Date: August 2015.
Source: SIGGRAPH 2015. Journal ACM Transactions on Graphics (TOG), Volume 34 Issue 4, Article No. 120.
Research Summary: Digital full-body avatars need realistic soft-tissue deformation to move like a real human. Using over 40,000 scans of ten subjects from a high-resolution 4D capture system, the Dyna parametric model aligns its template mesh to sequences of 3D scans and learns how soft-tissue motion causes mesh triangles to deform relative to a base 3D body model. Dyna represents this deformation with a low-dimensional linear subspace, using a second-order auto-regressive model that predicts soft-tissue movement from previous deformations, body velocity and acceleration, and limb angular velocity and acceleration. Dyna also models how deformation varies with body mass index, producing different results for people of different shapes, and generalizes realistically to subjects and motions not seen during training.
3dMD’s Role: Dyna’s soft-tissue deformation model was trained on over 40,000 real-world scans of 10 subjects captured with the dynamic-3D/4D 3dMDbody22.u System. This same dataset was reused directly to train DMPL, SMPL’s dynamic soft-tissue extension, and separately later refined into Dynamic FAUST.
Article: Dyna: A Model of Dynamic Human Shape in Motion.
Authors: Gerard Pons-Moll, Javier Romero, Naureen Mahmood, Michael J Black, Max Planck Institute for Intelligent Systems, Tübingen, Germany.