|
Research Interest: Human annotated groundtruth dataset, Learning prior model for object and scene
Education:
M.S. Institute of Electronics, Chinese Academy of Sciences, June, 2007
B.S. University of Science and Technology of China, June, 2000
Teaching:: Stat110 (Applied statistics) Disc 1A Tue. 2:00 to 2:50 p.m.. Office hour: Monday 4:00-5:30 p.m. MS 8145
UCLA Related Links: Active Bases,
R@UCLA, IPAM
Some Vision Datasets: ImageParsing, LabelMe, PASCAL(VOC), Caltech101, Berkeley Segmentation
Researches:
|
Learning the motion patterns for abnormal behaviour detection. Top-right: Training video sequence of a traffic scene. Bottom-right: Trajectory graph representation (tg). Right column: Diagram of learning and abnormal detection. A library of spatio-temporal relations are defined on the tg including the scene structural information, single object motion, pair-wise objects constraints and multiple objects interactions. New relations with largest information gain are pursued following a minimax entropy principle. More subtle abnormal behaviours detected as new relations added.
|
|
|
|
Tracking Multiple Category Objects in Outdoor Scene
|
|
|
Zhenyu Yao, Xiong Yang, and Song-Chun Zhu, “Introduction to a large scale general purpose ground truth dataset: methodology, annotation tool, and benchmarks.” 6th Int'l Conf on EMMCVPR [pdf]
|
|
