Károly Zsolnai-Fehér explains a new technique for simulating subsurface scattering, the way light travels through translucent materials like skin. Developed with Activision Blizzard and universities in Zaragoza and Vienna, the method runs efficiently enough for video games and scientific work.
Dear fellow scholars, this is two minute papers with subsurface scattering means that a portion of incoming light penetrates the surface of a material. Our skin is a little known, but nonetheless great example of that. But so are plant leaves, marble, milk or snails. To have a wackier example, subsurface scattering looks unbelievably beautiful. But at the same time, it is very expensive to compute because we have to simulate up to thousands and thousands of light scattering events for every ray of light. And we have to do this for millions of rays. It really takes forever. The lack of subsurface scattering is the reason why we have seen so many lifeless rubber looking human characters in video games and animated movies for decades now. This technique is a collaboration between the Activision Blizzard game development company, the University of Zaragoza in Spain and the Technical University of Vienna in Austria. And it can simulate this kind of subsurface light transport in half a millisecond per image. Let's stop for a minute and think about this. Earlier we talked about subsurface scattering techniques that were really awesome, but still took at least, let's say, four hours on a scene before they became useful. This one is half a millisecond per image. Almost nothing in one second. It can do this calculation 2,000 times. Now this has to be a completely different approach than just simulating many millions of rays of light, right? We can take a four hour long algorithm, do some magic, and get something like this. The first key thought is that we can set up some cool experiment where we play around with light sources and big blocks of translucent materials and record how light bounces off of these materials. Cool thing number one, we only need to do it once per material. Two, the results can be stored in an image. This is what we call a diffusion profile, and this is how it looks like. So we have an image of the diffusion profile and one image of the material that we would like to add subsurface scattering to. This is a convolution based technique, which means that it enables us not to add these two images together, but to mix them together in a way that the optical properties of the diffusion profiles are carried to the image. If we add the optical properties of an apple to a human face, it will look more like a face that has been carved out of a giant apple. A less asinine application is of course, if we mix it with the appropriate skin profile image, then we'll get photorealistic looking faces, as it is demonstrated quite aptly by this animation. This apple to skin example. By the way, you can actually try for yourself as the source code and an executable demo is also freely available for everyone to experiment with. Convolutions have so many cool applications I don't even know where to start. In fact, I think we should have an episode solely on that. Can't wait. It's going to be a lot of fun. These convolution computations are great, but they are still too expensive for real time video games. What this work gives us is a set of techniques that are able to compute this convolution not on these original images, but much smaller, tiny tiny strips which are much cheaper. But the result of the computations look barely distinguishable. Another cool thing is that the quality of the results is not only scientifically provable, but this technique also opens up the possibility of artistic manipulation. It is done in a way that we can start out with a physically plausible result and tailor it to our liking. You can see some exaggerated examples of that. The entire technique is so simple a computer program that executes it can fit on your business card. It also seems to have appeared in Blender recently. Also a big hello and shout out for the awesome people at intel who recently invited my humble self to chat a bit about this technique. If you would like to hear more about the details on how this algorithm works, I've put some videos in the description box. The most important take home message from this project, at least for me, is that it is possible to conduct academic research projects together with companies and create results that can make it to multi million dollar computer games. Also having proven results that are useful for the scientific community. Thanks for watching and for your generous support and I'll see you next time.
