Out-of-Distribution Detection on CIFAR-10 (AUROC vs OOD Sets)
99.92AUROC (SVHN)Shifting transformation learning
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Shifting transformation learningTraining Objective=Lssl2021.06 | 99.92 | 97.61 | 93.72 | 92.99 | 95.03 | 93.24 | 95.42 | |
| GeometricTraining Type=Self-supervised2021.06 | 97.96 | 96.25 | 92.57 | 92.06 | 93.57 | 91.91 | 94.05 | |
| CSI (ens)ensemble=true2021.06 | 97.38 | 97.18 | 93.11 | 92.49 | 94.02 | 92.06 | 94.37 | |
| SupSimCLRTraining Type=Contrastive Learning2021.06 | 97.22 | 94.21 | 91.11 | 92.1 | 92.13 | 88.36 | 92.52 | |
| Shifting transformation learningTraining Objective=Lssl + Lsup2021.06 | 96.6 | 96.91 | 98.73 | 93.57 | 94.12 | 94.07 | 95.67 | |
| CSIensemble=false2021.06 | 96.11 | 95.92 | 92.21 | 91.33 | 92.91 | 90.6 | 93.18 | |
| SSD+2021.06 | 93.8 | 94.05 | 91.77 | 90.28 | 94.4 | 90.4 | 92.45 | |
| BaselineTraining Type=Supervised2021.06 | 92.89 | 87.69 | 88.34 | 87.44 | 89.87 | 87.62 | 88.98 |