General-purpose models

Abstract

How can computer vision systems be designed to handle unfamiliar or unexpected scenes? Many of the current systems cope quite well with limited visual worlds, by making use of specialized knowledge about these worlds. But if we want to expand these systems to handle a wide range of visual domains, it will not be enough to simply employ a large number of specialized models. It will also be important to make use of what might be called "general-purpose models." Such models can be used to suggest reasonable descriptions for a given scene, e.g., in terms of features or regions or groups of these, in the absence of specific context.