Dynamic Algorithm Selection: A New Approach to the Real-Time Rendering of Complex Scenes Problem

Abstract

This paper presents a novel approach to the real-time rendering of complex scenes problem. Up to the present date, a huge number of acceleration techniques have been proposed, although most are geared towards a specific kind of scene. Instead of using a single, or a fixed set of rendering acceleration techniques, we propose the use of several, and to select the best one based on the current viewpoint. Thus, dynamically adapting the rendering process to the contents of the scene, it is possible to take advantage of all these techniques when they are better suited, rendering scenes that would otherwise be too complex to display at interactive frame rates. We describe a framework capable of achieving this purpose, consisting on a pre-processor and an interactive rendering engine. The framework is geared towards interactive applications were a complex and large scene has to be rendered at interactive frame rates. Finally, results and performance measures taken from our test implementation are given.