Date: July 2017.
Source: ACM Transactions on Graphics, Vol. 36, No. 4, Article 54, SIGGRAPH 2017.
Research Summary: Data-driven models can produce realistic human shapes, but they only model the visible surface and cannot capture skin deformation from interactions with the environment. Physics simulation can generalize to external forces, but its parameters are difficult to control. This paper introduces a layered volumetric body model that combines both approaches, an inner layer driven by VSMPL, a volumetric extension of SMPL, and an outer layer of soft tissue simulated using the finite element method (FEM). The model’s layer segmentation and physical parameters are learned directly from real-world 4D scans of people exhibiting soft-tissue deformation, using a relative distance field to control where each layer applies. The result is a full-body avatar that generalizes to new motions and reacts realistically to external forces like pushing and pulling, something purely data-driven models cannot do.
3dMD’s Role: This model’s physical parameters, layer segmentation, and soft-tissue elasticity, are learned from Dyna’s existing dataset, the same 40,000 real-world scans of 10 subjects captured on the 3dMDbody22.u System. Two of the fourteen available Dyna sequences per subject were used for training, with the remaining twelve held out for testing.
Article: Data-Driven Physics for Human Soft Tissue Animation.
Authors: Meekyoung Kim, Gerard Pons-Moll, Sergi Pujades, Seungbae Bang, Jinwook Kim, Michael J. Black, Sung-Hee Lee. KAIST and Max Planck Institute for Intelligent Systems.