What question did this study set out to answer?

To improve the quality and usability of SMPL-X models by enhancing them with photorealistic details and animation capabilities.

January 14, 2026Open Access

An embedding of realistic virtual human bodies created by SMPL-X through 3D scans and their application

Puntos clave

To improve the quality and usability of SMPL-X models by enhancing them with photorealistic details and animation capabilities.
Developed a pipeline integrating 3D scans for texture transfer and semantic material labeling.
Utilized a neural network for classifying material types in body regions.
Implemented skeleton insertion techniques for full animation functionality.
Enhanced SMPL-X models with realistic textures and improved animation capability.
Facilitated the accurate animation of garments on digital human models.
Supported export in standard formats like OBJ or FBX.

Resumen

Photorealistic, animatable digital humans are valuable across industries such as fashion, healthcare, and entertainment, yet generating high-quality models remains complex and time-consuming. While parametric models like SMPL-X enable automatic body reconstruction from images, they lack texture and rigging, limiting their use in modern rendering engines. This paper presents a lightweight pipeline that combines scan-based texture transfer, semantic material labeling, and skeleton insertion to enhance SMPL-X models with photorealistic detail and full animation capability. Body regions are segmented using skinning weights, then refined and classified with a neural network to identify material types. The resulting digital avatar can be exported in standard formats like OBJ or FBX. With the help of the generated labels it is possible to accurately animate garments and simulation clothing on generated body models.

Leer artículo completoexternamente

Me gusta

Guardar

Ver artículo completo