OOD Detection on ImageNet far-OOD (Textures)
27.78FPR@95DynProto
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DynProtoBackbone=Swin-B2026.04 | 27.78 | 89.78 | |
| ViMBackbone architectures=Average of ResNet-50, Swin-B, and ViT-B2026.05 | 37.09 | 91.06 | |
| Mahalanobis++Backbone architectures=Average of ResNet-50, Swin-B, and ViT-B2026.05 | 37.62 | 91.9 | |
| MahaVarBackbone architectures=Average of ResNet-50, Swin-B, and ViT-B2026.05 | 38.09 | 91.87 | |
| KNNBackbone architectures=Average of ResNet-50, Swin-B, and ViT-B2026.05 | 39.31 | 91.19 | |
| NNGuideBackbone architectures=Average of ResNet-50, Swin-B, and ViT-B2026.05 | 40.98 | 90.7 | |
| NCI+filterBackbone architectures=Average of ResNet-50, Swin-B, and ViT-B2026.05 | 45.19 | 88.95 | |
| CADRefBackbone=Swin-B2026.04 | 47.38 | 89.08 | |
| fDBDBackbone architectures=Average of ResNet-50, Swin-B, and ViT-B2026.05 | 50.35 | 88.29 | |
| SCALEBackbone architectures=Average of ResNet-50, Swin-B, and ViT-B2026.05 | 51.09 | 81.21 | |
| NCIBackbone architectures=Average of ResNet-50, Swin-B, and ViT-B2026.05 | 51.68 | 87.72 | |
| ReActBackbone architectures=Average of ResNet-50, Swin-B, and ViT-B2026.05 | 52.45 | 86.78 | |
| MahalanobisBackbone architectures=Average of ResNet-50, Swin-B, and ViT-B2026.05 | 52.51 | 88.09 | |
| GENBackbone architectures=Average of ResNet-50, Swin-B, and ViT-B2026.05 | 54.83 | 86 | |
| MaxLogitBackbone architectures=Average of ResNet-50, Swin-B, and ViT-B2026.05 | 57.78 | 80.66 | |
| ReActBackbone=Swin-B2026.04 | 57.83 | 84.07 | |
| EnergyBackbone architectures=Average of ResNet-50, Swin-B, and ViT-B2026.05 | 59 | 78.45 | |
| OptFSBackbone=Swin-B2026.04 | 61.68 | 85.1 | |
| DICEBackbone=Swin-B2026.04 | 63.22 | 73.15 | |
| KL-matchingBackbone architectures=Average of ResNet-50, Swin-B, and ViT-B2026.05 | 63.61 | 83.5 | |
| MSPBackbone architectures=Average of ResNet-50, Swin-B, and ViT-B2026.05 | 63.93 | 80.73 | |
| VIMBackbone=Swin-B2026.04 | 64.64 | 83.96 | |
| EnergyBackbone=Swin-B2026.04 | 65.95 | 69.58 | |
| MSPBackbone=Swin-B2026.04 | 66.2 | 78.46 | |
| ASH-SBackbone architectures=Average of ResNet-50, Swin-B, and ViT-B2026.05 | 70.14 | 50.15 | |
| ASH-SBackbone=Swin-B2026.04 | 98.65 | 18.41 |