Defect detection on watch components — case backs, dial markings, applied indices, hands, polished metal lugs. The watchmaking industry’s tolerance for cosmetic defects is among the strictest, and deep-learning vision handles part-to-part variability that fixed thresholds cannot.