Out-of-Distribution Detection on CIFAR-10 (test)
0.984AUROCMSP with OECC
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| MSP with OECCBackbone=WRN-40-2, Fine-tuning method=OECC, Detector type=Maximum Softmax Probability (MSP)2019.06 | 0.984 | 6.56 | 93.08 | — | — | — | — | — | — | |
| SupCLR + LoRot-IBatch size=512, Trial count=32022.07 | 0.9795 | — | — | — | — | — | — | — | — | |
| SupCLR + LoRot-EBatch size=512, Trial count=32022.07 | 0.9792 | — | — | — | — | — | — | — | — | |
| MSP with Outlier ExposureBackbone=WRN-40-2, Fine-tuning method=Outlier Exposure (OE), Detector type=Maximum Softmax Probability (MSP)2019.06 | 0.9781 | 9.5 | 90.48 | — | — | — | — | — | — | |
| SupCLR [32]Batch size=512, Trial count=3, Reproduced=true2022.07 | 0.9698 | — | — | — | — | — | — | — | — | |
| SupCLR + Rotation (PT)Batch size=512, Trial count=32022.07 | 0.969 | — | — | — | — | — | — | — | — | |
| SupCLR + Rotation (MT)Batch size=512, Trial count=32022.07 | 0.9628 | — | — | — | — | — | — | — | — | |
| DICEpost-hoc=true2021.11 | 0.9524 | 20.83 | — | — | — | — | — | — | — | |
| MLSBackbone=WideResNet-402021.10 | 0.951 | — | — | — | — | — | — | — | — | |
| ReActpost-hoc=true2021.11 | 0.9495 | 26.45 | — | — | — | — | — | — | — | |
| Cross Entropyaugmentation=APR-SP2021.08 | 0.947 | — | — | 0.977 | 0.895 | 0.979 | 0.963 | 0.937 | 0.928 | |
| Energypost-hoc=true2021.11 | 0.9457 | 26.55 | — | — | — | — | — | — | — | |
| VOSBackbone=WideResNet-402021.10 | 0.941 | — | — | — | — | — | — | — | — | |
| ATDAttack Source=OpenGAN-fea2022.09 | 0.94 | — | — | — | — | — | — | — | — | |
| ATDAttack Source=ViT (MD)2022.09 | 0.94 | — | — | — | — | — | — | — | — | |
| CSI2021.08 | 0.94 | — | — | 0.965 | 0.905 | 0.963 | 0.962 | 0.921 | 0.924 | |
| ODINpost-hoc=true2021.11 | 0.9371 | 24.57 | — | — | — | — | — | — | — | |
| MahalanobisBackbone=WideResNet-402021.10 | 0.933 | — | — | — | — | — | — | — | — | |
| ATDAttack Source=AT (MD)2022.09 | 0.933 | — | — | — | — | — | — | — | — | |
| ATDAttack Source=ALOE (OM)2022.09 | 0.933 | — | — | — | — | — | — | — | — | |
| Cross Entropyaugmentation=APR-P2021.08 | 0.931 | — | — | 0.981 | 0.889 | 0.937 | 0.952 | 0.914 | 0.911 | |
| ATDAttack Source=OSAD (OM)2022.09 | 0.927 | — | — | — | — | — | — | — | — | |
| ATDAttack Source=AOE (OM)2022.09 | 0.925 | — | — | — | — | — | — | — | — | |
| MSPpost-hoc=true2021.11 | 0.9246 | 48.73 | — | — | — | — | — | — | — | |
| SupCLR2021.08 | 0.92 | — | — | 0.973 | 0.886 | 0.928 | 0.914 | 0.916 | 0.905 | |
| CSI-ensTraining Mode=Unlabeled multi-class2021.06 | 0.9199 | — | — | — | — | — | — | — | — | |
| CSI-ensTraining Mode=Unlabeled multi-class, Ensemble=true2021.06 | 0.9199 | — | — | — | — | — | — | — | — | |
| Energy ScoreBackbone=WideResNet-402021.10 | 0.919 | — | — | — | — | — | — | — | — | |
| ATDAttack Source=HAT (OM)2022.09 | 0.914 | — | — | — | — | — | — | — | — | |
