Out-of-Distribution Detection on SVHN (ID) vs iSUN (OOD) (test)
99.9TNR @ TPR 95%Mahalanobis
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| MahalanobisBackbone=DenseNet2019.12 | 99.9 | 99.9 | 99.2 | |
| MahalanobisModel Architecture=DenseNet2019.12 | 99.9 | 99.9 | 99.2 | |
| MahalanobisBackbone=ResNet2019.12 | 99.7 | 99.8 | 98.3 | |
| MahalanobisModel Architecture=ResNet2019.12 | 99.7 | 99.8 | 98.3 | |
| OursBackbone=DenseNet2019.12 | 99.4 | 99.8 | 98.3 | |
| OursBackbone=ResNet2019.12 | 99.4 | 99.8 | 98.1 | |
| OursModel Architecture=DenseNet2019.12 | 99.4 | 99.8 | 98.3 | |
| OursModel Architecture=ResNet2019.12 | 99.4 | 99.8 | 98.1 | |
| ODINBackbone=DenseNet2019.12 | 82.2 | 94.7 | 89.7 | |
| ODINModel Architecture=DenseNet2019.12 | 82.2 | 94.7 | 89.7 | |
| ODINBackbone=ResNet2019.12 | 79.1 | 91.4 | 89.2 | |
| ODINModel Architecture=ResNet2019.12 | 79.1 | 91.4 | 89.2 | |
| BaselineBackbone=DenseNet2019.12 | 78.3 | 94.4 | 89.6 | |
| BaselineModel Architecture=DenseNet2019.12 | 78.3 | 94.4 | 89.6 | |
| BaselineBackbone=ResNet2019.12 | 77.1 | 92.2 | 89.7 | |
| BaselineModel Architecture=ResNet2019.12 | 77.1 | 92.2 | 89.7 |