Development of an AI-based garment design process that retrieves flat sketches with similar shapes based on an input flat sketch and automatically generates the corresponding patterns.
A labeled flat sketch dataset was built to identify key components automatically, and the YOLOv8 model was fine-tuned. A flat sketch-pattern dataset was created by generating pattern modules for each component. Software was developed for pattern combination and adjustment, enabling a modular pattern generation system. A comparative analysis validated the process by assessing the similarity between generated and actual garment patterns.
The AI-based process accurately retrieved similar flat sketches and generated corresponding patterns. Experiments showed that the YOLOv8-based detection and pattern combination system achieved an mAP50 of 0.849. The software effectively adjusted seam lengths for precise alignment, preserving garment structure. Three-dimensional reconstruction confirmed high similarity between generated and actual patterns in structure and design details.
This study introduces an end-to-end AI-driven approach to automate garment pattern-making. By building a flat sketch-pattern dataset and a modular pattern system, it enables the generation of structurally complete and production-ready patterns from a single flat sketch, thereby enhancing efficiency and streamlining modifications. The system helps designers save time, allowing more focus on creative detailing.
