Neural Head Avatars from Monocular RGB Videos

2022년 09월 02일 (금) | 발표자: 김소랑

This paper reconstruct a full head geometry (include hair) from the monocular RGB Video and performs a reenactment by adding motion to the 3d geometry. This disentangled about geometry and appearance, and added geometry networks and textures networks to improve the representation capabilities of each component and optimized them jointly. This lead it possible to represent 3D geometry and photorealistic textures that were not seen during learning at the state-of-the-art level.