Scale Up Video understanding with Deep Neural Networks
Mr. Chuang Gan of Tsinghua University
Thursday, September 29, 2016 · 11:00AM · HEC 101
Abstract
The increasing ubiquity of devices capable of capturing videos has led to an explosion in the amount of recorded video content. Instead of “eyeballing” the videos for potential useful information, it is desirable to develop automatic video analysis and understanding algorithms. Yet understanding video on a large scale remains challenging: large variations and complexities, time-consuming annotations, and wide range of involved video concepts. In light of these challenge, we proposed a Deep Event Network (DevNet) that simultaneously detects pre-defined events and provides key spatial-temporal evidences. To scale up video recognition, we pave two new research directions: 1) utilize Web images and videos returned by commercial search engines to replace human-annotated data to conduct labor-free video recognition and 2) teach machine to learn new video concept categories through zero-shot learning.
Biography
Chuang Gan is a fourth-year Phd student in Tsinghua University, supervised by Professor Andrew Chi-Chih Yao, the laureate of A.M. Turing Award. His research focus is on Computer Vision and Machine Learning. In particular, he is interested in deep learning for large-scale video understanding. He has also spent time working at Carnegie Mellon University, University of Southern California and Microsoft Research Redmond.
