Image of Ken Museth
This interview was contributed to the ACM SIGGRAPH Blog by Chris Davison.
The Academy of Motion Picture Arts and Sciences (AMPAS) hosts an annual celebration known as the Scientific and Technical Awards, dedicated to honoring those whose innovations have contributed in significant and lasting ways to motion pictures. In other words, the wizards who work behind the scenes to make movie magic.
A master magician in this field is Ken Museth, who has been honored not once but twice for his wizardly work. Ken was first honored in 2014 for his concept and development of OpenVDB. Ken was then honored by the Academy a second time in 2024, for the ongoing impact of OpenVDB within the motion picture industry.
So, what is OpenVDB?
The academic answer from the Academy: “For over a decade, OpenVDB’s core voxel data structures, programming interface, file format, and rich tools for data manipulation continue to be the standard for efficiently representing complex volumetric effects, such as water, fire, and smoke.”
And an entertaining explanation courtesy of Miles Teller and Margot Robbie:
I had a chance to interview Ken about modeling and rendering clouds, GPU acceleration, the Academy Software Foundation, top-down traversals, and starting from scratch.
Chris Davison (CD): When did it first become clear that an unmet need existed, and what was your development process like?
Ken Museth (KM): Before I joined the movie industry, I was actually a full professor in computer graphics. Back then it was obvious that both researchers and practitioners alike were struggling with inefficient volumetric data structures. Basically, exciting solutions either used far too much memory (e.g., dense volumes) or were too slow (e.g., octrees or tiled grids). I worked with my Ph.D. students on this problem for a number of years, and we came up with some interesting novel volumetric data structures that outperformed existing ones. However, once I left academia and joined the movie industry (in 2007), I quickly learned that our solutions had severe limitations that basically rendered them useless for several practical applications, like rendering and random access. It was a humbling experience since I basically had to start again from scratch. First, I developed a new data structure dubbed DB+Grid that was later improved and eventually became VDB, which forms the foundation for what was open sourced as OpenVDB. Other developers joined the project over time and helped both mature and extend the library to the point where today it has become an industry standard for sparse volumetric data.
CD: What is a favorite memory of working on OpenVDB in its early stages?
KM: One of the most exciting moments in the early development of VDB was probably when I discovered an acceleration technique that has since become one of its defining and distinctive features. Traditionally accessing values in a tree data structure requires a top-down traversal, which can be both slow and inefficient. However, because of VDB’s unique tree configuration, I discovered that — surprisingly — it allows for bottom-up traversal, which is much faster. It was achieved with the help of a small perfect hash table, and its only
requirement is that data access is spatially coherent, which turns out to be the case for virtually all operations in practical volume applications. In other words, this realization suddenly made VDB much faster, while still maintaining its small memory footprint. I can still remember the adrenaline rush from that day.
CD: OpenVDB began as a proprietary offering. When did the discussions for going open source begin, and what was the workflow thereafter?
KM: It was a very long and tedious process that involved a lot of business discussions and meetings with lawyers. Remember, OpenVDB development started back in 2009 and was eventually open source in 2012 by DreamWorks Animation. A lot of people were involved in these discussions, but ultimately our CTO, Lincoln Wallen, and CEO, Jeffrey Katzenberg, gave the green light. As I recall, the winning argument was the fact that we spent a lot of our development time integrating VDB into commercial software packages like Houdini. By open sourcing it, we could leave this work to others and instead focus on improving OpenVDB itself. I’m happy to say that it worked exactly as expected. Later, OpenVDB was adopted as the very first software project by the Academy Software Foundation.
CD: “Puss in Boots” was the first feature to use OpenVDB, and your work helped make it possible. When you first saw the completed film, how did you feel?
KM: Immensely proud. I had worked tirelessly with fellow developers and artists to create a lot of new tools for modeling and rendering of clouds at resolutions and scales never seen before. It was, for sure, another highlight of this project!
CD: How do you see the next few years unfolding with OpenVDB?
KM: We’re working on several new exciting directions including GPU acceleration, multi- resolution grids, and machine learning applications. I see no signs that OpenVDB is slowing down — on the contrary actually!
CD: What excites you? What technologies are on the horizon that you are looking forward to?
KM: Currently, I’m most excited about a new project that we hope to share at this year’s SIGGRAPH conference in Denver. It combines NanoVDB, our GPU port of OpenVDB, with deep learning operations that compute spatial intelligence from very large and sparse 3D data sets, like points and voxels. For instance, we can use this technology for Generative AI at scales never seen before!
CD: What advice would you give to someone considering a scientific or technical career in the motion picture industry?
KM: First, note there is no such thing as a perfect or even ideal education for someone seeking a technical career in the motion picture industry. This is, of course, both a challenge and opportunity! Sure, an argument can be made that degrees in fields like computer science, applied mathematics, or computational physics are handy, but it’s by no means a requirement. Second, VFX exists in the intersection of many different fields, like computer graphics, art, physics, vision, machine learning, etc., and most of these can be learned outside of the classroom. In my experience, the most important qualifications are passion and an appetite for knowledge and problem solving. The movie industry is extremely diverse in both its content creation and the skillsets of its crew and, as such, there are many different entry points and career trajectories. Find the one that you’re most passionate about and then start acquiring the skills along the way. It’s a long journey with many twists and turns, so passion and a thirst for new knowledge are your best companions.
Chris Davison is a scientist, consultant, and public speaker and can be reached at davison@intellcap.com.
