MGAvatar improves realistic head avatars with hybrid geometry representation
MGAvatar: Mesh-Bound Gaussians for Head Avatar Geometry and Appearance Modeling
Computer Vision and Pattern Recognition
Summary
Creating realistic 3D head models is hard because existing methods often rely on basic face shapes that miss personal details like hair or clothes. The authors developed MGAvatar, which combines a flexible mesh with Gaussian blobs to better capture detailed head shapes and appearances. They also introduced new ways to handle how the face changes with different expressions and views, making the avatars look more consistent and detailed. Tests showed MGAvatar generates higher quality and more lifelike head models than previous methods.
What this means in practice
- •For game developers: Create more detailed and responsive 3D head avatars with realistic hair and facial expressions for games and virtual environments.$Commercial implications: Enables production of high-fidelity, personalized avatar assets enhancing player immersion and customization features in games.
- •For animated film studios: Generate high-quality head models with dynamic facial textures and consistent appearances across multiple viewpoints for animated characters.$Commercial implications: Supports efficient creation of believable digital characters, reducing manual animation effort and improving visual quality.
Authors
Lei Shi, Sen Peng, Zhiyang Deng, Zhonggui Chen, Xiaohu Guo, Baorong Yang, Xiao Dong
Abstract
Accurate head modeling requires a stable yet expressive geometric representation. Existing Gaussian-based head avatars commonly rely on parametric templates (e.g., FLAME) for Gaussian initialization and deformation, but these templates lack personalized priors and struggle to represent structures such as hair and clothing. To address this issue, we propose MGAvatar, a Gaussian-mesh hybrid representation that jointly models geometry and appearance through two Gaussian-mesh binding modes. Specifically, we introduce vertex-bound Gaussians and constrain their learnable parameters, enabling progressive mesh deformation to represent complex head geometry, while a pose-dependent offset module accounts for non-rigid deformations. Once geometry is stabilized, MGAvatar switches to face-bound Gaussians for appearance modeling. To improve appearance consistency across novel poses and viewpoints, we introduce a view-conditioned neural color field that alleviates artifacts caused by independently optimized Gaussian colors. In addition, we design a Gaussian offset network to predict Gaussian offset maps in the observation space, providing greater flexibility for face-bound Gaussians to capture dynamic facial textures. Extensive experiments on multi-view and monocular videos show that MGAvatar outperforms existing methods in rendering quality, producing high-fidelity head avatars with rich texture details.