Károly Zsolnai-Fehér explains Deep Opacity Maps, a 2017 technique that renders realistic hair and fur in real time by using adaptive layers to capture how light gets blocked. Instead of flat layers, the method places them based on actual distance from the light source, producing accurate shadows with fewer computational steps.
Dear fellow scholars, this is two minute papers with zsolnai. In earlier episodes, we've seen plenty of video footage about hair simulations and rendering. And today we are going to look at a cool new technique that produces self shadowing effects for hair and fur. In this image pair, you can see this drastic difference that shows how prominent this effect is in the visual appearance of hair. Just look at that. Beautiful. But computing such a thing is extremely costly because since we have a dense piece of geometry, for instance hundreds of thousands of hair strands, we have to know how each one occludes the other ones. This would take hopelessly long to compute. To even get a program that executes in a reasonable amount of time, we clearly need to simplify the problem further. An earlier technique takes a few planes that cut the hair volume into layers. These planes are typically regularly spaced outwards from the light sources, and it is much easier to work with a handful of these volume segments than with the full geometry. The more planes we use, the more layers we obtain and the higher quality results we can expect. However, even if we can do this in real time, we will produce unrealistic images when using around 16 layers. Well, of course we should then crank up the number of layers some more. If we do that, for instance, by now using 128 layers, we can expect better quality results. But we'll be able to process an image only twice a second, which is far from competitive. And even then, the final results still contain layering artifacts and are not very close to the ground. Truth. There has to be a better way to do this. And with this new technique called Deep Opacity Maps, these layers are chosen more wisely. And this way we can achieve higher quality results with using only three layers and it runs easily in real time. It is also more memory efficient than previous techniques. The key idea is that if we look at the hair from the light source's point of view, we can record how far away different parts of the geometry are from the light source. Then we can create the new layers further and further away according to this shape. This way, the layers are not planar anymore. They adapt to the scene that we have at hand and contain significantly more useful occlusion information. As you can see, this new technique blows all previous methods away and is incredibly simple. I have found an implementation from Philip Rideout. The link to this is available in the video description. If you have found more, let me know and I'll include your findings in the video description. For the fellow tinkerers out there, the paper is ample in comparisons. Make sure to have a look at that too. And sometimes I get some messages saying Caroj, why do you bother covering papers from so many years ago? It doesn't make any sense. And here you can see that part of the excitement of Minute Papers is that the next episode can be about absolutely anything. The series has been mostly focusing on computer graphics and machine learning papers, but don't forget that we also have an episode on whether we are living in a simulation or the Dunning Kruger effect and so much more. I've put a link to both of them in the video description for your enjoyment. The other reason for covering older papers is that a lot of people don't know about them and if we can help just a tiny bit to make sure these incredible works see more widespread adoption, we've done our job well. Thanks for watching and for your generous support and I'll see you next time.
