URHead: A Unified UV-Space Representation for Joint Mesh-3DGS Optimization in Head Avatars

2026-07-08

AI Summary

Existing avatar modeling approaches struggle to simultaneously achieve geometric controllability and photorealistic fidelity: mesh-based methods offer structural precision but lack fine details, while 3D Gaussian splatting (3DGS) methods produce high realism at the cost of temporal or structural consistency. To address this trade-off, this work proposes a unified optimization framework that integrates mesh and 3DGS representations in UV space. By leveraging shared parameterization and adaptive Gaussian sampling, the method decouples and jointly optimizes the two representations, enabling them to complement each other effectively. This approach presents the first deep integration of meshes and 3DGS in the UV domain, preserving intricate appearance details while ensuring animation consistency. Extensive evaluations demonstrate superior performance over state-of-the-art methods in both reconstruction quality and drivability.

Abstract

We present URHead, a unified representation for high-fidelity and animatable head avatars that fundamentally redefines mesh-Gaussian integration. While mesh-based methods offer precise geometric control but lack photorealistic detail, and Gaussian-based approaches achieve photorealism but suffer from poor structural consistency, existing hybrid solutions fail to fully leverage their complementary strengths. Our key contribution is a UV-space unification where both representations share a common UV parameterization. Through joint optimization with adaptive gaussian sampling, our method automatically learns to disentangle and allocate appropriate roles to each component. URHead maintains full parametric controllability while preserving subject-specific details, and outperforms existing state-of-the-art methods in reconstruction quality and animation consistency.

Problem

• head avatars
• mesh-Gaussian integration
• photorealism
• structural consistency
• animatable modeling

Innovation

• UV-space unification
• joint mesh-3DGS optimization
• adaptive Gaussian sampling
• animatable head avatars
• parametric controllability