OOD detection on OOD Suite Average
0.9142AUROCGradPCA
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GradPCABackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 0.9142 | 0.3984 | — | — | |
| DICEBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 0.9055 | 0.4147 | — | — | |
| EnergyBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 0.9022 | 0.4162 | — | — | |
| ReActBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 0.9022 | 0.4129 | — | — | |
| MOSgrouping=taxonomy-based2021.05 | 0.9011 | 0.3997 | 0.9795 | — | |
| Max logitsBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 0.8943 | 0.4575 | — | — | |
| ODINBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 0.8943 | 0.4575 | — | — | |
| KNNBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 0.8837 | 0.4847 | — | — | |
| MahalanobisBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 0.8829 | 0.4694 | — | — | |
| KL Matching2021.05 | 0.8382 | 0.543 | 0.959 | — | |
| MSPBackbone=ViT-B/16, In-Distribution Dataset=ImageNet-1k2025.05 | 0.8379 | 0.5559 | — | — | |
| MM++Backbone=Swin-T, In-Distribution (ID) Dataset=ImageNet-LT2026.06 | 0.8341 | 0.5339 | — | — | |
| Energy2021.05 | 0.8274 | 0.7103 | 0.9642 | — | |
| ODIN2021.05 | 0.8256 | 0.7299 | 0.9637 | — | |
| MM++Backbone=ConvNeXt-T, In-Distribution Dataset=ImageNet-LT2026.06 | 0.8144 | 0.5975 | — | — | |
| rMaha++Backbone=Swin-T, In-Distribution (ID) Dataset=ImageNet-LT2026.06 | 0.8077 | 0.5841 | — | — | |
| Maha++Backbone=Swin-T, In-Distribution (ID) Dataset=ImageNet-LT2026.06 | 0.8045 | 0.5886 | — | — | |
| rMaha++Backbone=ConvNeXt-T, In-Distribution Dataset=ImageNet-LT2026.06 | 0.8 | 0.6357 | — | — | |
| Maha++Backbone=ConvNeXt-T, In-Distribution Dataset=ImageNet-LT2026.06 | 0.7999 | 0.6354 | — | — | |
| MSP2021.05 | 0.7929 | 0.7696 | 0.9537 | — | |
| rMahaBackbone=ConvNeXt-T, In-Distribution Dataset=ImageNet-LT2026.06 | 0.7904 | 0.6685 | — | — | |
| rMahaBackbone=Swin-T, In-Distribution (ID) Dataset=ImageNet-LT2026.06 | 0.7816 | 0.685 | — | — | |
| MahaBackbone=ConvNeXt-T, In-Distribution Dataset=ImageNet-LT2026.06 | 0.7734 | 0.707 | — | — | |
| MSPBackbone=Swin-T, In-Distribution (ID) Dataset=ImageNet-LT2026.06 | 0.7639 | 0.6967 | — | — | |
| MahaBackbone=Swin-T, In-Distribution (ID) Dataset=ImageNet-LT2026.06 | 0.763 | 0.7389 | — | — | |
| ReActBackbone=Swin-T, In-Distribution (ID) Dataset=ImageNet-LT2026.06 | 0.7611 | 0.683 | — | — | |
| MSPBackbone=ConvNeXt-T, In-Distribution Dataset=ImageNet-LT2026.06 | 0.7594 | 0.6808 | — | — | |
| EnergyBackbone=Swin-T, In-Distribution (ID) Dataset=ImageNet-LT2026.06 | 0.7415 | 0.7033 | — | — | |
| ReActBackbone=ConvNeXt-T, In-Distribution Dataset=ImageNet-LT2026.06 | 0.6642 | 0.8463 | — | — | |
| ODINBackbone=Swin-T, In-Distribution (ID) Dataset=ImageNet-LT2026.06 | 0.6587 | 0.8386 | — | — | |
| EnergyBackbone=ConvNeXt-T, In-Distribution Dataset=ImageNet-LT2026.06 | 0.6208 | 0.8635 | — | — | |
| Mahalanobis2021.05 | 0.6202 | 0.8169 | 0.8767 | — | |
| ODINBackbone=ConvNeXt-T, In-Distribution Dataset=ImageNet-LT2026.06 | 0.6151 | 0.8668 | — | — | |
| KNNBackbone=Swin-T, In-Distribution (ID) Dataset=ImageNet-LT2026.06 | 0.5896 | 0.8336 | — | — | |
| KNNBackbone=ConvNeXt-T, In-Distribution Dataset=ImageNet-LT2026.06 | 0.5059 | 0.9405 | — | — | |
| Baseline MSPFPR=5%2026.05 | — | — | — | 60.1 | |
| Combined (KL|Ent)FPR=5%2026.05 | — | — | — | 92 |