HFNeRF: Learning Human Biomechanic Features with Neural Radiance Fields

Abstract

HFNeRF introduces a novel approach to learning human biomechanic features using Neural Radiance Fields. By leveraging 2D pre-trained foundation models, HFNeRF learns to encode human biomechanic information — including skeletal structure and joint kinematics — directly within a 3D neural radiance field. This bridges the gap between neural rendering and biomechanical analysis, enabling comprehensive human understanding from multi-view images. The method demonstrates strong potential for downstream applications in augmented reality, virtual reality, and sports analytics, where accurate 3D biomechanical modeling from visual data is essential.

Download paper here

Recommended citation:

@inproceedings{dey2024hfnerf,
  title={HFNeRF: Learning Human Biomechanic Features with Neural Radiance Fields},
  author={Dey, Arnab and Yang, Di and Dantcheva, Antitza and Martinet, Jean},
  booktitle={2024 IEEE Conference on Artificial Intelligence (CAI)},
  year={2024},
  address={Singapore}
}