Out-of-Distribution Detection on iNaturalist (test)
98.15AUROCMOS
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| MOSBackbone=BiT-S-R101x1, In-distribution dataset=ImageNet-1k, Test Time (min)=3.2, Fine-tuned=true2021.05 | 98.15 | 9.28 | — | — | |
| KL MatchingBackbone=BiT-S-R101x1, In-distribution dataset=ImageNet-1k, Test Time (min)=20.6, Fine-tuned=true2021.05 | 93 | 27.36 | — | — | |
| OCOBackbone=DINOv2, Training Dataset=ImageNet-1k2026.05 | 90.75 | 28.03 | — | — | |
| ReActBackbone=T2T-ViT-242022.09 | 89.51 | 52.17 | — | — | |
| ODINBackbone=BiT-S-R101x1, In-distribution dataset=ImageNet-1k, Test Time (min)=23.6, Fine-tuned=true2021.05 | 89.36 | 62.69 | — | — | |
| ASH2026.03 | 89.23 | — | 36.82 | 43.79 | |
| MSPBackbone=T2T-ViT-242022.09 | 88.95 | 48.92 | — | — | |
| EnergyBackbone=BiT-S-R101x1, In-distribution dataset=ImageNet-1k, Test Time (min)=3.1, Fine-tuned=true2021.05 | 88.48 | 64.91 | — | — | |
| CoRPBackbone=DINOv2, Training Dataset=ImageNet-1k2026.05 | 88.34 | 29.7 | — | — | |
| ODINBackbone=T2T-ViT-242022.09 | 88.17 | 44.07 | — | — | |
| ReAct2026.03 | 87.93 | — | 40.64 | 39.6 | |
| RankFeatBackbone=T2T-ViT-242022.09 | 87.81 | 50.37 | — | — | |
| GradNorm2026.03 | 87.62 | — | 45.63 | 39.57 | |
| MSPBackbone=BiT-S-R101x1, In-distribution dataset=ImageNet-1k, Test Time (min)=3.1, Fine-tuned=true2021.05 | 87.59 | 63.69 | — | — | |
| FDBDBackbone=DINOv2, Training Dataset=ImageNet-1k2026.05 | 87.59 | 31.54 | — | — | |
| NECOBackbone=DINOv2, Training Dataset=ImageNet-1k2026.05 | 86.9 | 31.96 | — | — | |
| MaxLogitBackbone=DINOv2, Training Dataset=ImageNet-1k2026.05 | 86.53 | 32.99 | — | — | |
| SCALEBackbone=DINOv2, Training Dataset=ImageNet-1k2026.05 | 85.78 | 33.23 | — | — | |
| EnergyBackbone=DINOv2, Training Dataset=ImageNet-1k2026.05 | 85.6 | 33.84 | — | — | |
| FDBDBackbone=ViT, Training Dataset=ImageNet-1k2026.05 | 85.37 | 39.01 | — | — | |
| SHE2026.03 | 84.71 | — | 50.97 | 32.37 | |
| CoRPBackbone=ViT, Training Dataset=ImageNet-1k2026.05 | 84.34 | 40.1 | — | — | |
| RankFeatBackbone=SqueezeNet, Block=Block 42022.09 | 83.09 | 61.67 | — | — | |
| RankFeatBackbone=SqueezeNet, Block=Block 3 + 42022.09 | 83.06 | 65.81 | — | — | |
| EnergyBackbone=T2T-ViT-242022.09 | 82.93 | 52.95 | — | — | |
| RAS2026.03 | 82.69 | — | 42.11 | 25.68 | |
| DICE2026.03 | 82.52 | — | 54.89 | 29.47 | |
| RAS HL2026.03 | 82.26 | — | 42.47 | 24.66 | |
| NECOBackbone=ViT, Training Dataset=ImageNet-1k2026.05 | 81.85 | 47.47 | — | — | |
