Dr. Zsolnai reviews Viper, a new technique that simulates muscle and soft-body movement in real time by adding a scale property to particles. It runs at 8-10 milliseconds per frame, fast enough for character animation, and the code is available open source.
Questions this video answers
Approximately 8–10 milliseconds per frame (≈100 frames per second).
Viperization.
Yes; the video states the source code of the project is available.
Lambda GPU Cloud; the claim is it can train ImageNet to 93% accuracy for less than $19.
lambdalabs.compapers.
Chapters
Dear fellow scholars, this is two minute papers with Dr. Zsolnai. We have showcased this paper just a few months ago, which was about creating virtual characters with a skeletal system, adding more than 300 muscles and teaching them to use these muscles to kick, jump, move around and perform other realistic human movements. It came with really cool insights as it could portray how increasing the amount of weight to be lifted changes what muscles are being trained during a workout. These agents also learned to jump really high, and you can see a drastic difference between the movement required for a mediocre jump and an amazing one beyond that. It showed us how these virtual characters would move if they were hamstrung by bone deformities, a stiff ankle, or muscle deficiencies, and watch them learn to walk. Despite these setbacks, we could even have a look at the improvements after a virtual surgery takes place. So now how about an even more elaborate technique that focuses more on the muscle simulation part? The ropes here are simulated in a way that the only interesting property of the particles holding them together is position. Cosserat rod simulations are an improvement because they also take into consideration the orientation of the particles and hence can simulate twists as well. And this new technique is called Viper and adds a scale property to these particles and hence takes into consideration stretching and compression. What does that mean? Well, it means that this can be used for a lot of muscle related simulation problems that you will see in a moment. However, before that, an important part is inserting these objects into our simulations. The cool thing is that we don't need to get an artist to break up these surfaces into muscle fibers. That would not only be too laborious, but of course would also require a great deal of anatomical knowledge. Instead, this technique does all this automatically, a process that the authors call Viperization. So in goes the geometry and out comes a nice muscle model. This really opens up a world of really cool applications. For instance, one such application is muscle movement simulation. When attaching the muscles to bones, as we move the character, the muscles move and contract accurately. Two, it can also perform muscle growth simulations. And three, we get more accurate soft body physics. Or in other words, we can animate gooey characters like this octopus. Ok, that all sounds great, but how expensive is this? Do we have to wait a few seconds to minutes to get this? No, no, no, not at all. This technique is really efficient and runs in milliseconds, so. So we can throw in a couple more objects. And by couple, a computer graphics researcher always means a couple dozen more. Of course. And in the meantime, let's look carefully at the simulation timings. It starts from around 8 to 9 milliseconds per frame, and with all these octopi we are still hovering around 10 milliseconds per frame. That's 100 frames per second, which means that the algorithm scales with the complexity of these scenes really well. This is one of those rare papers that is written both very precisely and it is absolutely beautiful. Make sure to have a look in the video description. The source code of the project is also available. With this I hope we'll get even more realistic characters with real muscle models in our computer games and real time applications. What a time to be alive. This episode has been supported by Lambda. If you if you're a researcher or a startup looking for cheap GPU compute to run these algorithms, check out Lamda GPU Cloud. I've talked about Lambda's GPU workstations in other videos and am happy to tell you that they are offering GPU Cloud services as well. The Lambda GPU Cloud can train ImageNet to 93% accuracy for less than $19. Lambda's web based IDE lets you easily access your instance right in your browser. And finally, hold on to your papers because the Lambda GPU Cloud costs less than half of AWS and Azure. Make sure to go to lambdalabs.compapers and sign up for one of their amazing GPU instances today. Our thanks to Lambda for helping us make better videos for you. Thanks for watching and for your generous support and I'll see you next time.
