Dr. Zsolnai presents a faster light-transport simulation that renders caustics and glitter in minutes instead of days. The method fixes flickering problems from incorrect curvature handling and can produce a close-up caustic in two minutes.
Dear fellow scholars, this is two minute papers with Dr. Zsolnai. I am a light transport researcher by trade and due to popular request today, I am delighted to show you these beautiful results from a research paper. This is a new light simulation technique that can create an image like this. This looks almost exactly like reality. It can also make an image like this. And have a look at this. This is a virtual object that glitters. Oh my goodness. Absolutely beautiful, right? Well believe it or not, this third result is wrong. Now you see, it is not so bad, but there is a flaw in it somewhere. By the end of this video you will know exactly where and what is wrong. So, light transport, eh? How do these techniques work anyway? We can create such an image by simulating the path of millions and millions of light rays. And initially this image will look noisy. And as we add more and more rays, this image will slowly clean up over time. If we don't have a well optimized program, this can take from hours to days to compute for difficult scenes. For instance, this difficult scene took us several weeks to compute. Ok, so what makes a scene difficult? Typically caustics and specular light transport. What does that mean? Look here we have a caustic pattern that takes many, many millions, if not billions of light rays to compute properly. This can get tricky because these are light pairs that we are very unlikely to hit with randomly generated rays. So how do we solve this problem? Well, one way of doing it is not trusting random light rays, but systematically finding these caustic light paths and computing them. This fantastic paper does exactly that. So let's look at one of those classic close ups that are the hallmark of any modern light transport paper. Lets see. Yes, on this scene you see beautiful caustic patterns under these glossy metallic objects. Let's see what a simple random algorithm can do with this. With an allowance of two minutes of rendering time. Well, do you see any caustics here? Do you see any bright points? These are the first signs of the algorithm finding small point samples of the caustic pattern. But that's about would take at the very least several days for this algorithm to compute the entirety of it. This is what the fully rendered reference image looks like. This is the one that takes forever to compute. Quite different. Right, let's allocate two minutes of our time for the new method and see how well it does. Which one will it be closer to? Can it beat the naive algorithm? Now hold on to your papers and let's see together. What? On this part it looks almost exactly the same as the reference. This is insanity. A converged caustic region in two minutes. Wow. The green close up is also nearly completely done. Now not everything is sunshine and rainbows. Look. The blue closeup is still a bit behind, but it still beats the naive algorithm handily. That is quite something. And yes, it can also render these beautiful underwater caustics as well in as little as five minutes. Five minutes. And I would not be surprised if many people would think this is an actual photograph from the real world. Loving it. Now what about the glittery origami scene from the start of the video? This one? Was that footage really wrong? Yes it was. Why? Well, look here. These glittery patterns are unstable. The effect is especially pronounced around here. This arises from the fact that the technique does not take into consideration the curvature of this object correctly when computing the image. Let's look at the corrected version and oh my goodness. No unnecessary flickering anywhere to be seen. Just the beautiful glitter slowly changing as we rotate the object around. I could stare at this all day. Now note that these kinds of glints are much more practical than most people would think. For instance, it also has a really pronounced effect when rendering a vinyl record and many other materials as well. So from now on we can render photorealistic images of difficult scenes with caustics and glitter not in a matter of days, but in a matter of minutes. What a time to be alive. And when watching all these beautiful results. If you are thinking that this light transport thing is pretty cool and you would like to learn more about it, I held a master level course on this topic at the Technical University of Vienna. Since I was always teaching it to a handful of motivated students, I thought that the teachings shouldn't only be available for the privileged few who can afford a college education, but the teachings should be available for everyone. Free education for everyone. That's what I want. So the course is available free of charge for everyone. No strings attached. So make sure to click the link in the video description to get started. We write a full light simulation program from scratch there and learn about physics, the world around us and more. This episode has been supported by Weights and Biases. In this post they show you how you can get an email or Slack notification when your model crashes. With this you can check on your model performance on any device. Heavenly Weights and Biases provides tools to track your experiments in your deep learning projects. Their system is designed to save you a ton of time and money and it is actively used in projects at prestigious labs such as OpenAI, Toyota Research, GitHub, and more. And the best part is that Weights and Biases is free for all individuals, academics and open source projects. It really is as good as it gets. Make sure to Visit them through wnb.com papers or just click the link in the video description and you can get a free demo today. Our thanks to Weights and Biases for their long standing support and for helping us make better videos for you. Thanks for watching and for your generous support and I'll see you next time.
