Out-of-Distribution Detection on SSB hard
95.8AUROC (%)DisCoPatch-64
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DisCoPatch-64Number of Patches=642025.01 | 95.8 | 19.8 | — | |
| RankFeatBackbone=ResNetv2-1012025.01 | 89.4 | 47.9 | — | |
| MOODv2Backbone=BEiTv22025.01 | 85 | 58.1 | — | |
| NNGuideBackbone=RegNet2025.01 | 84.7 | 54.7 | — | |
| SCALEBackbone=ResNet-502025.01 | 77.4 | 67.7 | — | |
| WeiPer+KLDBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 74.73 | 74.12 | — | |
| ASHBackbone=ResNet-502025.01 | 74.1 | 80.6 | — | |
| ReActBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 73.03 | 77.55 | — | |
| WeiPer+MSPBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 73.01 | 75.16 | — | |
| ASHBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 72.89 | 73.66 | — | |
| TempScaleBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 72.87 | 73.9 | — | |
| RMDSBackbone=ViT16-B, Training Objective=Cross Entropy, Preprocessing=standard2024.05 | 72.87 | 84.52 | — | |
| MLSBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 72.51 | 76.2 | — | |
| MSPBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 72.09 | 74.49 | — | |
| EBOBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 72.08 | 76.54 | — | |
| GENBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 72.01 | 75.73 | — | |
| GradNormBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 71.9 | 78.24 | — | |
| RMDSBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 71.77 | 77.88 | — | |
| ODINBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 71.74 | 76.83 | — | |
| MDSBackbone=ViT16-B, Training Objective=Cross Entropy, Preprocessing=standard2024.05 | 71.57 | 83.47 | — | |
| NNGuideBackbone=ResNet-502025.01 | 71.5 | 82.8 | — | |
| KLMBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 71.38 | 84.71 | — | |
| OpenMaxBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 71.37 | 77.33 | — | |
| WeiPer+ReActBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 71.2 | 80.39 | — | |
| SHEBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 71.08 | 76.3 | — | |
| DICEBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 70.13 | 77.96 | — | |
| GENBackbone=ViT16-B, Training Objective=Cross Entropy, Preprocessing=standard2024.05 | 70.09 | 82.23 | — | |
| VIMBackbone=ViT16-B, Training Objective=Cross Entropy, Preprocessing=standard2024.05 | 69.42 | 90.04 | — | |
| WeiPer+MSPBackbone=ViT16-B, Training Objective=Cross Entropy, Preprocessing=standard2024.05 | 68.98 | 85.09 | — | |
| MSPBackbone=ViT16-B, Training Objective=Cross Entropy, Preprocessing=standard2024.05 | 68.94 | 86.41 | — | |
| OpenMaxBackbone=ViT16-B, Training Objective=Cross Entropy, Preprocessing=standard2024.05 | 68.6 | 89.19 | — | |
| TempScaleBackbone=ViT16-B, Training Objective=Cross Entropy, Preprocessing=standard2024.05 | 68.55 | 87.35 | — | |
| WeiPer+ReActBackbone=ViT16-B, Training Objective=Cross Entropy, Preprocessing=standard2024.05 | 68.52 | 85.48 | — | |
| FDBDBackbone=ResNet-502025.01 | 68.4 | 83.4 | — | |
| WeiPer+KLDBackbone=ViT16-B, Training Objective=Cross Entropy, Preprocessing=standard2024.05 | 68.26 | 85.6 | — | |
| KLMBackbone=ViT16-B, Training Objective=Cross Entropy, Preprocessing=standard2024.05 | 68.14 | 88.35 | — | |
