Personalized Avatars Creation Based on Realistic 3d Face Reconstruction

Abstract

Personalized 3D human avatars hold great potential in the digital era, and this also enables interesting fashion applications such as virtual try-on, virtual fashion shows, and immersive online shopping experiences. Nevertheless, personalized avatars are seldom used in the fashion industry due to the requirements of specialized equipment and the tedious and computationally demanding nature of traditional personalized face modeling processes. In this paper, a novel method is presented for creating personalized fashion avatars by reconstructing a realistic 3D face from a single image. The proposed method involves transferring customized 3D facial shape and appearance, reconstructed from a single face image, onto full-body avatars with varying topologies suitable for different fashion applications. Through a comprehensive comparison with state-of-the-art face reconstruction methods, the proposed method has demonstrated high realism and richer facial details. This technology has already been applied in an augmented reality (AR) mobile application, enabling users to engage in virtual try-on experience of fashion.