AI-shrink: intelligent recognition of markers for fabric dimensional changes

Fabric dimensional change rate is a critical indicator of textile quality and garment performance, yet accurate detection remains challenging due to complex textures, deformed markering points, and post-wash marker variations. To address these issues, this study proposes an enhanced YOLOv8m-based detection model by incorporating the Efficient Multi-Scale Attention (EMA) mechanism, Bidirectional Feature Pyramid Network (BiFPN), and MPDIoU loss to improve feature extraction and localization. A dedicated dataset comprising 1200 images from 158 fabric types and 8 marker types was constructed for training and validation. Experimental results demonstrate superior accuracy and robustness over baseline YOLOv8 models and traditional template matching, particularly for complex patterned fabrics and diverse marker categories. The proposed method provides an effective solution for intelligent and automated fabric dimensional change rate detection and supports the development of smart textile inspection systems.