Out-of-distribution detection on CIFAR-100 Gaussian
99.65AUROCDeep Abstaining Classifier
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Deep Abstaining ClassifierBackbone=Wide ResNet 28x102021.05 | 99.65 | — | — | — | — | — | — | 0.72 | |
| Deep Abstaining ClassifierBackbone=ResNet 342021.05 | 99.48 | — | — | — | — | — | — | 1.66 | |
| ODINBackbone=Wide ResNet 28x102021.05 | 98.5 | — | — | — | — | — | — | 1 | |
| Outlier ExposureBackbone=Wide ResNet 40x22021.05 | 95.7 | — | — | — | — | — | — | 12.1 | |
| Ensemble of Leave-out ClassifiersBackbone=Wide ResNet 28x102021.05 | 93.04 | — | — | — | — | — | — | 98.26 | |
| Deep Abstaining ClassifierBackbone=Wide ResNet 40x22021.05 | 90.03 | — | — | — | — | — | — | 13.26 | |
| OpenMaxBackbone=ResNet 342021.05 | 89.37 | — | — | — | — | — | — | 21.5 | |
| Softmax BaselineArchitecture=40-4 Wide ResNet, Nonlinearity=GELU, Optimizer=SGD with restarts, Training Epochs=502016.10 | 88 | 50 | 92 | 43 | 80 | 57 | 77 | — |