Out-of-Distribution Detection on OBJECTS vs MNIST 1.0 (test)
92.43AUROCAENIB
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| AENIBScore=log Dir_0.05(y) + log N(zn; 0, I)2023.03 | 92.43 | 89.38 | 48.1 | |
| AENIBScore=log Dir_0.05(y)2023.03 | 90.53 | 85.68 | 52.08 | |
| VIBScore=log Dir_0.05(y)2023.03 | 86.13 | 79.45 | 64.92 | |
| VIBScore=max p(y|x)2023.03 | 80.23 | 73.5 | 80.69 | |
| AENIBScore=max p(y|x)2023.03 | 79.67 | 71.5 | 80.22 | |
| MahalanobisBase model=Cross-entropy2023.03 | 77.04 | 65.31 | 84.59 | |
| Cross-entropyScore=log Dir_0.05(y)2023.03 | 76.75 | 66.26 | 83.51 | |
| SEMBase model=Cross-entropy2023.03 | 75.69 | 76.61 | 99.7 | |
| ODINBase model=Cross-entropy2023.03 | 70.31 | 49.58 | 82.04 | |
| Cross-entropyScore=max p(y|x)2023.03 | 66.98 | 52.66 | 93.54 | |
| Energy-basedBase model=Cross-entropy2023.03 | 54.55 | 34.14 | 92.23 |