Out-of-Distribution Detection on Average (iNaturalist, SUN, Places, Textures)
17.02FPR@95Catalyst(μ) + ReAct
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Catalyst(μ) + ReActBackbone=ResNet-502026.02 | 17.02 | 96.18 | |
| Catalyst(σ) + ReActBackbone=ResNet-502026.02 | 17.46 | 96.02 | |
| Catalyst(m) + ReActBackbone=ResNet-502026.02 | 17.64 | 95.93 | |
| GradOrthBackbone=ResNet-502026.02 | 18.57 | 96.31 | |
| SCALEBackbone=ResNet-502026.02 | 21.89 | 95.32 | |
| ASHBackbone=ResNet-502026.02 | 22.83 | 95.12 | |
| LAPSBackbone=ResNet-502026.02 | 23.68 | 94.78 | |
| ReAct+DICEBackbone=ResNet-502026.02 | 25.41 | 94.1 | |
| NN-GuideBackbone=ResNet-502026.02 | 26.86 | 92.69 | |
| BATSBackbone=ResNet-502026.02 | 27.11 | 94.2 | |
| GradOrthBackbone=MobileNet-v22026.02 | 27.65 | 93.25 | |
| Catalyst(μ)Backbone=ResNet-502026.02 | 28.42 | 93.23 | |
| Catalyst(m) + ReActBackbone=MobileNet-v22026.02 | 29.33 | 93.43 | |
| Catalyst(σ)Backbone=ResNet-502026.02 | 29.75 | 92.92 | |
| Catalyst(m)Backbone=ResNet-502026.02 | 29.89 | 92.82 | |
| Adaptive Multi-prompt Contrastive NetworkShot=82025.06 | 30.56 | 94.29 | |
| ReActBackbone=ResNet-502026.02 | 30.77 | 93.27 | |
| Catalyst(μ) + ReActBackbone=MobileNet-v22026.02 | 30.81 | 93.31 | |
| Adaptive Multi-prompt Contrastive NetworkShot=12025.06 | 30.87 | 92.47 | |
| ReAct+DICEBackbone=MobileNet-v22026.02 | 31.06 | 92.84 | |
| SCTShot=12025.06 | 31.09 | 92.04 | |
| Catalyst(σ) + ReActBackbone=MobileNet-v22026.02 | 31.56 | 92.71 | |
| SCTShot=82025.06 | 32.32 | 93.53 | |
| Catalyst(m)Backbone=MobileNet-v22026.02 | 33.15 | 92.33 | |
| Catalyst(σ)Backbone=MobileNet-v22026.02 | 33.63 | 92.27 | |
| SCALEBackbone=MobileNet-v22026.02 | 34.28 | 92.52 | |
| LAPSBackbone=MobileNet-v22026.02 | 34.99 | 92.03 | |
| GL-MCMShot=02025.06 | 35.59 | 90.83 | |
| DICEBackbone=ResNet-502026.02 | 35.65 | 90.94 | |
| LoCoOpShot=82025.06 | 36.09 | 92.4 | |
| Catalyst(μ)Backbone=MobileNet-v22026.02 | 36.71 | 91.69 | |
| RankFeatBackbone=ResNetv2-101, Pre-training=ImageNet-1k, Feature Layers=Block 3 + 42022.09 | 36.8 | 92.15 | |
| LoCoOpShot=12025.06 | 37.64 | 91.98 | |
| NPOSEvaluation Protocol=fine-tuned2023.03 | 37.93 | 91.22 | |
| NPOSShot=Full2025.06 | 37.94 | 91.22 | |
| ASHBackbone=MobileNet-v22026.02 | 38.68 | 90.95 | |
| CoOpShot=82025.06 | 39.36 | 90.61 | |
| RankFeatBackbone=ResNetv2-101, Pre-training=ImageNet-1k, Feature Layers=Block 42022.09 | 39.69 | 87.84 | |
| GradNormBackbone=ResNet-502026.02 | 40.08 | 87.7 | |
| SeTARShot=02025.06 | 40.24 | 91.05 | |
| ODIN+UMAPBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 40.94 | 89.24 | |
| DICEBackbone=MobileNet-v22026.02 | 41.07 | 89.94 | |
| BATSBackbone=MobileNet-v22026.02 | 41.09 | 90.39 | |
| VOS+Evaluation Protocol=fine-tuned2023.03 | 41.32 | 91.19 | |
| GradNormBackbone=MobileNet-v22026.02 | 41.84 | 89.2 | |
