2016.09.07 (Wed) ‘Unsupervised visual object discovery and learning’ – Minsu Cho 조민수교수(POSTECH 포항공과대학교)
-Titile: Unsupervised visual object discovery and learning
-Biograpy :Minsu Cho is an assistant professor in the department of Computer Science and Engineering at POSTECH. He obtained his PhD in 2012 from Seoul National University, Korea, and worked as a researcher (starting research position) in the Inria WILLOW team at the Department of Computer Science, École Normale Supérieure (ENS), Paris, France, until 2016. He is also a member of team GAYA, an associate research team jointly established with Inria and Carnegie Mellon University. His research has focused on the interplay between computer vision and machine learning, especially in the problems of graph-based object matching and learning. His recent research investigates unsupervised or minimally-supervised object discovery, localization, and tracking in images and videos.
-Abstract:Object recognition is one of the main problems in computer vision, which is highly challenging because of intra-class variations, background clutter, and occlusions present in real-world images and videos. While significant progress has been made in this area over the last decade, most state-of-the-art methods still rely on strong supervision in the form of manually-annotated bounding boxes on target instances. Since those detailed annotations are expensive to acquire and also prone to unwanted biases and errors, avoiding strong supervision is a matter of importance. In this talk, I will briefly introduce current attempts to reduce the degree of supervision in object recognition, i.e., weakly-supervsed or unsupervsed object localization, describe their underlying assumptions, and discuss major issues in the state of the art. In addition, I will briefly introduce our recent approaches to unsupervised object discovery and localization, and show how object matching, localization, and learning can be related each other.