OOD Detection on ImageNet 100 classes (test)
6.5FPR95Generalized (Neg)
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Generalized (Neg)Backbone=ResNet18-SimCLR, ID Dataset=STL102026.03 | 6.5 | 98.4 | — | — | — | — | — | — | |
| Single (Neg)Backbone=ResNet18-SimCLR, ID Dataset=STL102026.03 | 45.4 | 93.7 | — | — | — | — | — | — | |
| COMBOODBackbone=ResNet18-SimCLR, ID Dataset=STL102026.03 | 58.1 | 78 | — | — | — | — | — | — | |
| D-KNNBackbone=ResNet18-SimCLR, ID Dataset=STL102026.03 | 59.6 | 76 | — | — | — | — | — | — | |
| ET-OOD (Ours)threshold (tau)=0.8, clusters (K)=10242023.03 | 60.17 | 63.91 | 69.55 | 58.23 | 1.08 | 2.16 | 10.56 | 21.34 | |
| Energy (T=1)Backbone=ResNet18-SimCLR, ID Dataset=STL102026.03 | 63.7 | 74.3 | — | — | — | — | — | — | |
| MSPBackbone=ResNet18-SimCLR, ID Dataset=STL102026.03 | 65.7 | 72.5 | — | — | — | — | — | — | |
| CIDERBackbone=ResNet18-SimCLR, ID Dataset=STL102026.03 | 72.5 | 70.1 | — | — | — | — | — | — | |
| Generalized (Neg, InfoNCE)Backbone=ResNet18-SimCLR, ID Dataset=STL102026.03 | 79.3 | 82.2 | — | — | — | — | — | — | |
| ODIN2023.03 | 80.07 | 51.23 | 56.93 | 50.46 | 0.06 | 1.18 | 8.42 | 13.36 | |
| OE2023.03 | 80.52 | 55.87 | 55.93 | 51.94 | 1.06 | 2.63 | 7.67 | 15.13 | |
| UDG2023.03 | 81.89 | 54.74 | 57.85 | 52.53 | 0.95 | 2.06 | 9.18 | 16.35 | |
| EBO2023.03 | 82.77 | 50.17 | 55.31 | 49.84 | 0.49 | 1.49 | 8.87 | 13.59 | |
| KNN (ℓ2)Backbone=ResNet18-SimCLR, ID Dataset=STL102026.03 | 86.7 | 57.2 | — | — | — | — | — | — | |
| MCD2023.03 | 91.04 | 52.26 | 54.8 | 43.92 | 0.08 | 1.92 | 5.57 | 14.35 | |
| MahalanobisBackbone=ResNet18-SimCLR, ID Dataset=STL102026.03 | 92.2 | 49.6 | — | — | — | — | — | — |