Physical Human Intelligence
How can machines understand the human body — its shape, motion, and biomechanics — well enough to model, simulate, and reason about what a person is doing and how well they do it?
We build models that recover and reason about the human body from images and video: its 3D geometry, pose, motion, appearance, and biomechanical state. The goal is not pose estimation alone, but a complete physical understanding of people — their movement, skill, and interaction with the world — that can drive digital humans, health, and embodied AI.
What we work on
- 3D human reconstruction and digitization
- Human shape, pose, motion, and appearance modeling
- Biomechanics-grounded human understanding
- Human–object interaction
- Physical skill and task proficiency assessment
- Digital humans and human simulation
- Human understanding for robotics and embodied AI
Selected work
-
Distilling CLIP with Dual Guidance for Learning Discriminative Human Body Shape Representation
CVPR 2024 -
SapiensID: Foundation for Human Recognition
CVPR 2025 -
From 3D Pose to Prose: Biomechanics-Grounded Vision–Language Coaching
CVPR 2026 -
Learning Clothing and Pose Invariant 3D Shape Representation for Long-Term Person Re-Identification
ICCV 2023 -
FarSight: A Physics-Driven Whole-Body Biometric System at Large Distance and Altitude
WACV 2024 -
ROG-Grasp: Root-Oriented Geometry for Robotic Grasping and Placement
IEEE CASE 2026
Community leadership
Vision and generative models can reconstruct and synthesize humans with high visual fidelity, but rarely model how bodies move under real-world physical constraints. This workshop brings together computer vision, biomechanics, simulation, and XR researchers to make physical quantities — ground reaction forces, center-of-mass dynamics, joint torques, and contact/friction states — first-class targets for learning from video, IMU, and multimodal data, and to benchmark the physical plausibility of human perception, modeling, and generation. These physically grounded representations are increasingly central to embodied AI and humanoid robotics, where agents must perceive, imitate, and act with human-like contact and force.
Keynotes: Jiajun Wu (Stanford) · Xin Li (Texas A&M) · Christian Theobalt (MPI for Informatics) · Ehsan Adeli (Stanford) · Dima Damen (Bristol & Google DeepMind).