An animation studio can spend days rendering a scene that features water, smoke and other substances that affect light (and its simulation) in complex ways. Now, a team led by Disney Research Zürich has developed a computational algorithm that can reduce rendering times for such scenes by a factor of up to 1,000.

Animators call substances like smoke, fog, water, mist, and dust "participating media," due to their tendency to deflect or scatter light as it travels the distance separating its source from the camera (the shafts of light visible in the upper right hand corner of the Monsters University screenshot featured above are a good example of how we perceive illumination in the presence of participating media.) Existing rendering algorithms account for participating media by randomly sampling potential paths that light might take through a scene, and then averaging the results.

But the number of potential paths is usually quite numerous, and many of these paths will often have little or no effect on the look and feel of the final animated sequence. Imagine, for example, a path that intersects with an object in the scene that blocks the light entirely, or a light source that is too distant from the camera, or separated by too much dense fog, to be seen at all. Calculating paths like these are not only a waste of time and processing power, they can also introduce unwanted effects in the animation. Getting rid of these unwanted effects, of course, winds up taking even more time, and introduces still further opportunities for rendering errors to emerge.


Above: The interaction of light and dust in a rendered scene from Zero Dark Thirty, via fxguide.

Now, a team led by Disney Research scientist Wojciech Jarosz has developed a method called "joint importance sampling" that efficiently identifies which paths of light are most likely to contribute to what the camera actually sees. As its name suggests, the algorithm helps sort out which paths are important to include when rendering the final scene, and which can be excluded. The researchers will present their findings this week at the this year's ACM SIGGRAPH conference in Hong Kong.

Researchers have looked into the use of importance sampling (IS) techniques in the past. Some of the most sophisticated methods rely on a bidirectional process that traces potential light paths not just from the light source to the camera, but from the camera back to the light source. Traditionally, however, bidirectional techniques have sampled the light-source–>camera and camera–>light-source paths independently. Jarosz's team's joint importance sampling method, in contrast, identifies potential light paths with mutual knowledge of both camera and light source locations. The result, the researchers claim, is a boost in efficiency that reduces rendering times while improving animation quality.


In the figure above, the results of the team's joint importance sampling method is contrasted with traditional techniques for both unidirectional and bidirectional path tracing. The boxes along the bottom of the images show close-up views of the scene. The relative lack of noise highlights the improved accuracy of Jarosz's team's sampling method.

"There's always going to be noise," said Jarosz in a statement, "but with our method, we can reduce the noise much more quickly, which can translate into savings of time, computer processing and ultimately money."

It could also add considerably to the creative process itself. "Faster renderings allow our artists to focus on the creative process instead of waiting on the computer to finish," Jarosz explains. "This leaves more time for them to create beautiful imagery that helps create an engaging story."

In an exhaustive two-part essay on the latest trends in the VFX industry (which goes into much greater detail than I can provide you), fxguide's Mike Seymour cites from a speech delivered by Jarosz earlier this year, and highlights how novel techniques like joint importance sampling could help pave the way to a new era of animation:

[In "The Perils of Evolutionary Rendering Research: Beyond the Point Sample,"] the keynote by Jarosz at EGSR 2013, [Jarosz] argued that the way "we approach many difficult problems in rendering today is fundamentally flawed." Jarosz put forward the case that "we typically start with an existing, proven solution to a problem (e.g., global illumination on surfaces), and try to extend the solution to handle more complex scenarios (e.g., participating media rendering)."

While he feels that this "evolutionary approach is often very intuitive," it can lead to algorithms that are significantly limited by their evolutionary legacy. To make major progress, we may have to rethink (and perhaps even reverse) this evolutionary approach." He claimed that "a revolutionary strategy, one that starts with the more difficult, more general, and higher-dimensional problem – though initially more daunting, can lead to significantly better solutions.

For more information on joint importance sampling, visit Jarosz's team's research page. For more on how joint importance sampling fits into the broader field of visual effects, I highly recommend "The State of Rendering," Seymour's comprehensive overview of the latest trends in VFX, including, most notably, the industry-wide push to achieve increasingly plausible shading and lighting.