Table 2. Quantitative comparisons of AA-CLIP and our framework on industrial domain datasets (MVTec-AD and VisA) and medical domain datasets (Clinic DB).

Method	MVTec-AD
	Image-level		Pixel-level
	AUROC	AP	AUROC	AP
AA-CLIP [6]	99.2	99.6	99.6	86.3
AA-CLIP+ our framework	99.5	99.7	99.5	86.5