Synthetic data framework boosts defect detection in printing to 80.9% mAP

A team of researchers from Morocco and Italy has developed a synthetic data generation framework to overcome the extreme scarcity of real-world defect images in rotogravure printing quality control. Traditional manual inspection is slow, costly, and subjective, while deep learning models like YOLO and Vision Transformers require large labeled datasets that are difficult to obtain. The proposed pipeline automatically generates high-fidelity images of common printing defects—creases, streaks, misregistration—and outputs corresponding bounding boxes and annotations.

To validate the framework, the team generated a synthetic dataset of 7,533 images and trained the state-of-the-art object detection model RFDETR. Experimental results showed the model achieved a Mean Average Precision (mAP) of 80.9% on real industrial testing samples. This framework provides a zero-cost, rapid-deployment solution for automating defect inspection in printing lines without requiring massive manual data collection, potentially transforming quality control in the printing industry.