MGAvatar combines Gaussians and meshes for high-fidelity head avatars

Researchers have introduced MGAvatar, a novel hybrid representation for creating detailed head avatars. This system combines Gaussian-based rendering with mesh deformation to accurately model complex geometry, including hair and clothing, which parametric templates struggle with. MGAvatar uses vertex-bound Gaussians for progressive mesh deformation and then switches to face-bound Gaussians for appearance modeling, enhanced by a view-conditioned neural color field and a Gaussian offset network to capture dynamic facial textures. Experiments demonstrate that MGAvatar surpasses existing methods in rendering quality, producing high-fidelity avatars with rich texture details from multi-view and monocular videos. AI

IMPACT Introduces a new hybrid representation for more realistic and detailed head avatars, potentially improving applications in virtual reality and digital media.

RANK_REASON The item describes a new research paper detailing a novel method for head avatar modeling. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• alphaXiv
• arXiv
• CatalyzeX
• Connected Papers
• DagsHub
• Gotit.pub
• Hugging Face
• Litmaps
• MGAvatar
• ScienceCast
• Scite

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →