Towards practical and efficient home‑based facial geometry and appearance capture with consistent topology

Facial geometry and appearance capture are a foundational technique for digital human creation in the metaverse. Existing methods either rely on expensive hardware or suffer from low efficiency and inconsistent topology. This paper proposes a practical, low-cost, and fully automatic system for high-quality facial asset reconstruction from smartphone videos. The method adopts a sequential pipeline: multi-view stereo reconstructs detailed geometry, followed by non-rigid ICP template fitting to ensure topology consistency. A 2D multi-resolution hash grid in UV space is used to model reflectance, avoiding texture bleeding and hyperparameter tuning. Experiments show that the method reconstructs relightable facial assets in 8 minutes while maintaining quality comparable to state-of-the-art approaches. This work advances accessible and production-ready facial capture for everyday users. Code is available here .