← ArchivePaper2023

SAME: Skeleton-Agnostic Motion Embedding for Character Animation

Sunmin Lee, Taeho Kang, Jungnam Park, Jehee Lee, Jungdam Won

SIGGRAPH AsiaAcademic34 citesRetargeting Motion Synthesis

Learns an embedding space that disentangles skeleton structure from motion semantics, enabling cross-skeleton motion retargeting.

Open the paper ↗Trace in the lineage graph

Abstract

Learning deep neural networks on human motion data has become common in computer graphics research, but the heterogeneity of available datasets poses challenges for training large-scale networks. This paper presents a framework that allows us to solve various animation tasks in a skeleton-agnostic manner. The core of our framework is to learn an embedding space to disentangle skeleton-related information from input motion while preserving semantics, which we call Skeleton-Agnostic Motion Embedding (SAME). To efficiently learn the embedding space, we develop a novel autoencoder with graph convolution networks and provide new formulations of various animation tasks operating in the SAME space. We showcase various examples, including retargeting, reconstruction, and interactive character control, and conduct an ablation study to validate design choices made during development.

How to read this

Category: Method: a skeleton-agnostic motion embedding for animation
Contributions: Skeleton-Agnostic Motion Embedding (SAME) that disentangles skeleton structure from motion semantics
A graph-convolution autoencoder to learn the embedding plus new formulations of animation tasks operating in SAME space
Demonstrations of retargeting, reconstruction, and interactive character control with an ablation of design choices
Context: Addresses dataset skeleton heterogeneity in deep motion learning, in the lineage of Skeleton-Aware Networks for deep motion retargeting (Aberman et al.).Builds on: Skeleton-Aware Networks for Deep Motion Retargeting
Correctness: Assumes skeleton-related information can be cleanly separated from semantics while preserving motion meaning; validated across several tasks with an ablation, though the degree of disentanglement and behavior on very exotic skeletons should be read critically.
Clarity: Moderately accessible; a first pass conveys the disentanglement idea and applications, a second pass clarifies the graph-conv autoencoder and per-task formulations.
How to read it: Focus on what the embedding disentangles and how each task is recast in SAME space; a second pass on the autoencoder and ablation is worthwhile if comparing retargeting methods.

Builds on

Skeleton-Aware Networks for Deep Motion Retargeting 2020

Built upon by

Nothing yet.

Related work

Dog Code: Human to Quadruped Embodiment Using Shared Codebooks 2024 / MIG
WalkTheDog: Cross-Morphology Motion Alignment via Phase Manifolds 2024 / SIGGRAPH
Perpetual Humanoid Control for Real-time Simulated Avatars 2023 / ICCV
Universal Humanoid Motion Representations for Physics-Based Control 2024 / ICCV

Keywords

skeleton-agnostic motion embedding same motion retargeting graph convolution networks graph neural networks autoencoder motion embedding graph attention forward kinematics layer amass smpl motion matching interactive character control missing joint reconstruction motion classification skeletal variation dynamic time warping embedding consistency loss

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 →