Out-of-Distribution Detection on OBJECTS vs FashionMNIST (test)
84.85AUROCAENIB
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| AENIBScore=log Dir_0.05(y) + log N(zn; 0, I)2023.03 | 84.85 | 94.84 | 74.67 | |
| AENIBScore=log Dir_0.05(y)2023.03 | 84.56 | 94.61 | 74.24 | |
| Cross-entropyScore=log Dir_0.05(y)2023.03 | 82.88 | 93.97 | 77.19 | |
| VIBScore=log Dir_0.05(y)2023.03 | 81.11 | 93.12 | 77.82 | |
| ODINBase model=Cross-entropy2023.03 | 80.98 | 91.53 | 68.73 | |
| MahalanobisBase model=Cross-entropy2023.03 | 80.33 | 92.28 | 77.17 | |
| SEMBase model=Cross-entropy2023.03 | 79.4 | 93.14 | 93.72 | |
| AENIBScore=max p(y|x)2023.03 | 77.33 | 91.63 | 84.31 | |
| Energy-basedBase model=Cross-entropy2023.03 | 76.5 | 89.8 | 72.4 | |
| VIBScore=max p(y|x)2023.03 | 76.35 | 91.22 | 84.75 | |
| Cross-entropyScore=max p(y|x)2023.03 | 73.87 | 90.15 | 88.08 |