Skip to content

← ArchivePaper2024

TailorMe: Self-Supervised Learning of an Anatomically Constrained Volumetric Human Shape Model

Stephan Wenninger, Fabian Kemper, Ulrich Schwanecke, Mario Botsch

EurographicsAcademic2 citesMusclesSkinning

Self-supervised volumetric body shape model fitting anatomical template (bones and soft tissue) to surface scans, enabling localized shape manipulation and fast inference.

Abstract

Human shape spaces have been extensively studied, as they are a core element of human shape and pose inference tasks. Classic methods for creating a human shape model register a surface template mesh to a database of 3D scans and use dimensionality reduction techniques, such as Principal Component Analysis, to learn a compact representation. While these shape models enable global shape modifications by correlating anthropometric measurements with the learned subspace, they only provide limited localized shape control. We instead register a volumetric anatomical template, consisting of skeleton bones and soft tissue, to the surface scans of the CAESAR database. We further enlarge our training data to the full Cartesian product of all skeletons and all soft tissues using physically plausible volumetric deformation transfer. This data is then used to learn an anatomically constrained volumetric human shape model in a self‐supervised fashion. The resulting TailorMe model enables shape sampling, localized shape manipulation, and fast inference from given surface scans.

How to read this

Category
Method / model: anatomically constrained volumetric human shape model
Contributions
  • Registers a volumetric anatomical template of skeleton bones and soft tissue to surface scans of the CAESAR database
  • Enlarges training data via physically plausible volumetric deformation transfer across the Cartesian product of skeletons and soft tissues
  • Learns the TailorMe model self-supervised, enabling shape sampling, localized shape manipulation, and fast inference from scans
Context
Combines anatomical-template ideas from Dicko et al.'s Anatomy Transfer and Kadlecek et al.'s personalized anatomical models with the surface shape-space tradition of Loper et al.'s SMPL, swapping global PCA control for an anatomically grounded volumetric one.Builds on: Anatomy Transfer · Reconstructing Personalized Anatomical Models for Physics-based Body Animation · SMPL: A Skinned Multi-Person Linear Model
Correctness
Assumes the synthesized skeleton-by-soft-tissue combinations are physically plausible enough to train on; demonstrated on CAESAR-derived data with localized control and fast scan fitting, but augmented data realism and anatomical accuracy versus real subjects are the caveats to keep in mind.
Clarity
Clearly motivated against PCA shape models; a first pass conveys the anatomical-template idea, a second pass is needed for the registration and deformation-transfer details.
How to read it
Focus on the volumetric template and the deformation-transfer data augmentation that make localized control possible; a second pass on registration and the self-supervised objective is worthwhile if building or fitting body models.

Builds on

Built upon by

Nothing yet.

Related work

Keywords

This page summarises the entry and links to its original source. The archive never hosts or redistributes the publication itself.Show it in the full archive list →