| MaxLogitBackbone=ViT, Training Dataset=ImageNet-1k2026.05 | 81.34 | 47.75 | — | — | |
| OCOBackbone=ViT, Training Dataset=ImageNet-1k2026.05 | 81.34 | 49.69 | — | — | |
| RankFeatBackbone=SqueezeNet, Block=Block 32022.09 | 81.3 | 71.04 | — | — | |
| SCALEBackbone=ViT, Training Dataset=ImageNet-1k2026.05 | 81.22 | 45.76 | — | — | |
| KLM2026.03 | 80.88 | — | 58.6 | 27.91 | |
| EnergyBackbone=ViT, Training Dataset=ImageNet-1k2026.05 | 80.43 | 48.61 | — | — | |
| OpenMax2026.03 | 80.3 | — | 49.98 | 25.64 | |
| OODDBackbone=DINOv2, Training Dataset=ImageNet-1k2026.05 | 79 | 41.53 | — | — | |
| GEN2026.03 | 78.47 | — | 50.61 | 24.53 | |
| ODIN2026.03 | 77.43 | — | 57.1 | 19.81 | |
| NNguideBackbone=DINOv2, Training Dataset=ImageNet-1k2026.05 | 77.38 | 44.57 | — | — | |
| TempScale2026.03 | 77.33 | — | 58.16 | 20.64 | |
| NNguideBackbone=ViT, Training Dataset=ImageNet-1k2026.05 | 76.9 | 59.54 | — | — | |
| MSP2026.03 | 76.35 | — | 61.75 | 20.9 | |
| MLS2026.03 | 75.61 | — | 54.59 | 18.04 | |
| OODDBackbone=ViT, Training Dataset=ImageNet-1k2026.05 | 75.22 | 54.45 | — | — | |
| EBO2026.03 | 74.01 | — | 55.18 | 16.43 | |
| GradNormBackbone=SqueezeNet2022.09 | 73.92 | 76.31 | — | — | |
| RMDS2026.03 | 73.48 | — | 55.68 | 16.41 | |
| EnergyBackbone=SqueezeNet2022.09 | 73.3 | 79.27 | — | — | |
| SHEBackbone=ViT, Training Dataset=ImageNet-1k2026.05 | 73.09 | 71.73 | — | — | |
| VIM2026.03 | 72.22 | — | 55.59 | 16 | |
| Gram2026.03 | 71.36 | — | 72.69 | 17.33 | |
| ReActBackbone=SqueezeNet2022.09 | 68.56 | 76.78 | — | — | |
| KNN2026.03 | 67.79 | — | 62.46 | 13.53 | |
| ODINBackbone=SqueezeNet2022.09 | 65.75 | 90.79 | — | — | |
| MSPBackbone=SqueezeNet2022.09 | 65.41 | 89.83 | — | — | |
| SHEBackbone=DINOv2, Training Dataset=ImageNet-1k2026.05 | 64.77 | 74.72 | — | — | |
| MahalanobisBackbone=T2T-ViT-242022.09 | 58.13 | 90.5 | — | — | |
| MDS2026.03 | 54.06 | — | 83.38 | 9.86 | |
| MDSEns2026.03 | 53.32 | — | 87.47 | 9.26 | |
| MahalanobisBackbone=SqueezeNet2022.09 | 51.79 | 91.5 | — | — | |
| MahalanobisBackbone=BiT-S-R101x1, In-distribution dataset=ImageNet-1k, Test Time (min)=145.4, Fine-tuned=true2021.05 | 46.33 | 96.34 | — | — | |
| RankFeat2026.03 | 34.34 | — | 96.43 | 6.82 | |
| GradNormBackbone=T2T-ViT-242022.09 | 25.86 | 99.3 | — | — | |
| KNN (non-parametric)Backbone=ViT-B/16, Fine-tuned on=ImageNet-1k2022.04 | — | 7.3 | — | — | |
| Mahalanobis (parametric)Backbone=ViT-B/16, Fine-tuned on=ImageNet-1k2022.04 | — | 17.56 | — | — |