Synthetic Data Generation Replaces Subjective Human Inspection in Gravure Printing Quality Control

arXiv 2607.21577·low signal

Korota Arsène Coulibaly, Mohamed Hamlich and Khalid Hmali build a synthetic data generation framework for automating rotogravure printing quality control, a domain still dependent on slow, costly, and subjective manual inspection. The generalizable pattern is using synthetic defect generation to bootstrap a vision model where real defect examples are rare by definition — good manufacturing lines produce few defects. Applicable to any industrial CV problem with severe class imbalance on the failure class.