An advance in the algorithms used to model images of fog, smoke and water, due to be presented at an association of professional and academic programmers, comes not from a renowned computer-science Ph.D. program or videogame developer.
It comes from Disney.
Mathematicians and computer scientists at Disney Research in Zurich, Switzerland led a team of academic researchers in developing a method, known as joint importance sampling, which helps animators clean the noise from digital images far more easily.
The technique, which will be presented at the Association for Computing Machines SIGGRAPH Asia 2013 Nov. 19-22 in Hong Kong, is an improvement over existing algorithms based on the mathematics describing how light rays are deflected or scattered when they bounce off solid objects, and how that behavior changes when light passes through aerosols or liquids.
The new algorithms reduce the amount of compute time it takes to produce noise-free images within complex scenes by one-tenth, one one-hundredth or even one one-thousandth of the current standard, according to Wojciech Jarosz, a research scientist at Disney Research, Zurich.
“Faster renderings allow our artists to focus on the creative process instead of waiting on the computer to finish,” Jarosz said. “This leaves more time for them to create beautiful imagery that helps create an engaging story.”
The new code involves so-called Monte Carlo algorithms, which make the animation of light far quicker by extrapolating all the possible routes a ray of light could take through a scene – after analyzing only a random sample of data on the origin, intensity and other characteristics of the light near its source – and average the results to create an overall effect.
But Monte Carlo algorithms analyze all the possible paths of light through a scene, not only those that would actually be present, or which the animator prefers to emphasize. Disney’s approach expands on the Monte Carlo algorithms by identifying rays of light that would be absent from a scene after being blocked by a solid object or because the source of the light was too far away from the camera.
Rather than sampling data about the light only at its source, Disney’s approach is to sample the data at locations within the scene chosen because of their visual importance or as a landmark to identify paths with a particularly heavy load of light, those that intersect in important ways with the image of characters in a scene, or other factors.
Being able to choose which paths of light to optimize to eliminate visual noise or static makes the animation process much more efficient, and reduces the noise in the image, Jarosz said in a statement announcing the project’s results: “There’s always going to be noise, but with our method, we can reduce the noise much more quickly, which can translate into savings of time, computer processing and ultimately money.”
The Walt Disney Company launched Disney Research in 2008 to extend a similar organization at Pixar Animation Studios, which Disney bought in 2006.
Image: Walt Disney Co.