Dr. Zsolnai shows seven years of snow simulation progress, from early attempts to realistic physics. The improvements focus on computational speed and how snow actually behaves when you interact with it.
February 5, 2021.
A Disney paper from 2013.
About half an hour per frame in prior works.
A few seconds per frame for smaller scenes.
A few minutes per frame (for millions of particles).
Dear fellow scholars, this is Two Minute Papers with Dr. Zsolnai. Let's talk about snow simulations. Being able to simulate snow on our computers is not new. It's been possible for a few years now. For instance, this legendary Disney paper from 2013 was capable of performing that. So why do researchers write new papers on this topic? Well, because the 2013 paper had its limitations. Lets talk about them while looking at some really sweet footage from a new paper. Limitation number one for previous works is that their snow simulations were sinfully expensive. Most of them took not seconds and sometimes not even minutes, but half an hour per frame. Yes, that means all nighter simulations. They were also not as good in fracturing interactions. Simulating powder snow and avalanches were also out of reach. Until now. You see, simulating snow is quite challenging. It clumps up, deforms, breaks and hardens under compression. And even my favorite phase change from fluid to snow. These are all really challenging to simulate properly. And in a moment you will see that this new method is capable of even more. Beyond that, lets start with friction. First we turn on the snow machine and then engage the wipers. That looks wonderful. And now may I request seeing some tire marks. There you go. This looks really good. So how about taking a closer look at this phenomenon. VB Here is the boundary friction coefficient and it is a parameter that can be chosen freely by us. So let's see what that looks like. If we initialize this to a low value, we'll get very little friction. And if we crank this up, look, things get a great deal more sticky. A big clump of snow also breaks apart in a spectacular manner, also showcasing compression and fracturing beyond the boundary friction effect we just looked at. Oh my. This is beautiful. Ok, now this is a computer graphics paper. If you're a seasoned fellow scholar, you know what this means. It means that it is time to put some virtual bunnies into the oven. This is a great example for rule number one for watching physics simulations, which is that we discuss the physics part and not the visuals. So why are these bunnies blue? Well, let's chuck them into the oven and find out. Aha. They are color coded for temperature. Look, they start from minus 100 Celsius. That's the blue. And we see the colors change as they approach zero degrees Celsius. At this point they don't yet start melting, but they are already falling apart. So it was indeed a good design decision to show the temperatures because it tells us exactly what is going on here. Without this we would be expecting melting well, can we see that melting in action too?
You bet.
Now hold on to your papers and bring forth the soft isomat. This machine can not only create an exquisite dessert for for computer graphics researchers, but also showcases the individual contributions of this new technique. One by one. There is the melting. Yes, add a little frosting and there you go. Bon appetit. Now, as we feared, many of these larger scale simulations require computing the physics for millions of particles. So how long does that take? When we need millions of particles, we typically have to wait a few minutes per frame. But if we have a smaller scene, we can get away with these computations in a few seconds per frame. Goodness, we went from hours per frame to seconds per frame in just one paper. Outstanding work. And also, wait a second. If we are talking millions of particles, I wonder how much memory it takes to keep track of them. Let's see. Whoa. This is very appealing. I was expecting a few gigabytes, yet it only asks for a fraction of it. A couple hundred megabytes. So with this hefty value proposition, it is no wonder that this paper has been accepted to the Siggraph conference. This is the Olympic gold medal of computer graphics research, if you will. Huge congratulations to the authors of this paper. This was quite an experience. What a time to be alive.
This episode has been supported by Lamda. GPU Cloud if you're looking for inexpensive cloud GPUs for AI, check out Lambda GPU Cloud. They've recently launched Quadro RTX 6000, RTX 8000 and V100 instances. And hold on to your papers because Lambda GPU Cloud can cost less than half of AWS and Azure. Plus they are the only cloud service with 48 gigabyte RTX 8000. Join researchers at organizations like Apple, MIT and Caltech in using Lambda Cloud instances, workstations or servers. Make sure to go to lambdalabs.compapers to sign up for for one of their amazing GPU instances today. Our thanks to Lamda 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.
