Out-of-Distribution Detection on CIFAR100 (test) (OOD Targets: SVHN, TinyImageNet, LSUN, Places, Textures)
29.1Avg FPR95 (Overall)GAIA-Z
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| GAIA-ZBackbone=ResNet34, Pre-training loss=cross-entropy loss2023.11 | 29.1 | 15.73 | 97.06 | 63.85 | 89.17 | 33.33 | 94.18 | 16.78 | 97.17 | 15.82 | 97.09 | 94.93 | |
| GAIA-ABackbone=WRN40, Pre-training loss=cross-entropy loss2023.11 | 36.29 | 35.49 | 93.6 | 53.37 | 89.86 | 33.52 | 93.86 | 27.62 | 95.37 | 31.44 | 94.16 | 93.37 | |
| GAIA-ZBackbone=WRN40, Pre-training loss=cross-entropy loss2023.11 | 38.63 | 15.19 | 97.19 | 87.06 | 73.42 | 37.97 | 91.59 | 25.64 | 95.26 | 27.29 | 94.05 | 90.3 | |
| WeiPer+KLDBackbone=ResNet182024.05 | 48.78 | — | — | — | — | — | — | — | — | — | — | 83.54 | |
| WeiPer+MSPBackbone=ResNet182024.05 | 49.28 | — | — | — | — | — | — | — | — | — | — | 83.39 | |
| RMDSBackbone=ResNet182024.05 | 49.56 | — | — | — | — | — | — | — | — | — | — | 82.55 | |
| KNNBackbone=ResNet182024.05 | 49.65 | — | — | — | — | — | — | — | — | — | — | 83.34 | |
| WeiPer+ReActBackbone=ResNet182024.05 | 49.65 | — | — | — | — | — | — | — | — | — | — | 83.4 | |
| GENBackbone=ResNet182024.05 | 49.98 | — | — | — | — | — | — | — | — | — | — | 83.25 | |
| TempScaleBackbone=ResNet182024.05 | 50.26 | — | — | — | — | — | — | — | — | — | — | 82.79 | |
| MSPBackbone=ResNet182024.05 | 50.7 | — | — | — | — | — | — | — | — | — | — | 82.07 | |
| ReActBackbone=ResNet182024.05 | 51.47 | — | — | — | — | — | — | — | — | — | — | 82.88 | |
| MLSBackbone=ResNet182024.05 | 51.83 | — | — | — | — | — | — | — | — | — | — | 82.9 | |
| EBOBackbone=ResNet182024.05 | 52.03 | — | — | — | — | — | — | — | — | — | — | 82.76 | |
| OpenMaxBackbone=ResNet182024.05 | 52.99 | — | — | — | — | — | — | — | — | — | — | 78.44 | |
| VIMBackbone=ResNet182024.05 | 54.66 | — | — | — | — | — | — | — | — | — | — | 77.76 | |
| DICEBackbone=ResNet182024.05 | 54.93 | — | — | — | — | — | — | — | — | — | — | 80.72 | |
| ODINBackbone=ResNet182024.05 | 55.19 | — | — | — | — | — | — | — | — | — | — | 81.63 | |
| SHEBackbone=ResNet182024.05 | 57.74 | — | — | — | — | — | — | — | — | — | — | 79.74 | |
| ASHBackbone=ResNet182024.05 | 63.35 | — | — | — | — | — | — | — | — | — | — | 79.92 | |
| KNNBackbone=WRN40, Pre-training loss=cross-entropy loss2023.11 | 66.13 | 46.88 | 88.97 | 70.88 | 82.86 | 68.92 | 76.83 | 83.57 | 69.64 | 60.41 | 83.66 | 80.39 | |
| GAIA-ABackbone=ResNet34, Pre-training loss=cross-entropy loss2023.11 | 68.97 | 68.02 | 89.03 | 68.61 | 83.33 | 71.24 | 86.37 | 73.15 | 86.25 | 63.81 | 87.12 | 86.42 | |
| KLMBackbone=ResNet182024.05 | 71.07 | — | — | — | — | — | — | — | — | — | — | 79.22 | |
| OpenGANBackbone=ResNet182024.05 | 74.21 | — | — | — | — | — | — | — | — | — | — | 68.74 | |
| ReActBackbone=ResNet34, Pre-training loss=cross-entropy loss2023.11 | 74.51 | 77.53 | 83.17 | 71.18 | 78.6 | 73.36 | 84.37 | 78.41 | 80.12 | 72.06 | 82.54 | 81.76 | |
| ReActBackbone=WRN40, Pre-training loss=cross-entropy loss2023.11 | 74.67 | 75.04 | 82.36 | 76.09 | 75.83 | 66.64 | 83.06 | 77.94 | 78.18 | 77.66 | 78.33 | 79.55 | |
