Amap announced the launch of its HumanRig technology and the open-sourcing of the industry's first large-scale dataset specifically designed for 3D human character rigging. This technology aims to reduce the cost of character animation production, driving the animation industry towards greater efficiency and automation. It serves as a crucial infrastructure for Amap's development of immersive navigation experiences.
The core of HumanRig technology is its automatic rigging algorithm. It utilizes a Prior-guided Skeleton Estimator (PGSE) to project 2D prior information into 3D space, initializing a rough skeleton and significantly reducing the complexity of the rigging task. The algorithm also employs a U-shaped Point Transformer as a mesh encoder, eliminating the reliance on 3D mesh edge information and enhancing the robustness of rigging on complex meshes. Furthermore, through a Mesh-Skeleton Mutual Attention Network (MSMAN), it achieves joint optimization of skeleton construction and skinning.
The HumanRig dataset contains 11,434 T-pose meshes following a unified bone topology, encompassing a diverse range of character types from real people to cartoon characters and even anthropomorphic animals. Compared to existing datasets, HumanRig boasts significant advantages in scale, diversity, and skeletal consistency, providing high-quality and diverse data support for rigging research and applications.
Experimental results show that HumanRig technology significantly outperforms existing methods. It not only fills the gap in large-scale human rigging datasets but also provides the animation industry with an efficient automatic rigging solution. By simplifying the rigging process, HumanRig paves the way for automation and efficiency in character animation, opening a new chapter in creative expression and content creation.
In the future, Amap plans to apply HumanRig technology to more innovative business scenarios, such as IP voice packs (including 3D car logos and dynamic skins), personalized 3D digital humans, etc., to further enhance user experience. Amap will also continue to explore intelligent character generation and one-stop animation solutions, promoting technological advancements in 3D modeling and character animation.
The HumanRig dataset is open-source and can be accessed via the following links:
- Dataset: https://modelscope.cn/datasets/jellyczd/HumanRig
- Paper Link: https://arxiv.org/abs/2412.02317
- Project Homepage: https://c8241998.github.io/HumanRig/
