Skip to content

← ArchivePaper2021

A Deep Emulator for Secondary Motion of 3D Characters

Mianlun Zheng, Yi Zhou, Duygu Ceylan, Jernej Barbic

CVPRAcademic33 citesML DeformationCFX

Graph-network ODE emulator adds vivid secondary dynamics to skinned characters over 30x faster than full FEM simulation with topology-independent inference.

Open the paper ↗Trace in the lineage graph

Abstract

Fast and light-weight methods for animating 3D characters are desirable in various applications such as computer games. We present a learning-based approach to enhance skinning-based animations of 3D characters with vivid secondary motion effects. We design a neural network that encodes each local patch of a character simulation mesh where the edges implicitly encode the internal forces between the neighboring vertices. The network emulates the ordinary differential equations of the character dynamics, predicting new vertex positions from the current accelerations, velocities and positions. Being a local method, our network is independent of the mesh topology and generalizes to arbitrarily shaped 3D character meshes at test time. We further represent per-vertex constraints and material properties such as stiffness, enabling us to easily adjust the dynamics in different parts of the mesh. We evaluate our method on various character meshes and complex motion sequences. Our method can be over 30 times more efficient than ground-truth physically based simulation, and outperforms alternative solutions that provide fast approximations.

How to read this

Category
Method: learned emulator for secondary motion of 3D characters
Contributions
  • A neural network that adds vivid secondary-motion dynamics to skinning-based character animation by emulating the character-dynamics ODEs
  • A local patch encoding where mesh edges implicitly encode internal forces, predicting new vertex positions from current accelerations, velocities, and positions, making it topology-independent and generalizable to arbitrary meshes
  • Per-vertex constraints and material properties (e.g. stiffness) that let the dynamics be tuned across different mesh regions
Context
Sits in the lineage of learned deformation approximators for animation (related to Fast and Deep Deformation Approximations by Bailey et al.), here targeting physically inspired secondary dynamics rather than pose-space deformation.Builds on: Fast and Deep Deformation Approximations
Correctness
Reported as over 30x more efficient than ground-truth physically based simulation and outperforming fast-approximation alternatives; a reader should keep in mind it emulates rather than solves the dynamics, so it approximates the reference simulation it was trained against.
Clarity
Accessible at the local-patch / graph idea; a first pass conveys the emulation concept, a second pass covers the ODE-style prediction and constraint handling.
How to read it
First pass on why a local, topology-independent encoding generalizes across meshes; second pass on the acceleration/velocity/position prediction and material parameters if you plan to integrate it.

Builds on

Built upon by

Nothing yet.

Related work

Keywords

secondary motiondeep emulatorskinningvolumetric meshtetrahedral meshlocal patchphysically based simulationfinite element methodimplicit backward eulercentral differencesmultilayer perceptrongraph convolutional networkmeshgraphnetsinternal forcesstiffnesslinear blend skinningmixamovega femelastic energymaterial properties

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 →