SkinCells: Sparse Skinning using Voronoi Cells

2025-06-17

AI Summary

Existing automatic skinning methods struggle to simultaneously ensure high-quality weights for complex geometries and controllable sparsity (e.g., ≤4 bones per vertex), while lacking mechanisms for weight reuse across levels of detail (LODs) and asset variants. Method: We propose a fully automatic sparse skinning framework introducing SkinCells—a novel parametric function family—that jointly optimizes skinning weights and sparsity constraints in continuous space, overcoming limitations of discrete-point optimization and robustly handling complex topologies where biharmonic solvers fail. Voronoi-based spatial modeling and radial basis function parameterization enable consistent, LOD-invariant weight generation and efficient cross-variant reuse. Contribution/Results: Our method achieves high-fidelity, fully automatic skinning on intricate characters, drastically reducing or eliminating manual correction. It supports real-time multi-character rendering on mobile devices, with significantly improved inference efficiency.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Intelligent Robots: Learning & Optimization for ROB
• Machine Learning: Learning with Manifolds

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Web performance, measurement, and characterization

Abstract

For decades, efficient real-time skinning methods have played a crucial role in animating character rigs for visual effects and games. These methods remain a fundamental component of modern applications. However, animatable digital asset creation predominantly remains a manual process. Current automated tools often fall short of delivering the desired level of quality for intricate and complex geometries, requiring manual touch-ups. We propose a fully automatic and robust method for generating high quality skinning weights given a user-provided skeleton and mesh in A- or T-pose. Notably, our approach provides direct sparsity controls, limiting the number of bone influences per vertex, which is essential for efficient asset creation for large-scale mobile experiences with multiple concurrent users. Our method additionally addresses the need for level-of-detail (LoD) variations in performance-sensitive applications, which are exacerbated on mobile platforms. By optimizing weights in space rather than on discrete points, we enable a single optimization result to be seamlessly applied to all levels of detail of that asset or even variations of that asset. To achieve this, we introduce a novel parameterized family of functions called SkinCells. We demonstrate how our automatic method is able to robustly compute skinning weights in cases where biharmonic weight computation fails.

Problem

• Automates high-quality skinning weight generation for skeletons and meshes
• Provides sparsity control for efficient large-scale mobile asset creation
• Supports level-of-detail variations for performance-sensitive applications

Innovation

• Automatic skinning weights generation
• Direct sparsity controls for efficiency
• Space optimization for LoD variations