Out-of-Distribution Detection on ImageNet-1k vs OpenImage-O
95.58AUROC (%)DINOv2+MLS
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DINOv2+MLSBackbone=DINOv2, Training=Linear Probe, Postprocessor=MLS2024.10 | 95.58 | 16.88 | |
| ViT-B+CE+RMDSBackbone=ViT-B, Training=Cross Entropy (CE), Postprocessor=RMDS2024.10 | 92.32 | 29.57 | |
| ViMBackbone=Swin-T, ID Accuracy=81.60%2026.04 | 92.29 | 26.62 | |
| RMDSBackbone=Swin-T, ID Accuracy=81.60%2026.04 | 92.1 | 29.34 | |
| Energy Profile Divergence (EPD)Backbone=Swin-T, ID Accuracy=81.60%2026.04 | 91.71 | 27.9 | |
| MDSBackbone=Swin-T, ID Accuracy=81.60%2026.04 | 90.98 | 30.89 | |
| GENBackbone=Swin-T, ID Accuracy=81.60%2026.04 | 90.6 | 32.77 | |
| FDBDBackbone=ResNet-502026.02 | 90 | 57 | |
| ReActBackbone=ResNet-502026.02 | 89.5 | 55.4 | |
| KNNBackbone=Swin-T, ID Accuracy=81.60%2026.04 | 88.62 | 36.3 | |
| ViMBackbone=ResNet-502026.02 | 88.5 | 59.8 | |
| SHEBackbone=Swin-T, ID Accuracy=81.60%2026.04 | 88.04 | 52.88 | |
| ReActBackbone=Swin-T, ID Accuracy=81.60%2026.04 | 87.85 | 42.54 | |
| OpenMaxBackbone=Swin-T, ID Accuracy=81.60%2026.04 | 87.75 | 60.88 | |
| MaxLogitBackbone=ResNet-502026.02 | 87.2 | 64.7 | |
| EnergyBackbone=ResNet-502026.02 | 86.9 | 65.3 | |
| TempScaleBackbone=Swin-T, ID Accuracy=81.60%2026.04 | 86.2 | 50.3 | |
| MSPBackbone=Swin-T, ID Accuracy=81.60%2026.04 | 85.81 | 48.97 | |
| MLSBackbone=Swin-T, ID Accuracy=81.60%2026.04 | 83.93 | 73.96 | |
| D-KNNBackbone=ResNet-502026.02 | 83.9 | 63.1 | |
| MSPBackbone=ResNet-502026.02 | 83.8 | 68.2 | |
| KNNBackbone=ResNet-502026.02 | 82.3 | 65.9 | |
| EBOBackbone=Swin-T, ID Accuracy=81.60%2026.04 | 80.24 | 80.2 | |
| NecoBackbone=ResNet-502026.02 | 77.3 | 72 | |
| WDiscOODBackbone=ResNet-502026.02 | 72 | 83.9 | |
| MahalanobisBackbone=ResNet-502026.02 | 71.3 | 85.7 | |
| DICEBackbone=Swin-T, ID Accuracy=81.60%2026.04 | 58.67 | 96.36 | |
| ASHBackbone=Swin-T, ID Accuracy=81.60%2026.04 | 45.22 | 93.99 | |
| GradNormBackbone=Swin-T, ID Accuracy=81.60%2026.04 | 34 | 96.83 |