Selective Parallel Rendering for High-Fidelity Graphics

High-fidelity rendering of complex scenes is one of the primary goals of computer graphics. Unfortunately, high-fidelity rendering is notoriously computationally expensive. In this paper we present a framework for high-fidelity rendering in reasonable time through our Rendering on Demand system. We bring together two of the main acceleration methods for rendering: selective rendering and parallel rendering. We present a selective rendering system which incorporates selective guidance. Amongst other things, the selective guidance system takes advantage of limitations in the human visual system to concentrate rendering efforts on the most perceptually important features in an image. Parallel rendering helps reduce the costs further by distributing the workload amongst a number of computational nodes. We present an implementation of our framework as an extension of the lighting simulation system Radiance, adding a selective guidance system that can exploit visual perception. Furthermore, we parallelise Radiance and its primary acceleration data structure, the irradiance cache, and also use the selective guidance to improve load balancing of the distributed workload. Our results demonstrate the effectiveness of the implementation and thus the potential of the rendering framework.