Proposal of Anime-like Avatar Generation System Based on Topology-Independent 3D Data Analysis Method

Abstract

As the culture within the metaverse develops, the demand for 3D models continues to intensify. Particularly, communication through avatars in Virtual Reality (VR) environments is anticipated to gain increasing significance. Although the generation of photorealistic avatars can be accomplished by various methods, the automatic creation of anime-like avatars has not yet been fully established. Anime-like 3D avatars lack a unified topology, and due to their wide range and exaggerated characteristics, the difference in topologies can be considerable. In this research, to address this issue, we devised a method for generating datasets that extract features sharing a common topology from avatar data of varying topologies. Additionally, by separately extracting mesh data and texture data, we have made it possible to extract characteristics from different models, and freely distribute them. We also revealed that using these datasets, it is possible to establish an interface that allows users to generate a variety of avatars by using a small number of parameters.