Chenyi Kuang

Chenyi Kuang

Computer Vision Researcher · 3D Vision · Human Behavior Understanding

Now I am an Algorithm Engineer at SkyHive Network. I received my Ph.D. in Electrical Engineering from Rensselaer Polytechnic Institute in 2024.

My research focuses on human behavior understanding from visual data, including gaze estimation, human-object interaction recognition and anticipation, dynamic 3D face reconstruction, 3D eye modeling, and physics-informed machine learning. I am especially interested in building vision systems that understand human attention, action, and intent for assistive robotics, autonomous systems, and embodied AI applications.

News

Research Interests

Human Attention Understanding

Gaze estimation, saliency detection, visual attention modeling, and gaze anticipation.

Human Activity Understanding

Human-object interaction recognition, action anticipation, and intent analysis.

3D Human Modeling

Dynamic 3D face reconstruction, 3D eye modeling, expression analysis, and mesh-based learning.

Physics-informed Learning

Combining biomechanical constraints, physical dynamics, and deep neural networks.

Vision for Robotics

Human-aware perception systems for assistive robotics, aviation platforms, and autonomous systems.

Uncertainty-aware Modeling

Probabilistic prediction, uncertainty estimation, and robust multimodal perception.

Featured Projects

Physics-informed Dynamic 3D Face Reconstruction

A physics-driven framework for dynamic 3D facial reconstruction that integrates biomechanical constraints and facial motion dynamics for expression tracking in video sequences.

Demo 1: Dynamic 3D Face Reconstruction

Demo 2: Physics-informed Facial Motion / Force Visualization

3D Face Physics-informed ML Facial Dynamics

Interaction-aware Dynamic 3D Gaze Estimation

A dynamic gaze estimation framework that models temporal gaze behavior and social interaction patterns in videos, including multi-person attention and interaction-aware gaze dynamics.

Demo: Interaction-aware 3D Gaze Estimation

3D Gaze Human Interaction Temporal Modeling

SAGE: Synchronized Action and Gaze Estimation

A unified framework for gaze detection, gaze anticipation, action recognition, and action anticipation. The system models how human gaze supports action and intent reasoning in first-person and third-person videos.

Gaze Estimation Action Anticipation HOI Recognition Uncertainty

Exo-Cook Benchmark

A benchmark for third-person joint gaze-action understanding in cooking activities, including gaze labels, action labels, object information, and baseline models for comprehensive human activity analysis.

Benchmark Ego-Exo Video Gaze-Action Modeling

Experience

Feb. 2026 – Present

Algorithm Engineer

SkyHive Network, LLC · Newark, DE
  • Developing lightweight deep learning models for production user behavior prediction, including face recognition and user attention analysis.
  • Building systems that incorporate spatial-temporal pilot interaction data and pilot activity analysis into recommendation algorithms for an aviation platform.
Jan. 2025 – Present

Research Collaborator

Honda Research Institute USA
  • Collaborating on gaze-conditioned human-object interaction recognition and anticipation research.
  • Developed an uncertainty-aware model for joint gaze and action detection and anticipation.
  • Patent submission: Gaze-aware Human Activity Detection and Anticipation.
May 2024 – Aug. 2024

Computer Vision Research Intern

Honda Research Institute USA · San Jose, CA
  • Developed a unified system for joint human attention, action, and intent analysis.
  • Constructed a benchmark for comprehensive human activity analysis with human/object trajectories, action labels, and gaze labels.
May 2022 – Aug. 2022

Computer Vision Research Intern

IBM Almaden Lab · San Jose, CA
  • Developed an uncertainty-guided data-free knowledge distillation algorithm for model fusion.
  • Applied the method to computer vision tasks including image classification and object detection.
Sep. 2019 – May 2022

Research Assistant

Rensselaer Polytechnic Institute · Troy, NY
  • Developed geometry-aware facial expression recognition models using 3D Morphable Models.
  • Developed physics-driven 4D facial reconstruction for dynamic facial tracking and expression analysis.
  • Constructed deformable 3D eye models for accurate 3D gaze estimation under diverse head poses and illumination conditions.
  • Built multi-person gaze tracking systems for modeling group interaction dynamics and human attention patterns.

Education

Sep. 2019 – Dec. 2024

Ph.D. in Electrical Engineering

Rensselaer Polytechnic Institute · Troy, NY, USA
Sep. 2015 – Jun. 2019

Bachelor in Electrical Engineering

University of Science and Technology of China · Hefei, China

Selected Publications

Physics-informed Dynamic 3D Face Reconstruction
Chenyi Kuang, Jeffrey O. Kephart, Qiang Ji · FG 2026 · Oral
Interaction-aware Dynamic 3D Gaze Estimation in Videos
Chenyi Kuang, Jeffrey O. Kephart, Qiang Ji · NeurIPS 2024 GMML Workshop · Oral
AU-aware Dynamic 3D Face Reconstruction from Videos with Transformer
Chenyi Kuang, Jeffrey O. Kephart, Qiang Ji · WACV 2024
AU-aware 3D Face Reconstruction Through Personalized AU-specific Blendshape Learning
Chenyi Kuang, Zijun Cui, Jeffrey O. Kephart, Qiang Ji · ECCV 2022
Towards an Accurate 3D Deformable Eye Model for Gaze Estimation
Chenyi Kuang, Jeffrey O. Kephart, Qiang Ji · ICPR 2022

Technical Skills

Programming & Tools

Python, MATLAB, PyCharm, Jupyter Notebook, Spyder

Machine Learning Libraries

PyTorch, TensorFlow, Hugging Face Transformers, Scikit-learn, Matplotlib

3D Vision & Geometry

PyTorch3D, gsplat, Trimesh, Open3D, 3D mesh processing, 3D Gaussian Splatting

Research Areas

Gaze estimation, human-object interaction, action recognition, 3D face reconstruction, physics-informed learning

Contact

Email: kuangcy1998@gmail.com
Location: Philadelphia, PA, USA