The article introduces FLARE, a method for creating animatable and relightable 3D mesh avatars from a single monocular video. It aims to efficiently learn personalized 3D head avatars from videos that are geometrically accurate, realistic, relightable, and compatible with current rendering systems. By exploiting highly-optimized methods from traditional computer graphics and approximating some components with neural networks, FLARE creates high-quality avatars that are also efficient to train and render.
Publication date: 27 Oct 2023
Project Page: https://doi.org/10.1145/3618401
Paper: https://arxiv.org/pdf/2310.17519
