Animators can now accurately control how microscopic particles interact with light in their rendering of objects, thanks to a research collaboration between computer scientists at Dartmouth University and staff scientists at Pixar and Disney. The team will feature this new work next week at the SIGGRAPH Asia event in Tokyo, Japan; there will also be a paper in the magazine Transactions on images.
The breakthrough gives animation artists more creative latitude in designing the appearance of different objects by giving them the ability to customize the way light travels through them. It should have the greatest impact on renderings of so-called “volumetric materials” – for example, clouds, fog, mist, skin or marble statues. (Marble is a material that reflects some light off the surface, but allows it to pass through, giving it a translucent appearance.)
“[It’s] a change as dynamic as the transition from black-and-white images to color.”
“There’s a whole host of dramatically different manifestations that artists just haven’t been able to explore until now,” said Dartmouth co-author Wojciech Jarosz. “In the past, artists basically had one control that could influence the appearance of a cloud. Now it is possible to explore a much richer palette of possibilities, a change as dynamic as the transition from black and white images to color. “
Pixar has long been at the forefront of advanced animation R&D to translate the incredibly complicated underlying math into easy-to-use software that can render highly realistic physical objects and motion. In 2014, Tony DeRose, senior scientist and head of Pixar’s Animation Studios, gave a lively TED-Ed talk about the math of Pixar movies, and he was also interviewed by Numberphile. You can even take a free online course through Khan Academy.
For example, hair is very difficult to represent realistically because there are hundreds of thousands of individual hairs all bouncing off each other. In a 2013 lecture at the Museum of Math in New York, DeRose said that Merida’s bright red curls in brave were composed of 100,000 units, meaning there were 10 billion possible collisions. You cannot use standard compression algorithms for this; you lose too much of the fine detail. So Pixar created the equivalent of a PNG or FLAC for those animations.
The latest study focuses on how light travels through different materials, interacting with the individual particles within them. For example, a cloud contains billions of individual water droplets – far too many to plot individually. Animators can specify how densely those particles are packed in different parts of the cloud, determining its shape. But they can’t dictate how those individual droplets are arranged.
This is important because nature does not necessarily randomly arrange those microparticles; for example, there could be clumps, or the properties of certain materials would cause particles to be fairly evenly spaced. “By controlling only the density, current techniques basically assume that the particles are arranged randomly, without any interdependence,” Jarosz said. “But this limitation can have a dramatic effect on the final appearance.”
To solve the problem, the researchers found inspiration in atmospheric science and models of neutron transport in physics, where it is vital to know how water droplets, or particles of radioactive material, are arranged. First, the collaboration traced how a beam of light travels through a material made up of randomly arranged microparticles. They then compared that trajectory to how a light beam travels through a more naturally ordered material.
Based on the averages of millions of runs, they came up with an accurate model for how far photons can penetrate a given material before colliding with microparticles. If you have particles clumped together, as you would in a cloud, a beam of light can travel further than through materials with a random arrangement of microparticles. Programming that model into animation software makes it easier for animators to get a realistic look.
DOI: Transactions on images2018. 10.1145/3272127.3275103 (About DOIs).