Out-of-Distribution Detection on CIFAR-10 (Near/Far OOD FPR95/AUROC)
26.47Near-OOD FPR95T2FNorm
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| T2FNormTraining setup=Training-time regularization methods, Backbone=ResNet-18, Batch size=1282026.03 | 26.47 | 92.79 | 12.75 | 96.98 | 94.69 | |
| LogitNormTraining setup=Training-time regularization methods, Backbone=ResNet-18, Batch size=1282026.03 | 29.34 | 92.33 | 13.81 | 96.74 | 94.3 | |
| SPCPTraining setup=Training-time regularization methods, Backbone=ResNet-18, Batch size=1282026.03 | 31.67 | 91.57 | 20.54 | 94.55 | 94.91 | |
| CIDERTraining setup=Training-time regularization methods, Backbone=ResNet-18, Batch size=1282026.03 | 32.11 | 90.71 | 20.72 | 94.71 | — | |
| UMAPTraining setup=Training-time regularization methods, Backbone=ResNet-18, Batch size=1282026.03 | 33.12 | 91 | 21.7 | 94.2 | 95.06 | |
| SNNTraining setup=Training-time regularization methods, Backbone=ResNet-18, Batch size=1282026.03 | 37.21 | 90.25 | 26.05 | 92.49 | 95.11 | |
| MSPTraining setup=Post-hoc (vanilla training with cross-entropy), Backbone=ResNet-18, Batch size=1282026.03 | 48.17 | 88.03 | 31.72 | 90.73 | 95.06 | |
| EnergyTraining setup=Post-hoc (vanilla training with cross-entropy), Backbone=ResNet-18, Batch size=1282026.03 | 61.34 | 87.58 | 41.69 | 91.21 | 95.06 | |
| ReActTraining setup=Post-hoc (vanilla training with cross-entropy), Backbone=ResNet-18, Batch size=1282026.03 | 63.56 | 87.11 | 44.9 | 90.42 | 95.06 | |
| DICETraining setup=Post-hoc (vanilla training with cross-entropy), Backbone=ResNet-18, Batch size=1282026.03 | 70.04 | 78.34 | 51.76 | 84.23 | 95.06 | |
| SCALETraining setup=Post-hoc (vanilla training with cross-entropy), Backbone=ResNet-18, Batch size=1282026.03 | 80.45 | 82.55 | 67.53 | 86.39 | 95.06 | |
| ASHTraining setup=Post-hoc (vanilla training with cross-entropy), Backbone=ResNet-18, Batch size=1282026.03 | 86.78 | 75.27 | 79.03 | 78.49 | 95.06 |