Google+ post
Tim HuttonTim Hutton - 2012-06-07 08:50:32+0000 - Updated: 2012-06-07 08:50:32+0000
Originally shared by Sean LumlySIGGRAPH 2012 Presents A Game Changer

For those that love computer graphics, pay very close attention to this post. At this year's SIGGRAPH 2012, a technique for de-noising images  -- or removing graininess -- in real-time is to be featured called "Adaptive Manifolds for Real-Time High-Dimensional Filtering" ( While it may appear to be both a mouthful and nothing of significant consequence, it has a very real application that has the potential to turn the world of real-time computer graphics on its head.

Currently real-time 3D computer graphics (like those featured in 3D games) use a technique called rasterization. In this technique many small triangles are drawn to the screen to compose larger 3D shapes. This technique scales extremely well with performance, though while it handles geometry, light interactions (light, shadow, reflections, etc) or Global Illumination, must be calculated separately. Of course, using rasterization as a base, the computer science community has been hard at work refining and creating new techniques over the years to produce images of increasing realism, and current examples of cutting-edge 3D games are stunning. But they are still quite far from being photo-real.

Quality global illumination plays a tremendous role in the realism of a scene.

Enter path tracing. Path tracing, is a form of ray tracing that casts and bounces multiple rays around a scene in order to capture highly realistic light and shadow in a scene (high quality global illumination). It produces such high-fidelity, physically-real rendering that it is often used as a baseline upon which other technique are compared! But the technique comes at a high performance cost that modern off-the-shelf hardware is just-now  capable of handling. Despite the high computational expense, a huge benefit with path tracing is that it scales particularly well with geometry. Filling a scene with 1M or 10M triangles is unlikely to have a large effect on rendering performance!

Of course, the technique comes with a huge caveat: when progressively displayed before all of the pixels have been fully calculated (as is popular in real-time path-tracing applications), the image is very, very noisy ( This noise typically happens when something in the scene is moving. This can happen when the camera moves, a light moves, or objects in the scene move -- which are all very common in interactive 3D games and thus prevents path tracing from being used despite its quality and scaling benefits.

Thankfully, "Adaptive Manifolds" allows a jaw dropping reduction in noise commonly caused by a partially converged render. This means that the noisy images that path tracing produces when there is movement in a scene can be cleaned up significantly and in real-time. In fact the paper (and accompanying video) highlights the techniques effectiveness specifically with path-tracing when used in conjunction scene-geometry information.  The bottom line? 

Real-time, photo-real graphics are right around the corner.

Real-time path-tracing engines like Brigade are already accomplishing 30fps rendering speeds on year-old off-the-shelf hardware (at low resolutions), and there are tremendous amounts of optimizations yet to be done. I have been assured that the story is far from over, and high frame rates at HD resolutions should be possible toward the end of the year, and with cloud-rendering will be available to all devices very shortly. Techniques like Adaptive Manifolds shrinks that time horizon considerably 

At the point where interactive rates (think: 30 frames per second) and path tracing meet, rasterization will be bested. Real-time interactive media can begin to approach movies in quality. It looks like that point is only months away.

To watch the technical video of this technique:
To read the paper, click here:
Click here for some excellent real-time photo-real path tracing examples:

#siggraph2012   #adaptivemanifolds   #pathtracing   #photorealismImage
Shared with: Public

This post was originally on Google+