| Cross Entropyaugmentation=APR-S2021.08 | 0.913 | — | — | 0.904 | 0.868 | 0.961 | 0.942 | 0.909 | 0.891 | |
| ODINBackbone=WideResNet-402021.10 | 0.911 | — | — | — | — | — | — | — | — | |
| Cross Entropyaugmentation=Cutout2021.08 | 0.91 | — | — | 0.936 | 0.864 | 0.945 | 0.902 | 0.922 | 0.89 | |
| MSPBackbone=WideResNet-402021.10 | 0.909 | — | — | — | — | — | — | — | — | |
| G-ODINpost-hoc=false, requires model retraining=true2021.11 | 0.9061 | 34.25 | — | — | — | — | — | — | — | |
| Shifting Transformation Learning (Lssl)Training Mode=Unlabeled multi-class2021.06 | 0.898 | — | — | — | — | — | — | — | — | |
| Shifting Transformation LearningTraining Mode=Unlabeled multi-class, Loss=Lssl2021.06 | 0.898 | — | — | — | — | — | — | — | — | |
| Mahalanobispost-hoc=true2021.11 | 0.8915 | 31.42 | — | — | — | — | — | — | — | |
| Cross Entropy2021.08 | 0.881 | — | — | 0.886 | 0.858 | 0.907 | 0.883 | 0.875 | 0.874 | |
| GeometricTraining Mode=Unlabeled multi-class2021.06 | 0.8604 | — | — | — | — | — | — | — | — | |
| GeometricTraining Mode=Unlabeled multi-class2021.06 | 0.8604 | — | — | — | — | — | — | — | — | |
| SSDTraining Mode=Unlabeled multi-class2021.06 | 0.8454 | — | — | — | — | — | — | — | — | |
| SSDTraining Mode=Unlabeled multi-class2021.06 | 0.8454 | — | — | — | — | — | — | — | — | |
| SimCLRTraining Mode=Unlabeled multi-class2021.06 | 0.7784 | — | — | — | — | — | — | — | — | |
| SimCLRTraining Mode=Unlabeled multi-class2021.06 | 0.7784 | — | — | — | — | — | — | — | — | |
| Cross Entropyaugmentation=Mixup2021.08 | 0.778 | — | — | 0.781 | 0.749 | 0.807 | 0.765 | 0.807 | 0.76 | |
| LEM (S2)score=∥∇Vθ∥2026.05 | 0.641 | — | — | — | — | — | — | — | — | |
| LEM (S1)score=Vθ2026.05 | 0.533 | — | — | — | — | — | — | — | — | |
| EqM dot product2026.05 | 0.49 | — | — | — | — | — | — | — | — | |
| IGEBM2026.05 | 0.48 | — | — | — | — | — | — | — | — | |
| PixelCNN++2026.05 | 0.33 | — | — | — | — | — | — | — | — | |
| LEM (S3)score=flow displacement (τ = 30)2026.05 | 0.301 | — | — | — | — | — | — | — | — | |
| GLOW2026.05 | 0.27 | — | — | — | — | — | — | — | — | |
| Input ComplexityModel class=Diffusion2023.12 | — | — | — | 0.87 | 0.568 | — | — | — | — | |
| Input LikelihoodModel class=Diffusion2023.12 | — | — | — | 0.18 | 0.52 | — | — | — | — | |
| Likelihood RegretModel class=Diffusion2023.12 | — | — | — | 0.904 | 0.546 | — | — | — | — | |
| LMDModel class=Diffusion2023.12 | — | — | — | 0.992 | 0.607 | — | — | — | — | |
| LMDModel class=Consistency2023.12 | — | — | — | 0.979 | 0.62 | 0.734 | 0.686 | — | — | |
| MSMAModel class=Diffusion2023.12 | — | — | — | 0.992 | 0.579 | 0.587 | 0.716 | — | — | |
| MSMAModel class=Consistency2023.12 | — | — | — | 0.707 | 0.57 | 0.605 | 0.578 | — | — | |
| Projection RegretModel class=Consistency2023.12 | — | — | — | 0.993 | 0.775 | 0.837 | 0.814 | — | — |