Out-of-Distribution Detection on SUN
15.81FPR@95LAPS
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LAPSBackbone=ResNet-502026.02 | 15.81 | 96.18 | — | |
| Catalyst(μ) + ReActBackbone=ResNet-502026.02 | 18.46 | 95.82 | — | |
| Catalyst(m) + ReActBackbone=ResNet-502026.02 | 19.02 | 95.52 | — | |
| Catalyst(σ) + ReActBackbone=ResNet-502026.02 | 19.13 | 95.61 | — | |
| GradOrthBackbone=ResNet-502026.02 | 19.61 | 95.76 | — | |
| DynProtoBackbone=ConvNeXt-B2026.04 | 21.05 | 94.26 | — | |
| DynProtoBackbone=ViT-B/162026.04 | 21.55 | 93.76 | — | |
| BATSBackbone=ResNet-502026.02 | 22.62 | 95.33 | — | |
| Adaptive Multi-prompt Contrastive NetworkShot=82025.06 | 23.17 | 95.89 | — | |
| Adaptive Multi-prompt Contrastive NetworkShot=12025.06 | 23.26 | 94.85 | — | |
| SCTShot=82025.06 | 23.48 | 94.77 | — | |
| SCTShot=12025.06 | 23.52 | 94.58 | — | |
| ReActBackbone=ResNet-502026.02 | 23.68 | 94.44 | — | |
| ReAct+DICEBackbone=ResNet-502026.02 | 24.05 | 94.31 | — | |
| SCALEBackbone=ResNet-502026.02 | 25.78 | 94.54 | — | |
| RankFeatBackbone=ResNetv2-101, Pre-training=ImageNet-1k, Feature Layers=Block 42022.09 | 27.88 | 92.18 | — | |
| ASHBackbone=ResNet-502026.02 | 28.01 | 94.02 | — | |
| RankFeatBackbone=ResNetv2-101, Pre-training=ImageNet-1k, Feature Layers=Block 3 + 42022.09 | 29.27 | 94.07 | — | |
| LAPSBackbone=MobileNet-v22026.02 | 30.07 | 92.98 | — | |
| GL-MCMShot=02025.06 | 30.42 | 93.09 | — | |
| ReAct+DICEBackbone=MobileNet-v22026.02 | 30.6 | 92.98 | — | |
| Catalyst(μ)Backbone=ResNet-502026.02 | 30.79 | 92.67 | — | |
| GradOrthBackbone=MobileNet-v22026.02 | 30.82 | 93.18 | — | |
| NN-GuideBackbone=ResNet-502026.02 | 31.62 | 91.66 | — | |
| Catalyst(μ) + ReActBackbone=MobileNet-v22026.02 | 32.82 | 92.93 | — | |
| LoCoOpShot=12025.06 | 33.27 | 93.67 | — | |
| LoCoOpShot=82025.06 | 33.87 | 93.23 | — | |
| CoOpShot=82025.06 | 34.45 | 92.5 | — | |
| Catalyst(m) + ReActBackbone=MobileNet-v22026.02 | 34.77 | 92.26 | — | |
| SeTARShot=02025.06 | 35.57 | 92.79 | — | |
| KNNEvaluation Protocol=fine-tuned2023.03 | 35.62 | 92.67 | — | |
| KNNShot=Full2025.06 | 35.62 | 92.67 | — | |
| Catalyst(σ)Backbone=ResNet-502026.02 | 35.73 | 91.47 | — | |
| Catalyst(m)Backbone=ResNet-502026.02 | 35.79 | 91.4 | — | |
| DICEBackbone=ResNet-502026.02 | 36.11 | 91.01 | — | |
| VOS+Evaluation Protocol=fine-tuned2023.03 | 36.88 | 92.57 | — | |
| Catalyst(m)Backbone=MobileNet-v22026.02 | 37.41 | 91.37 | — | |
| Catalyst(σ) + ReActBackbone=MobileNet-v22026.02 | 37.53 | 91.22 | — | |
| MCMShot=02025.06 | 37.67 | 92.56 | — | |
| Catalyst(μ)Backbone=MobileNet-v22026.02 | 37.74 | 91.43 | — | |
| DICEBackbone=MobileNet-v22026.02 | 37.84 | 90.81 | — | |
| Catalyst(σ)Backbone=MobileNet-v22026.02 | 38.2 | 91.26 | — | |
| CoOpShot=12025.06 | 38.53 | 91.95 | — | |
| SCALEBackbone=MobileNet-v22026.02 | 38.74 | 91.64 | — | |
| MCMEvaluation Protocol=zero-shot2023.03 | 39.21 | 92.28 | — | |
| RankFeatBackbone=ResNetv2-101, Pre-training=ImageNet-1k, Feature Layers=Block 32022.09 | 39.91 | 92.01 | — | |
| BATSBackbone=MobileNet-v22026.02 | 41.68 | 90.21 | — | |
| GradNormBackbone=MobileNet-v22026.02 | 42.15 | 89.65 | — | |
| GradNormBackbone=ResNet-502026.02 | 42.81 | 87.26 | — | |
