← ArchivePaper2020
SoftSMPL: Data-driven Modeling of Nonlinear Soft-tissue Dynamics for Parametric Humans
Recurrent network regresses real-time soft-tissue dynamics as a function of body shape and motion encoded in a nonlinear deformation subspace.
Abstract
We present SoftSMPL, a learning‐based method to model realistic soft‐tissue dynamics as a function of body shape and motion. Datasets to learn such task are scarce and expensive to generate, which makes training models prone to overfitting. At the core of our method there are three key contributions that enable us to model highly realistic dynamics and better generalization capabilities than state‐of‐the‐art methods, while training on the same data. First, a novel motion descriptor that disentangles the standard pose representation by removing subject‐specific features; second, a neural‐network‐based recurrent regressor that generalizes to unseen shapes and motions; and third, a highly efficient nonlinear deformation subspace capable of representing soft‐tissue deformations of arbitrary shapes. We demonstrate qualitative and quantitative improvements over existing methods and, additionally, we show the robustness of our method on a variety of motion capture databases.
How to read this
- Category
- Method: learning-based soft-tissue dynamics for parametric human bodies
- Contributions
- A motion descriptor that disentangles the standard pose representation by removing subject-specific features
- A recurrent neural regressor that generalizes soft-tissue dynamics to unseen body shapes and motions
- An efficient nonlinear deformation subspace that represents soft-tissue deformation for arbitrary shapes
- Context
- Adds learned, motion-dependent soft-tissue dynamics on top of the SMPL parametric body model (Loper et al.), targeting realism beyond static pose-corrective deformation.Builds on: SMPL: A Skinned Multi-Person Linear Model
- Correctness
- Addresses the scarcity and cost of dynamics training data, with the disentangled descriptor and nonlinear subspace aimed at avoiding overfitting; a reader should keep in mind that generalization claims rest on the chosen mocap databases and on how well the subspace captures unseen shapes.
- Clarity
- Reasonably accessible if you already know SMPL; a first pass conveys the three-part design, a second pass clarifies the recurrent regressor and subspace.
- How to read it
- First pass on the three contributions and why each fights overfitting; second pass on the motion descriptor disentanglement and the recurrent dynamics regressor if you work with body dynamics.
Builds on
Built upon by
Nothing yet.
Related work
- SNARF: Differentiable Forward Skinning for Animating Non-Rigid Neural Implicit Shapes 2021 / CVPR
- SNUG: Self-Supervised Neural Dynamic Garments 2022 / CVPR
- Animatable Neural Radiance Fields for Modeling Dynamic Human Bodies 2021 / CVPR
- Learning Skeletal Articulations with Neural Blend Shapes 2021 / SIGGRAPH
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 →