Pushing further the state of the art in learning-based contour detection
Decanato - Facoltà di scienze informatiche
Data d'inizio: 14 Luglio 2011
Data di fine: 15 Luglio 2011
The Faculty of Informatics is pleased to announce a seminar given by Iasonas Kokkinos
DATE: Thursday, July 14th, 2011
PLACE: USI Università della Svizzera italiana, room SI-008, Black building (Via G. Buffi 13)
Shape-related features, such as object boundaries, can serve a host of vision tasks including segmentation, registration, and recognition. Their detection however is impeded by numerous factors, such as low contrast, texture, and lack of contextual information.
In this talk I will present recent advances on learning-based boundary detection that largely circumvent these problems. We use a carefully engineered combination of old and novel learning techniques to optimize the F-measure of our classifier, while distilling as much information as possible from the training set. In specific, we will see how boosting can be used to optimize the F-measure, how Multiple Instance Learning accommodates ambiguity in the ground truth labels, as well as certain new variants of boosting that can exploit large numbers of data, and densely sampled image features. Finally, we will see how descriptors can exploit contextual information to boost performance.
Our approach results in dramatic improvements compared to the previous state-of-the-art on boundary detection; using the F-measure as a performance metric, we achieve an increase from .7 to .74, with humans performing at .78, effectively cutting down by half the gap between human and machine performance.
Iasonas Kokkinos obtained the Diploma of Engineering in 2001 and the Ph.D. Degree in 2006, both from the School of Electrical and Computer Engineering of the National Technical University of Athens, Greece. In 2006 he joined the Center for Image and Vision Sciences in the University of California at Los Angeles as a postdoctoral scholar.
As of 2008 he is Assistant Professor at the Department of Applied Mathematics of Ecole Centrale Paris and is also affiliated with the Galen group of INRIA-Saclay in Paris.
His research interests are in the broader areas of computer vision, signal processing and machine learning, while he has worked on nonlinear speech processing, biologically motivated vision, texture analysis and image segmentation. His currently research activity is focused on shape-based object recognition and learning-based approaches to feature detection.
He has been awarded a young researcher grant by the French National Research Agency, and serves regularly as a reviewer for all major computer vision conferences and journals.
HOST: Prof. Michael Brostein