Volumetric path tracing makes smoke and haze look realistic in movies, but naive simulations are expensive. A technique called equiangular importance sampling focuses computation on the light paths that matter most, cutting rendering time dramatically. Marcos Fajardo and the Arnold Renderer team won a Technical Oscar for this work.
Dear fellow scholars, this is two minute papers with how do people create these beautiful computer generated images and videos that we see in Hollywood blockbuster movies? In the world of light simulation programs to obtain a photorealistic image, we create a digital copy of a scene, add the camera and the light source, and simulate the paths of millions of light rays between the camera and the light sources. The this technique we like to call path tracing, and it may take several minutes to obtain only one image on a powerful computer. However, in these simulations, the rays of light are allowed to bounce off of the surface of objects. In reality, many objects are volumes where the rays of light can penetrate their surface and scatter around before exiting or being absorbed. Examples include not only rendering amazingly huge smoke plumes and and haze, but all kinds of translucent objects like our skin, marble wax and many others. Such an extended simulation program we call not path tracing, but volumetric path tracing, and we can create even more beautiful images with it. However, this comes at a steep price if the classical path tracing took several minutes per image. This addition of complexity often bumps up the execution time to several hours. In order to save time, we have to realize that not all light paths contribute equally to our image. Many of them carry barely any contributions, and only a tiny, tiny fraction of these paths carry the majority of the information that we see in these images. So what if we could create algorithms that know exactly where to look for these high value light paths and systematically focus on them? This family of techniques we call important sampling methods. These help us finding the regions where light is concentrated, if you will. This piece of work is an excellent new way of doing importance sampling for volumetric path tracing. And it works by identifying and focusing on regions that are most likely to scatter light. And it beats the already existing importance sampling techniques with ease. Now to demonstrate how simple this technique is. A few years ago, during a discussion with one of the authors, Marcos Fajardo, I told him that I would implement their method in real time on the graphical card in a smaller than 4 KB program. So we made a bet. 4 kilobytes is so little we can store only a fraction of a second of MP3 music in it. Also, this is an empty file generated with Windows Word. Apparently, in some software systems, the definition of nothing takes several times more than four kilobytes. And after a bit of experimentation, I was quite stunned by the results because the final result was less than 2 kilobytes. Even if support for some rudimentary animations is added, the whole computer program that executes volumetric path tracing with this equiangular importance sampling technique fit on your business card twice. Absolute insanity I've put a link discussing some details in the video description. Now don't be fooled by the simplicity of the presentation here. The heart and soul of the algorithm that created the rocket launch scene in Men in Black 3 is the same as this one. Due to legal reasons, it is not advisable to show it to you in this video, but this is fortunately one more excellent reason for you to have a look at the paper. As always, the link is available in the video description. This is my favorite kind of paper where there are remarkably large gains to be had and it can be easily added to pretty much any light simulation program out there. I often like to say that the value over complexity ratio is tending towards infinity. This work is a prime example of that. By the way, Marcos and his team recently won a Technical Oscar award not only for this, but for their decades of hard work on their Arnold Renderer, which is behind many, many Hollywood productions. I've put a link to their website in the video description as well. Have a look. Congrats guys. Thanks for watching and for your generous support and I'll see you next time.
