Case Studies

Photorealistic Rendering

Using Monte Carlo Statistical Analysis Techniques, Frontier® Produces High-Quality Visual Images

Goal

Render photorealistic images using Monte Carlo methods in conjunction with physically-based light transport analysis.

Business

Film and Television Production Studios, Architectural Firms, Computer Game Developers

Industry

Computer Graphics and Animation

Challenge

Currently, in order to approach the realism of photographic images and video, animators in state-of-the-art film and television production studios use a variety of techniques to imitate the natural behavior of light when generating images and animations in order to give them a more realistic appearance. These techniques are labor-intensive and hence time-consuming and expensive, and are limited in their ability to achieve photorealism.

To generate the initial graphics, they use local illumination algorithms which don't have the capacity to account for a variety of physical properties of objects and the interplay of light between them, such as the glow cast by a brightly colored object onto those around it, or the play of light reflected off the surface of a river on a bright day. The human eye is an expert in how people and objects look in the real world, and images missing these subtle cues can be immediately spotted as fake and plastic.

In order to compensate for this shortcoming, studios have to enhance their images by imitating such effects, for instance by placing "invisible" lamps to simulate indirect lighting, "painting" textures to give the desired appearance, or fabricating reflections and shadows. While these tweaks improve computer-generated images, they're still a far cry from the appearance of naturally occurring objects. The challenge, therefore, is to virtually render photorealistic images using lighting models based on physics, and to do this within a reasonable timeframe.

Solution

Until now, computational capability hasn't been robust enough to provide the capacity necessary to generate photorealistic graphics in a practical manner. Drawing on the massive and flexible computational power of the Frontier Grid Platform, Parabon has developed revolutionary approaches to computer rendering. Parabon's research application is able to render breathtaking images that surpass the realism of those created by these state-of-the-art production studios, and does so in a more time- and cost-efficient manner. The application uses statistical mathematics, and a robust geometric analysis library, to generate photorealistic images.

The rendering application duplicates the natural world in a digital environment by following the photons as they are reflected — according to the laws of physics — and recording their interactions with objects in the image, using Monte Carlo algorithms to build up an internal "picture" of the scene, photon by photon. After enough photons have been tracked, the image begins to come to life.

Less advanced algorithmic approaches to rendering consider light and its effects, but do so in an unsophisticated way; light is treated as though it only travels along a straight path from the source an object's surface. But in the physical world, light emanates and bounces from all surfaces; Photons are reflected from even the "darkest" object in a room. If you have a frame with a wine glass in it, the glass and the wine itself absorb some of the light, reflects some, and scatters the rest. These processes are determined by natural laws, and can be closely duplicated in algorithms, but these algorithms require extreme computational power. The rendering application applies the full capabilities of the Frontier Grid Platform to this problem, and in so doing makes true physically-based rendering practical for the first time.

Results

No longer are Monte Carlo rendering algorithms out of reach of a movie studio that wants to create breathtakingly realistic special effects or an architectural firm that wants to show clients exactly how the atrium of a new building will look in the first rays of morning light. Parabon's rendering application automatically includes such effects as indirect lighting, reflections and refractions, soft shadows, and the nearly impossible-to-obtain effect known as caustics in every image it generates. By splitting a rendering job into many small, independent tasks and allowing desktop computers to render portions of an image, the computational power of the Frontier Grid Platform eliminates the final barrier to high-quality photorealistic rendering.

Future Endeavors

Parabon's research in photorealistic rendering has demonstrated how high-quality visual images may be produced efficiently given the capabilities and constraints of a distributed, heterogeneous system. The lessons learned here may be applied not only to produce highly realistic images for film and television effects, video games, and similar needs, but moreover to a diverse set of applications in such fields as architecture (lighting analysis), physics and materials research (energy transport simulations), telecommunications (network coverage analysis), military intelligence and tactical planning (line-of-sight analysis), and more. Even more significantly, Parabon has demonstrated that a grid environment can be extremely conducive to efficient and reliable execution of Monte Carlo algorithms, which are in wide use from the financial industry to environmental modeling.

Parabon's Watchman application also utilizes the geometric analysis library originally developed as part of our rendering research, but applies the algorithms it contains to the problem of optimal sensor placement in urban and mountainous environments. By analyzing lines of sight in a given environment, Watchman is able to automatically determine where to place cameras, radars, and similar sensors in order to maximize coverage of a particular area of interest using as few sensors as possible. This capability can be utilized for border control, physical facility security and military intelligence.