| SHEBackbone=ViT16-B, Training Objective=Cross Entropy, Preprocessing=standard2024.05 | 68.04 | 85.73 | — | |
| KNNBackbone=ViT16-B, Training Objective=Cross Entropy, Preprocessing=standard2024.05 | 65.98 | 86.22 | — | |
| VIMBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 65.54 | 80.41 | — | |
| NNGuideBackbone=MobileNetV22025.01 | 64.3 | 90.3 | — | |
| MLSBackbone=ViT16-B, Training Objective=Cross Entropy, Preprocessing=standard2024.05 | 64.2 | 91.52 | — | |
| ReActBackbone=ViT16-B, Training Objective=Cross Entropy, Preprocessing=standard2024.05 | 63.1 | 90.46 | — | |
| KNNBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 62.57 | 83.36 | — | |
| GramIn-Distribution Dataset=ImageNet-2002026.03 | 59.12 | 86.37 | 58.06 | |
| DICEBackbone=ViT16-B, Training Objective=Cross Entropy, Preprocessing=standard2024.05 | 59.05 | 89.77 | — | |
| EBOBackbone=ViT16-B, Training Objective=Cross Entropy, Preprocessing=standard2024.05 | 58.8 | 92.24 | — | |
| GramBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 57.39 | 89.39 | — | |
| RMDSIn-Distribution Dataset=ImageNet-2002026.03 | 56.24 | 89.44 | 57.49 | |
| RankFeatBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 55.89 | 89.63 | — | |
| KLMIn-Distribution Dataset=ImageNet-2002026.03 | 54.03 | 91.97 | 57.36 | |
| ASHBackbone=ViT16-B, Training Objective=Cross Entropy, Preprocessing=standard2024.05 | 53.9 | 93.5 | — | |
| GradNormIn-Distribution Dataset=ImageNet-2002026.03 | 53.5 | 91.16 | 54.35 | |
| SHEIn-Distribution Dataset=ImageNet-2002026.03 | 52.82 | 91.16 | 55.08 | |
| ASHIn-Distribution Dataset=ImageNet-2002026.03 | 50.96 | 90.29 | 52.35 | |
| MSPIn-Distribution Dataset=ImageNet-2002026.03 | 50.94 | 89.08 | 51.41 | |
| TempScaleIn-Distribution Dataset=ImageNet-2002026.03 | 50.05 | 89.33 | 50.83 | |
| MDSIn-Distribution Dataset=ImageNet-2002026.03 | 48.59 | 93.9 | 51.88 | |
| MDSBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 48.5 | 92.1 | — | |
| GENIn-Distribution Dataset=ImageNet-2002026.03 | 48.33 | 89.59 | 49.75 | |
| MLSIn-Distribution Dataset=ImageNet-2002026.03 | 48.18 | 90.68 | 49.82 | |
| DICEIn-Distribution Dataset=ImageNet-2002026.03 | 48.11 | 90.94 | 50.61 | |
| RAS HLIn-Distribution Dataset=ImageNet-2002026.03 | 47.81 | 90.86 | 49.76 | |
| OpenMaxIn-Distribution Dataset=ImageNet-2002026.03 | 47.64 | 91.55 | 50.93 | |
| EBOIn-Distribution Dataset=ImageNet-2002026.03 | 47.56 | 90.71 | 49.39 | |
| RASIn-Distribution Dataset=ImageNet-2002026.03 | 47.42 | 90.84 | 49.33 | |
| ReActIn-Distribution Dataset=ImageNet-2002026.03 | 47.25 | 91.22 | 49.48 | |
| RankFeatIn-Distribution Dataset=ImageNet-2002026.03 | 46.41 | 95.78 | 50.58 | |
| VIMIn-Distribution Dataset=ImageNet-2002026.03 | 45.34 | 91.61 | 48.87 | |
| ODINIn-Distribution Dataset=ImageNet-2002026.03 | 44.31 | 91.71 | 46.5 | |
| KNNIn-Distribution Dataset=ImageNet-2002026.03 | 44.05 | 91.73 | 47.7 | |
| MDSEnsBackbone=ResNet50, Training Loss=Cross Entropy, Preprocessing=Standard Preprocessing2024.05 | 43.92 | 95.19 | — | |
| GradNormBackbone=ViT16-B, Training Objective=Cross Entropy, Preprocessing=standard2024.05 | 42.96 | 93.62 | — | |
| OpenGANIn-Distribution Dataset=ImageNet-2002026.03 | 41.58 | 95.9 | 47.78 | |
| MDSEnsIn-Distribution Dataset=ImageNet-2002026.03 | 34.22 | 96.28 | 41.87 |