| KNNBackbone=ResNet34, Pre-training loss=cross-entropy loss2023.11 | 77.25 | 73.34 | 80.06 | 69.24 | 82.17 | 76.98 | 78.36 | 86.76 | 71.53 | 79.95 | 69.24 | 76.27 | |
| RankFeatBackbone=ResNet182024.05 | 78.4 | — | — | — | — | — | — | — | — | — | — | 65.72 | |
| MDSBackbone=ResNet182024.05 | 79.05 | — | — | — | — | — | — | — | — | — | — | 61.5 | |
| ASH-P@70Backbone=WRN40, Pre-training loss=cross-entropy loss2023.11 | 79.8 | 81.2 | 80.99 | 76.24 | 77.92 | 74.78 | 81.06 | 84.81 | 73.78 | 81.97 | 76.12 | 77.97 | |
| ODINBackbone=WRN40, Pre-training loss=cross-entropy loss2023.11 | 80.1 | 80.64 | 82.34 | 78.5 | 76.41 | 74.43 | 81.95 | 84.57 | 74.58 | 82.36 | 76.51 | 78.36 | |
| ASH-P@65Backbone=ResNet34, Pre-training loss=cross-entropy loss2023.11 | 80.7 | 81.21 | 79.46 | 74.26 | 81.17 | 82.84 | 74.93 | 85.49 | 72.91 | 79.7 | 77.33 | 77.16 | |
| MahalanobisBackbone=WRN40, Pre-training loss=cross-entropy loss2023.11 | 80.97 | 82.36 | 81.07 | 82.95 | 79.2 | 74.76 | 81.16 | 82.44 | 76.06 | 83.72 | 76.93 | 78.34 | |
| MSPBackbone=WRN40, Pre-training loss=cross-entropy loss2023.11 | 81.26 | 83.44 | 79.85 | 76.94 | 77.84 | 76.68 | 80.32 | 85.81 | 72.5 | 83.42 | 74.94 | 77.09 | |
| MahalanobisBackbone=ResNet34, Pre-training loss=cross-entropy loss2023.11 | 81.29 | 88.71 | 73.72 | 75.7 | 79.57 | 88.28 | 71.63 | 78.54 | 79.74 | 82.63 | 73.78 | 76.16 | |
| EnergyBackbone=WRN40, Pre-training loss=cross-entropy loss2023.11 | 81.55 | 84.58 | 79.72 | 76.77 | 77.9 | 76.32 | 80.45 | 86.13 | 72.35 | 83.95 | 74.83 | 77.05 | |
| ODINBackbone=ResNet34, Pre-training loss=cross-entropy loss2023.11 | 82.61 | 89.34 | 70.21 | 70 | 81.44 | 83.8 | 71.37 | 88.1 | 67.69 | 81.81 | 72.66 | 72.67 | |
| MSPBackbone=ResNet34, Pre-training loss=cross-entropy loss2023.11 | 82.84 | 86.21 | 74.13 | 75.21 | 79.31 | 83.58 | 72.8 | 87.19 | 70.6 | 82 | 74.46 | 74.26 | |
| EnergyBackbone=ResNet34, Pre-training loss=cross-entropy loss2023.11 | 83.29 | 87.55 | 73.91 | 73.46 | 79.83 | 84.38 | 72.58 | 88.53 | 70.17 | 82.54 | 74.69 | 74.24 | |
| GradNormBackbone=WRN40, Pre-training loss=cross-entropy loss2023.11 | 85.89 | 85.27 | 69.22 | 86.58 | 67.75 | 81.1 | 62.38 | 87.01 | 52.89 | 89.41 | 51.3 | 60.71 | |
| GradNormBackbone=ResNet34, Pre-training loss=cross-entropy loss2023.11 | 86.32 | 90.7 | 65.95 | 80.12 | 61.44 | 82.62 | 58.1 | 92.29 | 64.35 | 85.89 | 52.48 | 60.46 | |
| GradNormBackbone=ResNet182024.05 | 86.85 | — | — | — | — | — | — | — | — | — | — | 69.95 | |
| RankfeatBackbone=WRN40, Pre-training loss=cross-entropy loss2023.11 | 88.29 | 80.39 | 77.1 | 94.58 | 52.35 | 91.63 | 61.89 | 86.83 | 67.71 | 88 | 67.36 | 65.28 | |
| RankfeatBackbone=ResNet34, Pre-training loss=cross-entropy loss2023.11 | 89.75 | 92.94 | 65.55 | 87.46 | 74.98 | 90.84 | 70.65 | 90.77 | 72.68 | 86.72 | 73.99 | 71.57 | |
| GramBackbone=ResNet182024.05 | 91.85 | — | — | — | — | — | — | — | — | — | — | 53.91 | |
| MDSEnsBackbone=ResNet182024.05 | 95.82 | — | — | — | — | — | — | — | — | — | — | 48.78 |