| VOSEvaluation Protocol=fine-tuned2023.03 | 43.03 | 91.92 | — | |
| ASHBackbone=MobileNet-v22026.02 | 43.63 | 90.02 | — | |
| NPOSEvaluation Protocol=fine-tuned2023.03 | 43.77 | 90.44 | — | |
| NPOSShot=Full2025.06 | 43.77 | 90.44 | — | |
| GradNormBackbone=ResNetv2-101, Pre-training=ImageNet-1k2022.09 | 46.48 | 89.03 | — | |
| ReActBackbone=MobileNet-v22026.02 | 52.46 | 87.26 | — | |
| ReActBackbone=ResNetv2-101, Pre-training=ImageNet-1k2022.09 | 52.71 | 90.16 | — | |
| EnergyEvaluation Protocol=fine-tuned2023.03 | 53.18 | 87.33 | — | |
| ViMEvaluation Protocol=fine-tuned2023.03 | 54.01 | 87.19 | — | |
| ViMShot=Full2025.06 | 54.01 | 87.19 | — | |
| ODINEvaluation Protocol=fine-tuned2023.03 | 54.04 | 87.17 | — | |
| ODINShot=Full2025.06 | 54.04 | 87.17 | — | |
| ODINBackbone=MobileNet-v22026.02 | 54.07 | 85.88 | — | |
| MahalanobisBackbone=MobileNet-v22026.02 | 54.79 | 86.33 | — | |
| VIMBackbone=ViT-B/162026.04 | 56.77 | 83.99 | — | |
| EnergyBackbone=ResNet-502026.02 | 58.28 | 86.73 | — | |
| EnergyBackbone=MobileNet-v22026.02 | 59.36 | 86.24 | — | |
| VIMBackbone=ConvNeXt-B2026.04 | 59.57 | 85 | — | |
| ODINBackbone=ResNet-502026.02 | 60.15 | 84.59 | — | |
| CADRefBackbone=ConvNeXt-B2026.04 | 60.16 | 86.84 | — | |
| fDBDBackbone=ResNet-502026.02 | 60.6 | 86.97 | — | |
| GODINBackbone=ResNet-502026.02 | 60.83 | 85.6 | — | |
| OptFSBackbone=ConvNeXt-B2026.04 | 61.69 | 85.96 | — | |
| MSPBackbone=ConvNeXt-B2026.04 | 63.37 | 79.28 | — | |
| EnergyBackbone=ResNetv2-101, Pre-training=ImageNet-1k2022.09 | 65.33 | 85.32 | — | |
| CADRefBackbone=ViT-B/162026.04 | 65.85 | 85.12 | — | |
| OptFSBackbone=ViT-B/162026.04 | 66.36 | 84.19 | — | |
| VOSBackbone=ResNet-502026.01 | 66.42 | — | 39.26 | |
| MSPBackbone=ViT-B/162026.04 | 66.65 | 80.88 | — | |
| OTISBackbone=ResNet-502026.01 | 68.17 | — | 38.8 | |
| CEDABackbone=ResNet-502026.01 | 68.36 | — | 46.25 | |
| MSPBackbone=ResNet-502026.02 | 68.58 | 81.75 | — | |
| BaselineBackbone=ResNet-502026.01 | 68.58 | — | 46.15 | |
| KNN (α = 100%)Backbone=ResNet-502026.02 | 68.82 | 80.72 | — | |
| KNN (α = 1%)Backbone=ResNet-502026.02 | 69.53 | 80.1 | — | |
| CODESBackbone=ResNet-502026.01 | 70.23 | — | 39.36 | |
| ACETBackbone=ResNet-502026.01 | 71.02 | — | 46.18 | |
| ODINBackbone=ResNetv2-101, Pre-training=ImageNet-1k2022.09 | 71.67 | 83.92 | — | |
| ReActBackbone=ViT-B/162026.04 | 72.38 | 78.93 | — | |
| EnergyBackbone=ViT-B/162026.04 | 72.81 | 70.16 | — | |
| MSPEvaluation Protocol=fine-tuned2023.03 | 73.37 | 78.03 | — | |
| MSPBackbone=MobileNet-v22026.02 | 74.2 | 78.88 | — | |
| ReActBackbone=ConvNeXt-B2026.04 | 78.02 | 78.51 | — | |
| NN-GuideBackbone=MobileNet-v22026.02 | 79.57 | 76.1 | — | |
| MSPBackbone=ResNetv2-101, Pre-training=ImageNet-1k2022.09 | 79.89 | 78.34 | — | |
| ViMBackbone=ResNet-502026.02 | 81.79 | 81.07 | — | |
| GradNormEvaluation Protocol=fine-tuned2023.03 | 82 | 72.86 | — | |
| MahalanobisBackbone=ResNetv2-101, Pre-training=ImageNet-1k2022.09 | 88.43 | 65.2 | — | |
| ViMBackbone=MobileNet-v22026.02 | 88.67 | 66.37 | — | |
| DICEBackbone=ConvNeXt-B2026.04 | 93.95 | 41.27 | — | |
| DICEBackbone=ViT-B/162026.04 | 94.51 | 65.23 | — |