3D Garment Reconstruction from Multi-view Dressed Images
Journal of Fiber Bioengineering and Informatics · 10.08.2026 · 19 (2) · 177-189
DOI: 10.3993/jfbim26012 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Fiber Bioengineering and Informatics, online erschienen am 10.08.2026. Verfasst von Xin-Yue Hu und 2 weiteren Autor:innen. Die Arbeit nennt 20 Literaturangaben. Gefördert von: Special Fund Project of Wuhan Textile University, Wuhan Textile and Garment Digital Engineering Technology Research Centre.
Abstract (Originalsprache)
3D dressed-modelling has become a core technology in fashion design, digital human modelling, and virtual try-on applications. Existing image-based multi-view garment modelling methods primarily emphasise visual realism while often overlooking dimensional accuracy. To address this issue, this study proposes a high-precision 3D garment reconstruction framework based on multi-view images of dressed subjects. A Mask-CNN neural network is utilised for instance segmentation of clothed human images, enabling the extraction of 2D garment-only regions. The model achieves a low training loss of 0.121 3, indicating reliable segmentation performance. Subsequently, a multi-view reconstruction approach is employed to generate the 3D garment model, followed by Poisson surface reconstruction to enhance geometric smoothness and realism. Comparative analysis between the reconstructed 3D garment and the actual physical garment demonstrates a waist circumference error of only 2.01%, confirming the high geometric accuracy and practical applicability of the proposed method. Unlike reconstruction pipelines that primarily target visual plausibility, the present work explicitly evaluates whether the reconstructed dressed-garment geometry preserves physically meaningful garment dimensions. This metric-oriented perspective is important for subsequent use in digital garment engineering and pattern-related applications.
Förderung
- Special Fund Project of Wuhan Textile University
- Wuhan Textile and Garment Digital Engineering Technology Research Centre
Zitieren
APA
Hu, X., Zhang, M., & Zhang, J. (2026). 3D Garment Reconstruction from Multi-view Dressed Images. Journal of Fiber Bioengineering and Informatics, 19(2), 177-189. https://doi.org/10.3993/jfbim26012
Vancouver
Hu X, Zhang M, Zhang J. 3D Garment Reconstruction from Multi-view Dressed Images. Journal of Fiber Bioengineering and Informatics. 2026;19(2):177-189. doi:10.3993/jfbim26012
BibTeX
@article{Hu2026,
author = {Hu, Xin-Yue and Zhang, Meng-Yun and Zhang, Jun},
title = {3D Garment Reconstruction from Multi-view Dressed Images},
journal = {Journal of Fiber Bioengineering and Informatics},
year = {2026},
volume = {19},
number = {2},
pages = {177--189},
doi = {10.3993/jfbim26012}
}
EUVIMED