Image Classification, OOD, and Adversarial Detection on CIFAR-10 vs SVHN (test)
98.4ID Accuracy (%)C-EDL
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| C-EDLUncertainty Strategy=Monte Carlo2025.06 | 98.4 | 47.04 | 6.66 | 9.39 | |
| C-EDLUncertainty Strategy=Metamorphic Transformation2025.06 | 97.98 | 54.7 | 4.69 | 1.25 | |
| EDL++Uncertainty Strategy=Monte Carlo2025.06 | 97.39 | 56.65 | 12.01 | 15.22 | |
| EDL++Uncertainty Strategy=Metamorphic Transformation2025.06 | 97.29 | 60.53 | 6.59 | 2.35 | |
| R-EDL2025.06 | 96.82 | 62.59 | 8.92 | 14.61 | |
| I-EDL2025.06 | 96.19 | 66.57 | 9.68 | 19.25 | |
| H-EDL2025.06 | 96.09 | 66.74 | 9.84 | 19.64 | |
| S-EDL2025.06 | 96.07 | 65.57 | 10.4 | 3.32 | |
| EDL2025.06 | 95.88 | 67.34 | 10.91 | 20 | |
| Posterior Network2025.06 | 95.63 | 65.74 | 11.82 | 15.57 | |
| DA-EDL2025.06 | 91.25 | 43.83 | 20.4 | 21.48 |