Action Representation and Purpose: Re-evaluating the Foundations of Computational Vision

The traditional goal of computer vision, to reconstruct, or recover properties of, the scene has recently been challenged by advocates of a new purposive approach in which the vision problem is defined in terms of the goals of an active agent. In the starkest light the debate can be characterized as one about the role of explicit representations. The extreme traditionalists strive for a detailed representation of the 3D world while the other extreme adopts a strict behaviorist stance which eschews representations in favor of “direct sensing.” This panel will explore the roles of action, representation, and purpose in computer vision and, in doing so, will hopefully discover areas of agreement.