“Microfacet Models for Refraction through Rough Surfaces,” a 2007 paper by Steve Marschner and Bruce Walter has won a test of time award from the Eurographics Symposium on Rendering convened this year in Saarbrücken, Germany and virtually. Coauthored with Hongsong Li and Kenneth E. Torrance, the paper appeared in Rendering Techniques (2007) from Springer-Verlag. In a video recording of the awards ceremony, Elmar Eisemann emphasizes that the award is not simply for a very high citation count over time (more than a decade), but also for the paper's specific achievements as a "technique or method that explains how to generalize standard reflection microfacet theory to refraction."
Steve Marschner is Professor in the Computer Science Department; he received his Ph.D. from Cornell in 1998 and joined the faculty in 2002. Bruce Walters, who also received his doctorate in Computer Science from Cornell in 1998, was advised by Donald P. Greenberg, and since 1999 has been Senior Research Associate in the Program of Computer Graphics. Hongsong Li received his Ph.D. in Computer Graphics from Cornell in 2005. Ken Torrance, a member of the Cornell faculty for more than forty years, was the Joseph C. Ford Professor of Mechanical and Aerospace Engineering at the time of his death in 2010; in 1994, he received the ACM SIGGRAPH Computer Graphics Award.
CS News asked Marschner about his sense of the importance of the paper, in short, why he thinks it has stood the test of time. With customary graciousness, his first thoughts were of his collaborators:
Ken Torrance, as part of his Ph.D. thesis research and later as one of the very earliest physics-based reflection models for computer graphics, was instrumental in developing an approach to modeling light reflection from surfaces known as the Microfacet model. The Cook-Torrance model (coauthored with Rob Cook, then a Cornell student and later a founder and decades-long technical leader of Pixar) was one of the banner early contributions of the Cornell Program of Computer Graphics in the early 80s. Ken continued working with us on reflectance modeling and measurement up until he died in 2010.
And now your collaborative work in "Microfacet Models" has become an industry standard.
Bruce Walters did a really great job writing up the math, including all the details needed for implementation, so that people used this paper as a reference for how to implement microfacet models at a time they were becoming the leading type of reflection model in the computer graphics industry.
In fact, the Cornell team had to invent a new mathematical description of surface roughness, correct?
Because we made measurements to confirm the model, and found it was hard to get the classic version of the model to fit the data, we introduced a different mathematical description of surface roughness, which we named GGX after the measured sample that motivated it. (GGX stood for "ground glass, unknown" because it was a ground glass sample we bought from a glass shop but did not know how it had been prepared.) The visual effects industry really liked the subtly different look that this model gave and it's become the default in most applications today.
When asked about the origins and subsequent influence of the research, Walter, like Marschner, thinks first of the team and their generative collaboration:
It is really surprising this paper has turned out to have so much longevity and impact. It started with a casual lab conversation I had with Hongsong. He mentioned that he was interested in rough refraction but hadn't found any satisfactory models. I thought perhaps microfacet theory (which I had learned about in part thanks to Ken Torrance's work) might provide a good framework and spent a few weeks (or months?) working out the math. The resulting model looked nice, but I had no idea if it matched the appearance of any real materials. We had a number of material measurement devices, but nothing that could measure rough refraction. So I talked to Steve and he came up with cool measurement schemes where we bonded hemispherical lens to surfaces to be able to measure the refraction.
What were your initial findings?
When we compared the measured data to the model, we found that the general behavior seemed right, but for some of the samples, especially the ground glass, the estimated normal distribution did not fit the then standard Beckmann model well. I spent a little time playing with other simple functions until I found one that fit the ground glass sample well and called it GGX (for ground-glass-unknown). I added it to the paper as it improved some of our fits, but it was almost an afterthought. But you never know how things will be received, and it turns out that GGX is perhaps the thing that the paper is most remembered for. Much of that achievement is also thanks to later researchers, such as Brent Burley at Disney, who further tested, adopted, and popularized GGX.
Looking back now, what is your key takeaway?
That sometimes the best work comes from finding interesting problems and ideas and following them wherever they may lead. This was originally just a small side project for each of us that we personally got excited about and pursued. I am gratefully that it turned out so well and that so many other people in the community have found it useful. Another person I’d like to thank is Don Greenberg, the director of the Cornell Program of Graphics. I was (and still am) a Research Associate in his lab, and he has always encouraged a very open-ended and collaborative approach to research, that enabled these kinds of projects to flourish.
In related news, read about two other papers by Steve Marschner that won test of time awards.