This page is not updated, Link to my lab research page





Projects in my group are divided in five categories.


1, Image Parsing

Inspired by Marr's quest for a general vision solution, we pose image parsing aa computing process that infers the following representations from a single image under a common framework. click the buttons/text to view projects texture texton prmskt scale segmentation grouping gestalt 2.1d 2.5d 3d his object scene

2, Video Parsing

Video parsing extends image parsing to larger scenes and longer duration to account for the interactions between agents and manipulable objects. vtexture vtexton vprmskt vscale action activity event

3, Modeling, Learning, Inference Algorithms, and Basic Theories

The progress of computer vision as a scientific discipline should be measured by its development of models, algorithms, and theories, so that its problems and solutions can be understood analytically.
Modeling and Learning
Inference Algorithms
Basic Theories
  • Minimax Entropy Learning
  • Prior Learning
  • Learning by Information Projection
  • Learning Implicit and Explicit Manifolds
  • Stochastic Image Grammar
  • And-or-graph Learning
  • PAC-learning of And-or graph models
  • Energy landscape of non-convex optimization problems

4, Real World Applications

Only when commercial requests are clearly presented, the progress shall be measured by specific datasets and benchmarks.
We have produced commericial systems based on projects at our Lotus Hill Institute.
Persistent Surveillance
Look at Humans
Computational Photography and arts
Aerial Image Understanding
Intelligent Vehicle
  • Background Modeling
  • Tracking in AoG
  • Actions
  • Events and Ontology
  • Image to Text (I2T)
  • PTZ Camera Tracking
  • Counting People
    • Scene modeling
    • Top-down/ Bottom-up Inference
    • Driving Log (I2T)

    5, Core and Long Standing Issues in Human and Computer Vision

    These are long-standing debates which can only be answered numerically using statistical and information theoretical approaches.

    Issue I: Discriminative vs. Generative methods
    Issue II: Top-down vs. Bottom-up computing processes
    Issue III: 2D view-based vs. 3D object-based representations
    Issue IV: Texture vs. Texton
    (similarly Context v.s. Hierarchy, Sketchable v.s. Non-Sketchable, Trackable v.s. Non-Trackable)
    © S.-C. Zhu