Out-of-Distribution Detection on CIFAR-10 In-Distribution
0.997AUROC (SVHN)Label-Free Mahalanobis
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Label-Free MahalanobisBackbone=DINOv3 ViT-S, Computational cost (#F + #J)=1F + 0J2026.05 | 0.997 | 1 | — | — | |
| ReSCOPEDBackbone=DINOv3 ViT-S, Computational cost (#F + #J)=2F + 1J2026.05 | 0.994 | 0.999 | — | — | |
| LMDComputational cost (#F + #J)=10^4F + 0J2026.05 | 0.992 | 0.557 | — | — | |
| Label-Free MahalanobisBackbone=DINOv3 ViT-B, Computational cost (#F + #J)=1F + 0J2026.05 | 0.99 | 1 | — | — | |
| Label-Free MahalanobisBackbone=DINOv3 ViT-L, Computational cost (#F + #J)=1F + 0J2026.05 | 0.983 | 1 | — | — | |
| ReSCOPEDBackbone=DINOv3 ViT-L, Computational cost (#F + #J)=2F + 1J2026.05 | 0.969 | 0.997 | — | — | |
| ReSCOPEDBackbone=DINOv3 ViT-B, Computational cost (#F + #J)=2F + 1J2026.05 | 0.968 | 0.996 | — | — | |
| MSMAComputational cost (#F + #J)=10F + 0J2026.05 | 0.957 | 1 | — | — | |
| DoSE2026.05 | 0.955 | 0.995 | — | — | |
| Unified Few-shotTraining data regime=Few-shot, samples per task=80-1502026.05 | 0.951 | 0.965 | 0.547 | 0.883 | |
| DiffPathTraining data regime=Full-data, Baselines trained on=SVHN2026.05 | 0.91 | 0.897 | 0.59 | 0.923 | |
| Improved CD2026.05 | 0.91 | — | — | — | |
| DiffPathComputational cost (#F + #J)=10F + 0J2026.05 | 0.91 | 0.897 | — | — | |
| MSMATraining data regime=Full-data, Baselines trained on=SVHN2026.05 | 0.887 | 0.574 | 0.469 | 0.685 | |
| TT2026.05 | 0.87 | 0.848 | — | — | |
| SCOPEDComputational cost (#F + #J)=2F + 2J2026.05 | 0.814 | 0.94 | — | — | |
| IGEBMTraining data regime=Full-data, Baselines trained on=SVHN2026.05 | 0.526 | 0.666 | 0.56 | 0.555 | |
| DDPM-OODComputational cost (#F + #J)=350F + 0J2026.05 | 0.39 | 0.659 | — | — | |
| VAEBMTraining data regime=Full-data, Baselines trained on=SVHN2026.05 | 0.35 | 0.427 | 0.474 | 0.371 | |
| WAIC2026.05 | 0.143 | 0.928 | — | — | |
| NLLComputational cost (#F + #J)=1000F + 0J2026.05 | 0.091 | 0.574 | — | — | |
| LR2026.05 | 0.064 | 0.914 | — | — | |
| DDPM-OODTraining data regime=Full-data, Baselines trained on=SVHN2026.05 | 0.014 | 0.483 | 0.496 | 0.349 | |
| LMDTraining data regime=Full-data, Baselines trained on=SVHN2026.05 | 0.005 | 0.549 | 0.517 | 0.589 | |
| DoSTraining data regime=Full-data, Baselines trained on=SVHN2026.05 | 0.004 | 0.373 | 0.519 | 0.55 | |
| NLLTraining data regime=Full-data, Baselines trained on=SVHN2026.05 | 0.002 | 0.356 | 0.526 | 0.556 |