score2021.07 | 0.9134 | 49.89 | — | — | — | 98.13 | — | — | — | — | — | — | |
| EnergyExtra data=Yes, Intrusive=Yes2024.04 | 0.912 | — | — | — | — | — | — | — | — | — | — | — | |
| EnergyExtra data=Yes, Intrusive=Yes2024.04 | 0.912 | — | — | — | — | — | — | — | — | — | — | — | |
| OpenMaxBackbone=ResNet 342021.05 | 0.9072 | 23.67 | — | — | — | — | — | — | — | — | — | — | |
| AENIBScore=log Dir_0.05(y)2023.03 | 0.9 | — | — | — | — | — | — | — | — | — | — | — | |
| BaselineBackbone=DenseNet, Validation Setup=OOD samples2018.07 | 0.899 | 40.2 | 83.2 | — | — | — | — | — | — | — | — | — | |
| BaselineBackbone=DenseNet, Validation Setup=adversarial samples2018.07 | 0.899 | 40.2 | 83.2 | — | — | — | — | — | — | — | — | — | |
| BaselineBackbone=ResNet, Validation Setup=OOD samples2018.07 | 0.899 | 32.5 | 85.1 | — | — | — | — | — | — | — | — | — | |
| BaselineBackbone=ResNet, Validation Setup=adversarial samples2018.07 | 0.899 | 32.5 | 85.1 | — | — | — | — | — | — | — | — | — | |
| BaselineBackbone=DenseNet2018.07 | 0.899 | 40.2 | 83.2 | 83.1 | 94.7 | — | — | — | — | — | — | — | |
| BaselineBackbone=ResNet2018.07 | 0.899 | 32.5 | 85.1 | 85.4 | 94 | — | — | — | — | — | — | — | |
| BaselineBackbone=ResNet2019.12 | 0.899 | — | 85.1 | — | — | — | 50.5 | — | — | — | — | — | |
| BaselineBackbone=DenseNet2019.12 | 0.899 | — | 83.2 | — | — | — | 40.2 | — | — | — | — | — | |
| MbLSObjective Category=MbLS-based2026.06 | 0.899 | — | — | — | — | — | — | — | — | — | — | — | |
| MCDBackbone=ResNet-18, Optimizer=SGD2024.12 | 0.8978 | 60.27 | — | 85.33 | 94.25 | — | — | — | 20.05 | 38.23 | 55.43 | 74.01 | |
| Perturb θ2024.04 | 0.8931 | — | — | — | — | 82.22 | — | — | — | — | — | — | |
| ACLSObjective Category=Baseline2026.06 | 0.8914 | — | — | — | — | — | — | — | — | — | — | — | |
| Perturb θBackbone=ViT2024.04 | 0.8913 | — | — | — | — | 81.26 | — | — | — | — | — | — | |
| ExGrad V Term2024.04 | 0.8904 | — | — | — | — | 88.06 | — | — | — | — | — | — | |
| JEMtype=generative-based, variant=best results2021.04 | 0.89 | — | — | — | — | — | — | — | — | — | — | — | |
| JEMScore=max_y p(y|x)2023.03 | 0.89 | — | — | — | — | — | — | — | — | — | — | — | |
| Exgrad2024.04 | 0.8887 | — | — | — | — | 84.85 | — | — | — | — | — | — | |
| Entropy2024.04 | 0.8886 | — | — | — | — | 84.64 | — | — | — | — | — | — | |
| Inserted Dropout2024.04 | 0.8884 | — | — | — | — | 84.1 | — | — | — | — | — | — | |
| LEM (S2)score=∥∇Vθ∥2026.05 | 0.885 | — | — | — | — | — | — | — | — | — | — | — | |
| Vanilla HNNsScore=Energy score2021.07 | 0.8845 | 57.33 | — | — | — | 97.44 | — | — | — | — | — | — | |
| AENIBScore=max_y p(y|x)2023.03 | 0.88 | — | — | — | — | — | — | — | — | — | — | — | |
| SWAGOptimizer=SWAG2025.11 | 0.879 | 19.7 | — | 84.1 | 92.4 | — | — | 17.5 | — | — | — | — | |
| SWAGOptimizer=SWAG2025.11 | 0.879 | 19.7 | — | — | 92.4 | — | — | — | — | — | — | — | |
| MC-AA2024.04 | 0.8756 | — | — | — | — | 82.96 | — | — | — | — | — | — | |
| uCBOptvartheta=-> 0+2025.11 | 0.869 | 20.8 | — | 82.4 | 91.8 | — | — | 18.6 | — | — | — | — | |
| uCBOptOptimizer=uCBOpt2025.11 | 0.869 | 20.8 | — | — | 91.8 | — | — | — | — | — | — | — | |
| uCBOpt-adaptOptimizer=uCBOpt-adapt2025.11 | 0.866 | 21 | — | 82.1 | 91.5 | — | — | 18.7 | — | — | — | — |