Using Expert Knowledge for Distributed Rendering Optimization

Abstract

The generation of virtual images, which do not differ from those taken from the real world, from an abstract description of a 3D scene is defined as Photorealistic Image Synthesis. Since achieving greater realism is the ultimate goal, the rendering of a single image may take hours or days even on powerful computers. To face this challenge, in this work we discuss the potential benefits of combining the use of expert knowledge and the adoption of a multi-agent architecture in order to optimize the rendering of complex 3D scenes. Within this context, we apply novel techniques based on the use of expert knowledge to distribute the different work units in which the input scene is divided in a balanced way, to automatically generate the rendering engine setting parameters, and to optimize the configuration of the rendering parameters given by the user. The conducted experiments demonstrate that our approach can drastically reduce the rendering time with unnoticeable quality loss.