Simulating volumetric participating media visual effects with Monte Carlo methods

点击翻译按钮获取中文摘要

The realistic simulation of lighting in realistic environments has applications broadly spanning many industries: architectural visualization, product design, medical imaging, visual effects for feature films, video games, virtual reality and augmented reality. In these contexts, faithfully simulating the physical behaviour of light in volumetric participating media, like clouds or particulate fluids, is essential in order to generate realistic images and/or image sequences. Monte Carlo-based techniques rely on statistical methods to synthesize this behaviour, by simulating the scattering of many billions-to-trillions of (virtual) photons in a (virtual) environment in order to generate a single image. As such, incorporating the effects of volumetric media remains a costly computation and a continuing area of focus in the fields of realistic image synthesis and radiative transfer over the last five decades. This collaborative project will address this problem by developing novel Monte Carlo-based solutions. Specifically, these solutions will adapt and apply concepts from the computational statistics literature, leveraging the vast amount of work on variance reduction and correlated sampling, two mathematical tools that have proven successful in the context of similar problems outside of image synthesis. Our work will be incorporated into an industrial-caliber software platform, Autodesk's Arnold Rendering Engine, which will allow us to benchmark the efficacy of the techniques we develop against the state-of-the-art of existing high-end software solutions. Moreover, we will leverage and build atop of Autodesk's existing expertise in these areas, working hand-in-hand with their rendering engineers and digital content creators in order to build a larger set of testing scenarios in order to stress-test the solutions we develop. One side-effect of such rigorous testing is that we will develop more consistent and representative benchmarks for the community to base future developments on. The techniques we develop will have immediate impact on the workflows and content generation capabilities of Autodesk's product suite, starting most immediately with their Arnold Rendering Engine.