PANDA: Adapting Pretrained Features for Anomaly Detection and Segmentation

Benchmark Model Rank Results
anomaly-detection-on-fashion-mnistPANDA#1ROC AUC: 95.6
anomaly-detection-on-fashion-mnistSelf-Supervised One-class SVM, RBF kernel#5ROC AUC: 92.8
anomaly-detection-on-fashion-mnistPANDA-OE#9ROC AUC: 91.8
anomaly-detection-on-fashion-mnistSelf-Supervised DeepSVDD#10ROC AUC: 84.8
anomaly-detection-on-hyper-kvasir-datasetPANDA#3AUC: 0.937
anomaly-detection-on-one-class-cifar-10PANDA-OE#4AUROC: 98.9
anomaly-detection-on-one-class-cifar-10PANDA#10AUROC: 96.2
anomaly-detection-on-one-class-cifar-10Self-Supervised DeepSVDD#27AUROC: 64.8
anomaly-detection-on-one-class-cifar-10Self-Supervised One-class SVM, RBF kernel#28AUROC: 64.7
anomaly-detection-on-one-class-cifar-100PANDA-OE#3AUROC: 97.3
anomaly-detection-on-one-class-cifar-100PANDA#5AUROC: 94.1
anomaly-detection-on-one-class-cifar-100Self-Supervised Multi-Head RotNet#9AUROC: 80.1
anomaly-detection-on-one-class-cifar-100Self-Supervised DeepSVDD#11AUROC: 67
anomaly-detection-on-one-class-cifar-100Self-Supervised One-class SVM, RBF kernel#12AUROC: 62.6