Neurally-inspired computational models can potentially be applied to a large number of visual recognition problems. Applications being researched today include:
Models of visual processing that are applied to real-world problems face particular challenges. Data for these problems are usually very high-dimensional, which can severely strain computational resources. The choice of a ``good'' image feature for a particular recognition problem is ill-defined. Should most effort be spent on developing specialized feature detectors using a priori notions of the problem space? Or is a flexible recognition architecture of primary importance? Related to this issue are questions of normalization. How robust is a system to variations in translation, scale, and direction of illumination?
In this workshop, these questions are addressed for a range of difficult, real-world applications. The problems include: integration of multiple sources of visual information, visual object tracking, fMRI analysis, image retrieval, character recognition, and object detection. The solutions to these problems promise to have substantial commercial and scientific value. Come to this workshop to see the future of applied statistical modeling!
The goal of this workshop is to explore the state-of-the-art in models applied to real-world visual recognition problems. Because of the wide range of approaches currently used, and open questions about optimal architectures for this problem, we believe that this topic will provide a rich set of issues for discussion and interaction. The workshop is one day in length, and will include short talks and time for discussion in both morning and afternoon sessions.
The workshop co-chairs are Michael Gray (University of California, San Diego and the Salk Institute) and Javier Movellan (University of California, San Diego). The following speakers will be presenting at this workshop:
Page by Suhas Chelian.