From Attraction to Equilibrium: Physics-Inspired Semantic Gravitons for Zero-Shot Anomaly Detection

Benchmark Model Rank Results
zero-shot-anomaly-detection-on-br35h-zero-shot-cross-domainSGNet–Detection AUROC: 97.8
zero-shot-anomaly-detection-on-brainmri-zero-shot-cross-domainSGNet–Detection AUROC: 96.4
zero-shot-anomaly-detection-on-btad-zero-shot-cross-domainSGNet–Detection AUROC: 92.9Segmentation AUROC: 97.2
zero-shot-anomaly-detection-on-cvc-clinicdb-zero-shot-cross-domainSGNet–Segmentation AUROC: 88.5
zero-shot-anomaly-detection-on-dagm-zero-shot-cross-domainSGNet–Detection AUROC: 99.2Segmentation AUROC: 99.5
zero-shot-anomaly-detection-on-dtd-synthetic-zero-shot-cross-domainSGNet–Detection AUROC: 98.7Segmentation AUROC: 99.3
zero-shot-anomaly-detection-on-mpdd-zero-shot-cross-domainSGNet–Detection AUROC: 80.8Segmentation AUROC: 97.5
zero-shot-anomaly-detection-on-mvtec-ad-zero-shot-cross-domainSGNet–Detection AUROC: 93.5Segmentation AUROC: 92.8
zero-shot-anomaly-detection-on-visa-zero-shot-cross-domainSGNet–Detection AUROC: 85.9Segmentation AUROC: 95.9