Displaying present location in the site.

Vision AI

ML research team is committed to cutting-edge and applied research related to computational vision. Main research achievements include:

  • Developed uncertainty inference tech for state-of-the-art deep models to deliver highly reliable AI systems
  • Developed data-efficient learning tech (Data Refiner) for addressing the challenge of data and label defects, and achieved AI-enabled process automation to deliver high quality AI training data
  • In-depth collaboration with leading hospitals in medical diagnosis AI,  and developed Endowise® endoscopy quality control system that obtained medical device registry (京械注准20232210470)  and was applied in clinical practice

Current Research Focuses:

  • Deliver high-accuracy video analysis unit technologies for pedestrian & object detecting and tracking  in complex public spaces, realizing real-time digitization and visualization of human activity
  • Synthesize images and simulate virtual environments to enhance the accuracy, robustness, and effectiveness of video analysis systems

Video Analysis

Human & object tracking

ReID retrieval

Human & object association

Image Recognition

Attribute analysis

Pose & action recognition

People & vehicle counting


  • 【IJCNN 2021 (Accepted)】Model Performance Inspection of Deep Neural Networks by Decomposing Bayesian Uncertainty Estimates
  • 【IJCNN 2021 (Accepted)】Layerwise Approximate Inference for Bayesian Uncertainty Estimates on Deep Neural Networks
  • 【MICCAI 2020】An Effective Data Refinement Approach for Upper Gastrointestinal Anatomy Recognition
  • 【ICIP 2020】Loss Rescaling by Uncertainty Inference for Single-stage Object Detection
  • 【IJCNN 2020】A Layer-wise Adversarial Training Approach to Improve Adversarial Robustness