Out-of-Distribution Detection on CIFAR-10 (FPR/AUROC vs SVHN, LSUN, iSUN, Texture, Places365)
1.51FPR (SVHN)SSD+
Evaluation results
| Method | Links | |||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SSD+Contrastive Learning=Yes, Backbone=ResNet-18, Training Data=CIFAR-102022.04 | 1.51 | 99.68 | 6.09 | 98.48 | 33.6 | 95.16 | 12.98 | 97.7 | 28.41 | 94.72 | 16.52 | 97.15 | 95.07 | — | — | — | — | |
| KNN+Contrastive Learning=Yes, Backbone=ResNet-18, Training Data=CIFAR-102022.04 | 2.42 | 99.52 | 1.78 | 99.48 | 20.06 | 96.74 | 8.09 | 98.56 | 23.02 | 95.36 | 11.07 | 97.93 | 95.07 | — | — | — | — | |
| MahalanobisBackbone=DenseNet-101, Pre-trained on ID data=true2023.02 | 6.42 | 98.31 | — | — | 9.78 | 97.25 | 21.51 | 92.15 | 85.14 | 63.15 | 31.42 | 89.15 | — | 56.55 | 86.96 | 9.14 | 97.09 | |
| MahalanobisContrastive Learning=No, Backbone=ResNet-18, Training Data=CIFAR-102022.04 | 9.24 | 97.8 | 67.73 | 73.61 | 6.02 | 98.63 | 23.21 | 92.91 | 83.5 | 69.56 | 37.94 | 86.5 | 94.21 | — | — | — | — | |
| KNNBackbone=DenseNet-101, Pre-trained on ID data=true2023.02 | 13.51 | 96.68 | — | — | 10.79 | 97.91 | 24.5 | 95.19 | 63.88 | 85 | 25.83 | 94.39 | — | 30.95 | 93.82 | 11.37 | 97.72 | |
| VRA+Backbone=DenseNet-101, Pre-trained on ID data=true2023.02 | 13.54 | 97.45 | — | — | 6.15 | 98.71 | 27.07 | 95.03 | 39.97 | 91.96 | 15.85 | 96.91 | — | 2.03 | 99.56 | 6.37 | 98.72 | |
| GODINContrastive Learning=No, Backbone=ResNet-18, Training Data=CIFAR-102022.04 | 18.72 | 96.1 | 11.52 | 97.12 | 30.02 | 94.02 | 33.58 | 92.2 | 55.25 | 85.5 | 29.82 | 92.97 | 93.64 | — | — | — | — | |
| VRABackbone=DenseNet-101, Pre-trained on ID data=true2023.02 | 18.75 | 96.68 | — | — | 5.7 | 98.69 | 34.89 | 93.42 | 39.98 | 91.69 | 17.74 | 96.47 | — | 1.32 | 99.63 | 5.8 | 98.69 | |
| ODINBackbone=DenseNet-101, Pre-trained on ID data=true2023.02 | 25.29 | 94.57 | — | — | 3.98 | 98.9 | 57.5 | 82.38 | 52.85 | 88.55 | 24.57 | 93.71 | — | 4.7 | 98.86 | 3.09 | 99.02 | |
| DICEBackbone=DenseNet-101, Pre-trained on ID data=true2023.02 | 25.99 | 95.9 | — | — | 4.36 | 99.14 | 41.9 | 88.18 | 48.59 | 89.13 | 20.84 | 95.25 | — | 0.26 | 99.92 | 3.91 | 99.2 | |
| KNNContrastive Learning=No, Backbone=ResNet-18, Training Data=CIFAR-102022.04 | 27.97 | 95.48 | 18.5 | 96.84 | 24.68 | 95.52 | 26.74 | 94.96 | 47.84 | 89.93 | 29.15 | 94.55 | 94.21 | — | — | — | — | |
| SHEBackbone=DenseNet-101, Pre-trained on ID data=true2023.02 | 28.12 | 94.72 | — | — | 10.99 | 97.95 | 51.98 | 83.07 | 59.35 | 84.16 | 26.82 | 92.98 | — | 0.76 | 99.84 | 9.73 | 98.15 | |
| CSIContrastive Learning=Yes, Backbone=ResNet-18, Training Data=CIFAR-102022.04 | 37.38 | 94.69 | 5.88 | 98.86 | 10.36 | 98.01 | 28.85 | 94.87 | 38.31 | 93.04 | 24.16 | 95.89 | 94.38 | — | — | — | — | |
| EnergyBackbone=DenseNet-101, Pre-trained on ID data=true2023.02 | 40.61 | 93.99 | — | — | 10.07 | 98.07 | 56.12 | 86.43 | 39.4 | 91.64 | 26.55 | 94.57 | — | 3.81 | 99.15 | 9.28 | 98.12 | |
| ReActBackbone=DenseNet-101, Pre-trained on ID data=true2023.02 | 41.64 | 93.87 | — | — | 12.72 | 97.72 | 43.58 | 92.47 | 43.31 | 91.03 | 26.45 | 95.3 | — | 5.96 | 98.84 | 11.46 | 97.87 | |
| MSPBackbone=DenseNet-101, Pre-trained on ID data=true2023.02 | 47.27 | 93.48 | — | — | 42.31 | 94.52 | 64.15 | 88.15 | 63.02 | 88.57 | 48.74 | 92.46 | — | 33.57 | 95.54 | 42.1 | 94.51 | |
| ODINContrastive Learning=No, Backbone=ResNet-18, Training Data=CIFAR-102022.04 | 53.78 | 91.3 | 10.93 | 97.93 | 28.44 | 95.51 | 55.59 | 89.47 | 43.4 | 90.98 | 38.43 | 93.04 | 94.21 | — | — | — | — | |
| EnergyContrastive Learning=No, Backbone=ResNet-18, Training Data=CIFAR-102022.04 | 54.41 | 91.22 | 10.19 | 98.05 | 27.52 | 95.59 | 55.23 | 89.37 | 42.77 | 91.02 | 38.02 | 93.05 | 94.21 | — | — | — | — | |
| MSPContrastive Learning=No, Backbone=ResNet-18, Training Data=CIFAR-102022.04 | 59.66 | 91.25 | 45.21 | 93.8 | 54.57 | 92.12 | 66.45 | 88.5 | 62.46 | 88.64 | 57.67 | 90.86 | 94.21 | — | — | — | — |