Human Skeleton-aware Virtual Try-on via Fashion Landmarks and Garment Deformation
Abstract
Virtual Try-On technology (VTON) enablesusers to visualize how a specific garment would appear on their bodies, servingas an important tool to support online purchase of clothing. The GAN models havebeen recently applied in VTON, generating visually pleasing fitting images. Nevertheless,these methods generate clothing images using a learned pixel-level translationtechnique, without modelling the complex relationship between human andclothing. This study introduces a novel human skeleton-aware fashion landmarksdeformation VTON method, called SLD-VTON, which transfers the wearing position fromthe source garment to the target person at the desired position, for a size faithful try-oneffect. Extensive experiment has demonstrated that SLD-VTON can size faithfuland realistic try-on results. Compared with other state-of-the-art VTONmethods, the proposed method is more robust in garments deformation and excelsin maintaining the details.
Keywords: Virtual Try-on, Fashion Landmarks, Pose Estimation, Human Parsing, Skeleton-aware, thin plate spline (TPS)
How to Cite:
He, H., Zhou, Y., Sun, Z., Fan, J. & Mok, P., (2024) “Human Skeleton-aware Virtual Try-on via Fashion Landmarks and Garment Deformation”, International Textile and Apparel Association Annual Conference Proceedings 80(1). doi: https://doi.org/10.31274/itaa.17592
Downloads:
Download updated version
View PDF
362 Views
116 Downloads