Multi-class OOD detection on CIFAR-100 (test) vs SVHN, CIFAR-10, LSUN
98.2OOD Accuracy (SVHN)SSD+
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| SSD+2023.02 | 98.2 | 78.3 | 79.8 | 85.4 | |
| Residual Flows2023.02 | 97.5 | 77.1 | — | 87.3 | |
| MOOD2023.02 | 96.5 | 98.3 | 96.3 | 97 | |
| Gram Matrix2023.02 | 96 | 67.9 | — | 82 | |
| Contrastive loss2023.02 | 95.6 | 78.3 | — | 87 | |
| Mahalanobis2023.02 | 94.5 | 55.3 | — | 74.9 | |
| Outlier exposure2023.02 | 86.9 | 75.7 | — | 81.3 | |
| Baseline OOD2023.02 | 81.9 | 81.1 | 86.6 | 83.2 | |
| ODIN2023.02 | 60.9 | 77.9 | — | 69.4 |