| KNNEvaluation Protocol=fine-tuned2023.03 | 42.19 | 90.97 | |
| KNNShot=Full2025.06 | 42.19 | 90.97 | |
| ODIN+UMBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 42.77 | 86.25 | |
| MCMShot=02025.06 | 43 | 90.65 | |
| VOSEvaluation Protocol=fine-tuned2023.03 | 43.24 | 90.86 | |
| MCMEvaluation Protocol=zero-shot2023.03 | 43.55 | 90.62 | |
| CoOpShot=12025.06 | 44.91 | 90.07 | |
| RankFeatBackbone=ResNetv2-101, Pre-training=ImageNet-1k, Feature Layers=Block 32022.09 | 45.8 | 90.8 | |
| Energy+UMBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 47.29 | 84.72 | |
| EnergyEvaluation Protocol=fine-tuned2023.03 | 47.67 | 88.9 | |
| ODINEvaluation Protocol=fine-tuned2023.03 | 47.75 | 88.8 | |
| ODINShot=Full2025.06 | 47.75 | 88.8 | |
| ReActBackbone=MobileNet-v22026.02 | 48.91 | 88.75 | |
| MSP+UMBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 49.76 | 87.07 | |
| ViMEvaluation Protocol=fine-tuned2023.03 | 50.2 | 87.82 | |
| ViMShot=Full2025.06 | 50.2 | 87.82 | |
| fDBDBackbone=ResNet-502026.02 | 51.19 | 89.26 | |
| ODINBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 51.29 | 84.75 | |
| MSP+UMAPBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 53 | 87.24 | |
| KNN (α = 100%)Backbone=ResNet-502026.02 | 53.97 | 85.01 | |
| ODINBackbone=MobileNet-v22026.02 | 54.2 | 85.81 | |
| KNN (α = 1%)Backbone=ResNet-502026.02 | 54.32 | 84.59 | |
| GradNormBackbone=ResNetv2-101, Pre-training=ImageNet-1k2022.09 | 54.7 | 86.71 | |
| Energy+UMAPBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 54.96 | 85.26 | |
| MSPBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 55.13 | 85.15 | |
| ODINBackbone=ResNet-502026.02 | 56.48 | 85.41 | |
| EnergyBackbone=ResNet-50, Pre-trained=PyTorch2023.06 | 57.45 | 82.78 | |
| EnergyBackbone=ResNet-502026.02 | 57.48 | 87.05 | |
| ReActBackbone=ResNetv2-101, Pre-training=ImageNet-1k2022.09 | 57.66 | 86.67 | |
| EnergyBackbone=MobileNet-v22026.02 | 58.87 | 86.59 | |
| ViMBackbone=ResNet-502026.02 | 62.91 | 85.93 | |
| MSPBackbone=ResNet-502026.02 | 64.76 | 82.82 | |
| MahalanobisBackbone=MobileNet-v22026.02 | 64.83 | 72.4 | |
| GODINBackbone=ResNet-502026.02 | 66.07 | 82.02 | |
| NN-GuideBackbone=MobileNet-v22026.02 | 67.12 | 80.43 | |
| MSPEvaluation Protocol=fine-tuned2023.03 | 67.31 | 80.64 | |
| MSPBackbone=MobileNet-v22026.02 | 70.49 | 80.67 | |
| EnergyBackbone=ResNetv2-101, Pre-training=ImageNet-1k2022.09 | 71.03 | 82.74 | |
| ODINBackbone=ResNetv2-101, Pre-training=ImageNet-1k2022.09 | 72.99 | 82.56 | |
| MSPBackbone=ResNetv2-101, Pre-training=ImageNet-1k2022.09 | 76.96 | 79.29 | |
| ViMBackbone=MobileNet-v22026.02 | 77.1 | 71.99 | |
| GradNormEvaluation Protocol=fine-tuned2023.03 | 80.82 | 72.35 | |
| MahalanobisBackbone=ResNetv2-101, Pre-training=ImageNet-1k2022.09 | 81.69 | 62.02 | |
| MahalanobisBackbone=ResNet-502026.02 | 87.43 | 55.47 |