Example-based approach for automatic garment pattern generation

R. Q. Hong

doi:10.1177/00405175251358497

Example-based approach for automatic garment pattern generation

2025-09-01

0

PRE

AI

DOI：10.1177/00405175251358497

Original

Original request for help

Save

Abstract

En 中文

This paper proposes an example-based method for automated garment pattern generation, addressing challenges in craftsmanship standardization and geometric fidelity in existing 2D pattern techniques. This approach integrates graph neural network (GNN)-based garment panel segmentation with manufacturing-constrained flat pattern modeling to establish a seamless bridge between digital fashion design and traditional garment craftsmanship. More specifically, a sparse graph transformer is employed to efficiently segment 3D garment meshes into individual panels. Leveraging the technique of virtual node sparsification, this method remarkably reduces the computational complexity, enabling a more efficient segmentation process. To ensure its practical viability in industrial applications, the methodology incorporates two critical types of manufacturing constraints. Symmetry constraints are imposed on the internal boundaries of panels and seams, while boundary constraints are applied to guarantee smooth and production-friendly edges. A hybrid boundary optimization strategy, which combines geometric constraints with B-spline fitting, is then utilized to refine the generated 2D patterns. Comprehensive experimental evaluations demonstrate the superiority of the proposed method. On a self-constructed dataset, it achieves an impressive 99.99% segmentation accuracy, and on cross-domain models, the accuracy reaches 99.91%. Moreover, compared with conventional approaches, the training time is reduced by 34%. For dresses and T-shirts, the generated patterns exhibit 100% structural similarity to template patterns, significantly outperforming the compared methods. Although the proposed method results in a slightly higher stretching ratio (ranging from 0.0157 to 0.0378) compared with the baseline methods (0.0112-0.0189), it ensures well-organized panel layouts and smooth boundaries, strictly adhering to industry standards and effectively preventing cutting errors caused by irregular shapes. By maintaining regular panel layouts and enforcing geometric constraints explicitly, the generated patterns preserve high fidelity during 3D-to-2D flattening while meeting industrial production standards.

Keywords:

Garment mesh model

example-based approach

manufacturing constraint

graph neural network (GNN)

pattern generation

Example-based approach for automatic garment pattern generation

Abstract

Journal

Organization