Synthetic Data Framework for Rotogravure Defect Detection

A new pipeline generates high-fidelity synthetic images of printing defects like creases and streaks with automated bounding box annotations. The framework was used to train an RFDETR model using a newly generated dataset of 7,533 images to overcome real-world data scarcity.