Learning object models in visual semantic networks

Abstract

Visual semantic networks, a representation scheme for a library of visual object models, are introduced. New models are learned in the library with the help of a knowledge engineer, who informs the system of the generic class of each new example, and then the system discovers potential cases of further classification, and gets them confirmed by the knowledge engineer. The details of discovery are discussed, and it is argued that the system behavior is more or less independent of the order of presentation of the examples. Experiments with a small number of mechanical tools confirm this.<>