Verfahren zur Rekonstruktion eines Animationsfähigen 3D-Avatars aus Monokularen Bildern

Details

Veröffentlichungs-Nr.

EP4760640

Anmeldetag

13. Dezember 2024

Veröffentlichung

17. Juni 2026

Rechtsraum

EP

IPC

G06T15/20, G06T17/00

Abstract

A computer-implemented method for reconstructing an animatable 3D avatar of a subject from a monocular video (110) comprising: (S10) acquiring a monocular video (110); (S20) generating a parametric morphable representation (120) based on the monocular video (110); (S30) associating 3D geometric primitives of the subject (130) to the parametric morphable representation (120); (S40) generating rendered images (140) of the subject under target views (109) based on the 3D geometric primitives (130); (S50) generating one normal map set (150) rendered under the target views (109) based on the parametric morphable representation (120); (S60) generating a pseudo-ground truth image set (160) under the target views (109) based on an image from the monocular video (110) using a multi-view diffusion model (170) conditioned by the normal map set (150); and (S70) adjusting the 3D geometric primitives by optimizing the rendered images (140) using the pseudo-ground truth image set (160).

Anmelder

Firmen

TOYOTA JIDOSHA KABUSHIKI KAISHA
Technical University of Munich

Land

🇯🇵 Japan