OOD Detection on CIFAR-100 vs Places365 (test)
95.85AUROCHopfield Boosting
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Hopfield BoostingBackbone=ResNet-182024.05 | 95.85 | 19.36 | |
| POEMBackbone=ResNet-182024.05 | 95.03 | 18.39 | |
| DOSBackbone=ResNet-182024.05 | 91.73 | 32.13 | |
| EBO-OEBackbone=ResNet-182024.05 | 91.35 | 26.68 | |
| NGCSupervised detection=false, IND noise level=50% sym., OOD samples in training set=20k, OOD samples in test set=10k2021.08 | 91.2 | — | |
| DALBackbone=ResNet-182024.05 | 91.1 | 33.43 | |
| ODINSupervised detection=true, IND noise level=50% sym., OOD samples in training set=20k, OOD samples in test set=10k2021.08 | 89.93 | — | |
| MixOEBackbone=ResNet-182024.05 | 89.2 | 47.01 | |
| DivOEBackbone=ResNet-182024.05 | 88.28 | 44.2 | |
| MSP-OEBackbone=ResNet-182024.05 | 87.77 | 45.96 | |
| MSPSupervised detection=true, IND noise level=50% sym., OOD samples in training set=20k, OOD samples in test set=10k2021.08 | 85.82 | — | |
| DOEBackbone=ResNet-182024.05 | 83.47 | 58.68 | |
| DREAM-OOD2023.09 | 79.94 | 70.85 | |
| KNN2023.09 | 78.21 | 79.62 | |
| Mahalanobis2023.09 | 77.9 | 76 | |
| ViM2023.09 | 77.81 | 79.2 | |
| GODIN2023.09 | 77.19 | 80.65 | |
| MSP2023.09 | 76.71 | 81.65 | |
| ReAct2023.09 | 76.25 | 81.75 | |
| SSDSupervised detection=true, IND noise level=50% sym., OOD samples in training set=20k, OOD samples in test set=10k2021.08 | 76.16 | — | |
| Energy2023.09 | 76 | 82.05 | |
| DICE2023.09 | 75.92 | 85.05 | |
| VOS2023.09 | 75.85 | 84.55 | |
| ODIN2023.09 | 74.87 | 79.3 | |
| NPOS2023.09 | 71.3 | 79.08 | |
| SSDSupervised detection=false, IND noise level=50% sym., OOD samples in training set=20k, OOD samples in test set=10k2021.08 | 68.97 | — | |
| MDSupervised detection=true, IND noise level=50% sym., OOD samples in training set=20k, OOD samples in test set=10k2021.08 | 68.08 | — | |
| GAN2023.09 | 66.76 | 88.75 | |
| RotSupervised detection=true, IND noise level=50% sym., OOD samples in training set=20k, OOD samples in test set=10k2021.08 | 59.9 | — | |
| RotSupervised detection=false, IND noise level=50% sym., OOD samples in training set=20k, OOD samples in test set=10k2021.08 | 44.85 | — | |
| CIDERlayer=penultimate2023.10 | — | 79.63 | |
| CIDERrepresentation=hypersphere2023.10 | — | 89.92 | |
| CIDER (hypersphere) + Diffusionmodel=Riemannian Diffusion Model2023.10 | — | 76.54 | |
| KNN+2023.10 | — | 80.74 | |
| SSD+2023.10 | — | 77.74 |