Voxel cone tracing is a GPU-friendly way to simulate realistic lighting in games in real time. Instead of tracing continuous light paths, it discretizes the scene into sparse voxel octrees. Unreal Engine 4 and NVIDIA Gameworks have already integrated it, though light can leak through thin geometry.
Dear fellow scholars, this is two minute papers with I consider this one to be one of the most influential papers in the field of light transport. Normally, to create a photorealistic image we have to create a digital copy of a scene and simulate the paths of millions of light rays between the camera and the light sources. This is a very laborious task that usually takes from minutes to hours on a complex scene, noting that there are many well known corner cases that can take up to days, as well as the rays of light can bounce around potentially indefinitely. And if we add that realistic materials and detailed scene geometry descriptions are not easy to handle mathematically, it is easy to see why this is a notoriously difficult problem. Simulating light transport in real time has been an enduring problem and is still not solved completely. For every possible material model and light transport effect, however, voxel cone tracing is as good of a solution as one can wish for. At the moment. The original formula of the problem is continuous, which means that rays of light can bounce around in infinitely many directions. And the entirety of this digital world is considered to be a continuum. If we look at the mathematical formulation, we see infinities everywhere we look. If you would like to learn more about this, I am holding a master level course at the Technical University of Vienna, the entirety of which is available on YouTube. As always, the link is available in the video description. For the more curious fellow scholars out there, if we try to approximate this continuous representation with tiny, tiny cubes, we get a version of the problem that is much less complex and easier to tackle. If we do this well, we can make it adaptive, which means that these cubes are smaller where there's a lot of information, so we don't lose out on many details. This data structure we call a sparse voxel octree. For such a solution, mathematicians like to say that this technique works on a discretized version of the continuous problem. And since we are solving a vastly simplified version of the problem, the question is always whether this way we can remain true to the original solution. And the results show beauty unlike anything we've seen in computer game graphics. Just look at this absolutely amazing footage. Ice cream for my eyes. And all this runs in real time on your consumer graphics card. Imagine this in the virtual reality applications of the future. The My goodness, I've chosen the absolute best profession. Also, this technique maps really well to the graphical card and is already implemented in Unreal Engine 4. And Nvidia has a framework Gameworks where they are experimenting with this in their project by the name vxgi I had a very pleasant visit at Nvidia's gameworks lab in Switzerland not so long ago. Friendly greetings to all the great and fun people in the team. Some kinks still have to be worked out. For instance, there are still issues with light leaking through thin objects. Beyond that, the implementation of this algorithm contains a multitude of tiny little distinct elements. It is indeed true that many of the elements are puzzle pieces that are interchangeable and can be implemented in a number of different ways. And that's likely one of the reasons why Nvidia and others are still preparing their implementation for widespread industry use. Soon we'll be able to solidify the details some more and see what the best practices are. I cannot wait to see this technique appear in the video games of the future. Note that this and future episodes will be available in 4K resolution for a significant bump in the visual quality of the series. It takes a ton of resources to produce these videos, but is now possible through the support of you fellow scholars on Patreon. There's a more detailed write up on that. I've included it in the video description. Thank you so much for supporting the show throughout 2016 and looking forward to continuing our journey together in 2017. Thanks for watching and for your generous support and I'll see you next time.
