Cost prediction for global illumination using a fast rasterised scene preview

Abstract

The media industry is demanding increasing fidelity for their rendered images. Despite the advent of modern GPUs, the computational requirements of physically based global illumination algorithms are such that it is still not possible to render high-fidelity images in real time. The time constraints of commercial rendering are such that the user would like to have an idea as to just how long it will take to render an animated sequence, prior the actual rendering. This information is necessary to determine whether the desired quality is achievable in the time available or indeed if it is possible to afford to carry out the work on a render farm for example. This paper presents a comparison of different pixel profiling strategies which may be used to predict the overall rendering cost of a high fidelity global illumination solution. A fast rasterised scene preview is proposed which provides a more accurate positioning and weighting of samples, to achieve accurate cost prediction.

A Multiagent Architecture for 3d Rendering Optimization

“…This analysis is carried out in a fast way independently of the rendering algorithm used. For each submitted 3D scene (Figure 4a), the Analyst generates an importance map as shown in Figure 4b to determine what parts of the scene are complex and what parts are not (see Gillibrand et al 2005Gillibrand et al , 2006 for similar approaches).…”