Exploring Intrinsic Normal Prototypes within a Single Image for Universal Anomaly Detection

Benchmark Model Rank Results
multi-class-anomaly-detection-on-mvtec-adINP-Former-Large#3Detection AUROC: 99.8Segmentation AUROC: 98.6
multi-class-anomaly-detection-on-mvtec-adINP-Former-Base#4Detection AUROC: 99.7Segmentation AUROC: 98.5
anomaly-detection-on-visaINP-Former ViT-B (model-unified multi-class)#6Detection AUROC: 98.9Segmentation AUPRO (until 30% FPR): 94.4…
anomaly-detection-on-mvtec-adINP-Fomer ViT-L (model-unified multi-class)#7Detection AUROC: 99.8Segmentation AUPRO: 95.6Segmentation AUROC: 98.6…