Out-of-Distribution Detection on SVHN Near-OOD (AUROC, FPR@95%TPR)
99.3AUROCProj. Regret
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Proj. Regret2026.05 | 99.3 | — | |
| LMD2026.05 | 99.2 | — | |
| k-NNBackbone=DINOv2, k=502026.05 | 97.8 | 15.7 | |
| E1 normed-forkBackbone=CLIP, Feature Dimension=512-d2026.05 | 97.7 | 13.9 | |
| k-NNBackbone=ResNet-50, k=502026.05 | 97 | 16.8 | |
| MahalanobisBackbone=ResNet-50, In-Distribution Dataset=CIFAR-102026.05 | 95.7 | 23.8 | |
| EncMin2L2026.05 | 95.1 | 30.9 | |
| DM Dual Threshold2026.05 | 94.4 | — | |
| VDM log pTBackbone=ResNet-182026.05 | 93.9 | 21.3 | |
| k-NNBackbone=CLIP, k=502026.05 | 93 | 56.4 | |
| Score Network (Hk = 4dk)Hk=4dk2026.05 | 92.7 | 38.8 | |
| MSPBackbone=ResNet-50, In-Distribution Dataset=CIFAR-102026.05 | 91.6 | 54.4 | |
| EnergyBackbone=ResNet-50, In-Distribution Dataset=CIFAR-102026.05 | 91.6 | 40.9 | |
| DiffPath-6DTraining=CelebA-trained, Zero-shot=true2026.05 | 91 | — | |
| E2 raw-forkBackbone=DINOv2, Feature Dimension=768-d2026.05 | 86.8 | 41 | |
| EigenScore2026.05 | 81 | — | |
| Score Network (Hk = 2dk)Hk=2dk2026.05 | 79.9 | 85.9 | |
| E3 normed-forkBackbone=ResNet-50, Feature Dimension=2048-d2026.05 | 76.2 | 81.1 |