Lightweight Quaternion Transition Generation with Neural Networks
PubDate: May 2021
Teams: IT University of Copenhagen
Writers: Romi Geleijn; Adrian Radziszewski; Julia Beryl van Straaten; Henrique Galvan Debarba
This paper introduces the Quaternion Transition Generator (QTG), a new network architecture tailored to animation transition generation for virtual characters. The QTG is simpler than the current state of the art, making it lightweight and easier to implement. It uses approximately 80% fewer arithmetic operations compared to other transition networks. Additionally, this architecture is capable of generating visually accurate rotation-based animations transitions and results in a lower Mean Absolute Error than transition generation techniques that are commonly used for animation blending.