Out-of-Distribution Detection on CIFAR-10 (In-distribution) vs SVHN (OOD) (test)
98.3AUROCMMCE
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| MMCEBackbone=ResNet-502026.05 | 98.3 | 12.1 | 6.9 | 98.4 | 98.2 | |
| AUCM + RefCalBackbone=ResNet-502026.05 | 98.2 | 10.2 | 6.6 | 98.5 | 98 | |
| CE+TS + RefCalBackbone=ResNet-502026.05 | 97.9 | 14.1 | 6.8 | 98.2 | 97.6 | |
| NLL (CE) + RefCalBackbone=ResNet-502026.05 | 97.5 | 16.3 | 7.1 | 97.7 | 97 | |
| CRL + RefCalBackbone=ResNet-502026.05 | 97.5 | 16.3 | 7.1 | 97.7 | 97 | |
| MMCE + RefCalBackbone=ResNet-502026.05 | 97.3 | 20.8 | 10.3 | 85.8 | 95.6 | |
| FL+ MDCA + RefCalBackbone=ResNet-502026.05 | 97.3 | 18 | 7.2 | 97.4 | 96.5 | |
| MbLS + RefCalBackbone=ResNet-502026.05 | 97.2 | 16.9 | 7.8 | 97.7 | 96.6 | |
| LS + RefCalBackbone=ResNet-502026.05 | 97.1 | 17 | 8.2 | 97.6 | 96.3 | |
| FL+ MDCABackbone=ResNet-502026.05 | 97.1 | 20.5 | 8 | 97.7 | 96.6 | |
| CE+TSBackbone=ResNet-502026.05 | 97 | 22.2 | 8.1 | 97.6 | 96.4 | |
| MbLSBackbone=ResNet-502026.05 | 96.4 | 25 | 8.2 | 97.1 | 95.1 | |
| CRLBackbone=ResNet-502026.05 | 96.3 | 27.9 | 8.2 | 97.2 | 95.3 | |
| NLL (CE)Backbone=ResNet-502026.05 | 95.9 | 30.5 | 8.8 | 96.9 | 94.7 | |
| L2EOOD score=Maximum Softmax Probability (MSP)2024.08 | 94.4 | — | — | — | — | |
| AUCMBackbone=ResNet-502026.05 | 93.4 | 38.3 | 12.1 | 93.6 | 92.4 | |
| LSBackbone=ResNet-502026.05 | 92.6 | 29.3 | 9.3 | 86.8 | 92.4 | |
| HMCOOD score=Maximum Softmax Probability (MSP)2024.08 | 89.2 | — | — | — | — | |
| CSGMCMCOOD score=Maximum Softmax Probability (MSP)2024.08 | 85.8 | — | — | — | — | |
| DEOOD score=Maximum Softmax Probability (MSP)2024.08 | 83.7 | — | — | — | — |