Abstract

The first two talks will be on ‘classification’, the next two on ‘beyond’. The first part will examine visual recognition at different levels: natural classes, individual objects, as well as their parts and sub-parts at multiple levels. Related issues will include object segmentation, and the integration of bottom-up with top-down processing in recognition. The second part will discuss the use of vision to understand the world beyond the recognition of single objects, for example, people’s actions and goals. Current methods for object classification are insufficient for dealing with broader aspects of visual interpretations. These limitations will be examined, and some extensions and future directions will be outlined.

Series is sponsored by the Intelligence Initiative (I2) and the Center for Biological & Computational Learning (CBCL).