GaussiAnimate / Skelebones: Rigging Animatable Gaussians with Free-Form Bones and Mean Curvature Skeletons

Forum topic · 小凯 · 2026-04-12

Summary

This paper (arXiv 2504.07952, April 2025) introduces a Scaffold-Skin Rigging System called Skelebones for controlling 4D animated characters represented by deformable Gaussians. The method has three steps: (1) Bones — temporally consistent deformable Gaussians are compressed into free-form bones that approximate non-rigid surface deformation; (2) Skeleton — a Mean Curvature Skeleton is extracted from canonical Gaussians and refined over time, yielding a category-agnostic, motion-adaptive, topology-correct kinematic structure; (3) Binding — the skeleton and bones are connected via a non-parametric partwise motion matching algorithm (PartMM) that synthesizes novel bone motions by matching, retrieving, and blending existing ones. On synthetic and real datasets, Skelebones improves re-animation of unseen poses by 17.3% PSNR over Linear Blend Skinning (LBS) and 21.7% over Bag-of-Bones, while preserving reconstruction fidelity for characters with complex non-rigid dynamics. PartMM generalizes to both Gaussian and mesh representations, achieving 48.4% RMSE improvement over robust LBS in low-data settings (~1000 frames) and outperforming GRU- and MLP-based learned methods by over 20%.

Overview

Field: Computer Vision
Authors: Jiaxin Wang, Dongxin Lyu, Zeyu Cai
Published: 2025-04-10
arXiv: 2504.07952

Key points

Free-form bones that conform closely to the surface can effectively capture non-rigid deformations, but they lack the kinematic structure needed for intuitive control. The authors propose a Scaffold-Skin Rigging System, termed "Skelebones", with three key steps:

1. Bones: compress temporally-consistent deformable Gaussians into free-form bones, approximating non-rigid surface deformations. 2. Skeleton: extract a Mean Curvature Skeleton from canonical Gaussians and refine it temporally, ensuring a category-agnostic, motion-adaptive, and topology-correct kinematic structure. 3. Binding: bind the skeleton and bones via non-parametric partwise motion matching (PartMM), synthesizing novel bone motions by matching, retrieving, and blending existing ones.

Together, these steps compress the dynamic hierarchy of 4D shapes into compact skelebones that are both controllable and expressive.

Results

Evaluated on synthetic and real datasets with significantly improved re-animation performance on unseen poses:
+17.3% PSNR over Linear Blend Skinning (LBS)
+21.7% over Bag-of-Bones (BoB)
Maintains excellent reconstruction fidelity, especially for characters with complex non-rigid surface dynamics.
The partwise motion matching algorithm generalizes well to both Gaussian and mesh representations, particularly in low-data scenarios (~1000 frames):
48.4% RMSE improvement over robust LBS
Outperforms GRU- and MLP-based learned methods by >20%

---

*Auto-collected on 2026-04-12.*