As computer graphics rendering becomes more realistic, it is increasingly critical to our economy, safety, and daily lives: from diagnosing illnesses, to self-driving cars, and to designing, visualizing and manufacturing products. These application domains are fueled by rendering algorithms which simulate how light sources emit photons that then scatter around in a scene before making their way into a camera (or our eyes) to form a virtual image. Unfortunately, the number of such photons that even the fastest computers are able to simulate is so tiny in comparison to nature that rendering high-quality images remains an incredibly time-consuming enterprise. This project will develop new ways of expressing numerical light transport simulations which are not restricted to operate in direct analogy to nature, thereby generalizing a broad range of rendering algorithms used in computer graphics so they can operate more efficiently. Project outcomes will transform the field by fundamentally changing the definition of Monte Carlo integration, and by providing a suite of new tools to design, implement, and analyze such algorithms, with far-reaching and broad impact on applications such as those mentioned above and many others. An integrated educational and outreach program will leverage ubiquitous familiarity with light to teach otherwise abstract mathematical and physical concepts to students from diverse educational and socio-economic backgrounds.
Accurate and efficient light transport simulation has always been a central problem of computer graphics, and it continues to be challenging because the currently prevailing Monte Carlo (MC) algorithms operate in direct analogy to nature by randomly point-sampling paths from the light to the sensor. While intuitive, these algorithms are too slow because each sample contributes little to the answer so that many samples are required, but even the fastest computers cannot hope to compete with the computational speed of nature in the foreseeable future. This project will address that challenge by: establishing a generalized theory of MC integration enabling new rendering algorithms that simulate light more efficiently than nature by leveraging higher-order samples such as lines (1D), planes (2D), and beyond (nD); by developing the necessary tools to analyze and optimally leverage such samples in general MC integration; and by applying these benefits to inverse light transport problems. Project outcomes will have far-reaching impact in all application domains relying on computer graphics rendering, as well as in other fields that rely on MC integration and physically based light transport in general.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
