In the age of big data, it is critical to develop scalable methods to address challenges in the real-world. In the first part of my talk, I will present a novel method to efficiently and robustly classify a large number of object categories. The proposed "find the best path" method achieves a significantly better trade-off between classification accuracy and time-efficiency than other existing methods. The proposed method is beneficial for various applications which require different trade-off between accuracy and efficiency. In the second part of my talk, I will present a fully automatic system to retrieve domain-specific highlights in raw personal videos. Given a domain-specific query such as "surfing", our system mines the Youtube database to analyze how users generate edited surfing videos. Then, the mined information is used to train our domain-specific ranker to retrieve highlights from unseen raw person videos. This work addresses the critical need toward automatic video editing to increase audience retention of personal videos.
Min Sun is a postdoctoral researcher in the Computer Science and Engineering department at the University of Washington (UW) working with Steve Seitz and Ali Farhadi. Before joining UW, he graduated from the University of Michigan at Ann Arbor with a Ph.D. degree and Stanford University with a M.Sc. degree. His research interests include object recognition, human pose estimation, and scene understanding in both 2D and 3D. Most recently, he focuses on developing scalable methods for object categorization and video analysis in the age of big data. He has won the best paper award in 3DRR and was also a recipient of W. Michael Blumenthal Family Fund Fellowship.
Sponsored by Tandent Vision Science, Inc.
kkitani [atsymbol] cs.cmu.edu