Skip to content

← ArchivePaper2016

A Deep Learning Framework for Character Motion Synthesis and Editing

Daniel Holden, Jun Saito, Taku Komura

SIGGRAPHAcademic24 descendantsMotion Synthesis

Convolutional autoencoder for human motion that enables synthesis and editing of character motion sequences in a compact latent space.

Abstract

We present a framework to synthesize character movements based on high level parameters, such that the produced movements respect the manifold of human motion, trained on a large motion capture dataset. The learned motion manifold, which is represented by the hidden units of a convolutional autoencoder, represents motion data in sparse components which can be combined to produce a wide range of complex movements. To map from high level parameters to the motion manifold, we stack a deep feedforward neural network on top of the trained autoencoder. This network is trained to produce realistic motion sequences from parameters such as a curve over the terrain that the character should follow, or a target location for punching and kicking. The feedforward control network and the motion manifold are trained independently, allowing the user to easily switch between feedforward networks according to the desired interface, without re-training the motion manifold. Once motion is generated it can be edited by performing optimization in the space of the motion manifold. This allows for imposing kinematic constraints, or transforming the style of the motion, while ensuring the edited motion remains natural. As a result, the system can produce smooth, high quality motion sequences without any manual pre-processing of the training data.

How to read this

Category
Method: a deep-learning framework for motion synthesis and editing
Contributions
  • Learns a motion manifold via a convolutional autoencoder trained on a large motion-capture dataset, representing motion as sparse combinable components
  • Stacks a feedforward network on top of the manifold to map high-level parameters (terrain curves, target locations for punching/kicking) to realistic motion
  • Enables editing by optimizing in the manifold space, imposing kinematic constraints or transforming style while keeping motion plausible
Context
Extends the authors' earlier convolutional-autoencoder motion manifold work (holden-deep-learning-motion-2015) into a full synthesis-and-editing framework with high-level control.Builds on: Learning Motion Manifolds with Convolutional Autoencoders
Correctness
Assumes the learned manifold captures the space of natural human motion from the training corpus, so synthesized and edited motion stays plausible only within that distribution; out-of-distribution parameters or styles may degrade results.
Clarity
Reasonably accessible; a first pass conveys the manifold-plus-control idea, a second pass clarifies the autoencoder architecture and the editing optimization.
How to read it
Focus first on the separation of motion manifold from control network (trained independently); a second pass is worth it to understand the convolutional autoencoder and the manifold-space optimization for editing.

Builds on

Built upon by

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 →