Automatic generation of 3D natural anime-like non-player characters with machine learning

Published by Issei FUJISHIRO on

Ruizhe Li, Masanori Nakayama, Issei Fujishiro

in Proceedings of Cyberworlds 2020 (online), pp. 110–116, September 29–October 1, 2020

[doi: 10.1109/CW49994.2020.00023]

In role-playing games, creators always use specific models to generate non-player characters (NPCs). The prefabricated models are materialized in the scenes, and if more NPCs are included than character models prepared, the players may easily find many NPCs with similar appearances, which makes the scenes unnatural. In this paper, we propose a novel system for generating a rich variety of 3D anime-like NPCs in real time to make the scenes look more natural. We combine a proprietary character customization system with machine learning, where the customizing parameters are treated as feature vectors input in the neural network. The parameters are trained to avoid generating bad-looking models. We demonstrate that the proposed system can generate a natural school classroom scene with a variety of good-looking female student NPCs in a uniform.

Publication page in 2020 is here


Leave a Reply

Avatar placeholder

Your email address will not be published. Required fields are marked *