Using Self-Supervised Learning Can Improve Model Robustness and Uncertainty

Benchmark Model Rank Results
anomaly-detection-on-one-class-cifar-10SSOOD#15AUROC: 90.1
anomaly-detection-on-one-class-imagenet-30RotNet + Translation + Self-Attention + Resize#6AUROC: 85.7
anomaly-detection-on-one-class-imagenet-30RotNet + Translation + Self-Attention#7AUROC: 84.8
anomaly-detection-on-one-class-imagenet-30RotNet + Self-Attention#8AUROC: 81.6
anomaly-detection-on-one-class-imagenet-30RotNet + Translation#9AUROC: 77.9
anomaly-detection-on-one-class-imagenet-30RotNet#10AUROC: 65.3
anomaly-detection-on-one-class-imagenet-30Supervised (OE)#11AUROC: 56.1
anomaly-detection-on-unlabeled-imagenet-30-vs-cub-200ROT+Trans#2ROC-AUC: 74.5Network: ResNet-18
anomaly-detection-on-unlabeled-imagenet-30-vs-flowers-102ROT+Trans#3ROC-AUC: 86.3Network: ResNet-18
out-of-distribution-detection-on-cifar-10WRN 40-2 + Rotation Prediction#9AUROC: 96.2FPR95: 16.0
out-of-distribution-detection-on-cifar-10-vs-cifar-100WRN 40-2 + Rotation Prediction#10AUROC: 90.9AUPR: 67.7