AAAI'08 Tutorial on Visual Recognition

Kristen Grauman (University of Texas at Austin) and Bastian Leibe (ETH Zurich)

The visual recognition problem is central to artificial intelligence. From robotics to information retrieval, many desired applications demand the ability to identify and localize categories, places, and objects. This tutorial will overview computer vision algorithms for visual object recognition and image classification. We will introduce primary representations and learning approaches, with an emphasis on recent advances in the field. The target audience consists of researchers/students working in AI and robotics who would like to understand what methods and representations are available for these problems. Our intent is for attendees to walk away understanding what is and isn't possible to do reliably today, and to gain key concepts that could be employed in their own systems or research.

Topics will include recognizing specific objects versus recognizing visual categories; global representations and sliding-windows classifiers; local invariant features: detection and description; pose clustering, voting, Hough transform; indexing features, visual vocabularies; bag-of-words representations; constellation models, part-based models; implicit shape models; distance measures and kernels; supervised and weakly supervised model learning; and current challenges and research directions.

We will assume some familiarity with probability, basic machine learning ideas, and linear algebra.

Syllabus

Please refer to the conference page in case of last-minute schedule changes.