Component-Based Adaptive Sampling

Abstract

High-fidelity renderings of virtual environments is a notoriously computationally expensive task. One commonly used method to alleviate the costs is to adaptively sample the rendered images identifying the required number of samples according to the variance of the area thus reducing aliasing and concurrently reducing the total number of rays shot. When using ray tracing algorithms, the traditional method is to shoot a number of rays and depending on the difference between the radiance of the samples further rays may be shot. This approach fails to take into account that different components of a material reflecting light may exhibit more coherence than others. With this in mind we present a component-based adaptive sampling algorithm that renders components individually and adaptively samples at the component level, finally composting the result to produce a full solution. Results demonstrate a significant improvement in performance without any perceptual loss in quality.