Why Gamers Will Never See Hair The Same Way Again
6:35

Why Gamers Will Never See Hair The Same Way Again

Two Minute Papers 29.09.2025 140 322 просмотров 9 638 лайков

Machine-readable: Markdown · JSON API · Site index

Поделиться Telegram VK Бот
Транскрипт Скачать .md
Анализ с AI
Описание видео
❤️ Check out the Fully Connected Conference by Weights & Biases - https://wandb.me/fclon2025-2min 20% discount code: FCLON2025-2MIN 📝 The paper is available here: https://www.cemyuksel.com/research/hairmesh_rendering/ Try the demo and try to break it, it is super fun: https://www.cemyuksel.com/research/hairmesh_rendering/demo/ 📝 My paper on simulations that look almost like reality is available for free here: https://rdcu.be/cWPfD Or this is the orig. Nature Physics link with clickable citations: https://www.nature.com/articles/s41567-022-01788-5 🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible: Benji Rabhan, B Shang, Christian Ahlin, Gordon Child, John Le, Juan Benet, Kyle Davis, Loyal Alchemist, Lukas Biewald, Michael Tedder, Owen Skarpness, Richard Sundvall, Steef, Sven Pfiffner, Taras Bobrovytsky, Thomas Krcmar, Tybie Fitzhugh, Ueli Gallizzi If you wish to appear here or pick up other perks, click here: https://www.patreon.com/TwoMinutePapers My research: https://cg.tuwien.ac.at/~zsolnai/ X/Twitter: https://twitter.com/twominutepapers Thumbnail design: Felícia Zsolnai-Fehér - http://felicia.hu

Оглавление (2 сегментов)

Segment 1 (00:00 - 05:00)

This is an amazing piece of work, but it's not about what you think. It's not about the physics and movement of hair, which is what we often do around here. No, no. This is about how to store hair, or even better, how to get an entire army of teapotss with glorious 80s rock band hair flapping around like it's a shampoo commercial. And this research work also describes how to render all this on a computer. super quickly. Now, hold on to your papers, fellow scholars, because all this is done in two milliseconds per frame, not 2 milliseconds per character. No, 2 milliseconds for all 100 of them. Yes, that is 500 frames per second. Wow. And you can try it yourself, too. And the geometry. Now, that is where things get even crazier. This only needs about 18 kilobytes per model. What? You know how small that is? That is approximately the same storage needed as 1 second of music. That is insanity. At this point, I really wanted to know more and looked into the research paper. And boy, is it amazing. So, how on earth does it do all this so quickly and so efficiently? Well, listen closely now because it uses no AI whatsoever, only human brilliance. Here's how. Dear fellow scholars, this is two minute papers with Dr. Carola. Dr. Carol H. If you look at the title, it says hair mesh rendering. What do you mean hair mesh rendering? Meshes are a terrible way of storing and rendering hair. You see, hair is made of countless individual thin strands while meshes are collections of flat polygons. This is a fundamentally wrong tool for the job because you'd need an astronomical number of tiny polygons to represent each strand that is heavy on storage and rendering. So, how on earth did they solve this? Well, this paper gets around this by not using the mesh to be the hair, but to act as a hair growing pot from which strands are generated on the fly on the GPU. The core idea is brilliantly simple. Instead of storing millions of individual hair strands, which is a data nightmare, instead they store a much simpler thing that they call the hair mesh that defines the overall volume and the flow of the hairstyle. Think of it like the blueprint for the hair. This blueprint is then converted into a special 3D texture. When it's time to render, the GPU uses this texture map to generate all 100,000 hair strands from scratch and in real time. It's like having a super efficient hair factory on the graphics card that reads the blueprint and creates a new complex hairstyle for every frame. And here's the best part. After the frame is rendered, the data for all those strands is simply discarded, saving a massive amount of memory. And it gets better because the hair is generated on the fly. They can easily implement level of detail. As a character moves further away, the system automatically generates fewer thicker strands, reducing the geometric complexity. And super nice. I think there might be no noticeable drop in quality. The left shows what you would see in the game and the right shows what is secretly being rendered under the hood, but you never see this one. It's just a bunch of noodles. Now, a huge performance saver that works seamlessly. What a time to be alive. Now, these videos flicker a bit. We all see that. I think that has to do more with the rendering technique than with the geometry. But I wanted to make a note of that. But wait, wait. If all this is really that fast, that means that it can generate this kind of hair geometry on the fly. Well, prove it then. Oh my, they put out a demo that can do all this kind of craziness in real time. You can try it yourself too in the video description. Okay, so here is how the demo works. First, well, grow some hair, dude. Okay, this will do. And there are a bunch of parameters we can play with. And if we continue like this, we can make this gentleman into a rockstar metal musician. And as he ages out of it, and as we play with the physical parameters, he quickly grows out of it. And he has now entered a bit of a conductor phase. I played with it too

Segment 2 (05:00 - 06:00)

here. And since it gives you a lot of freedom, kind of too much freedom, I recommend that you also play with it and try to break it. Now, a limitation of this technique is that this only works if the hairstyle is built with their special mesh system. But it is absolutely brilliant. Wow. And strangely, almost no one seems to be talking about this paper, even despite the fact that it's literally rendering what would otherwise be billions of triangles of hair in real time. This deserves way more attention. If you wish to help us, give these papers a voice because frankly, I am not sure if anyone else does it. Please like, subscribe, hit the bell icon, and leave a really kind comment. We've tried it before and it is the only thing that helps. So, hair on teapotss, hair on gators, faster than your graphic shaders. Subscribe to Two Minute Papers on November 4th and 5th in London. The Weights and Biases fully connected conference returns. Yes, join top AI engineers for 2 days of agent building and model training. It's a fantastic opportunity for networking and learning. Perfect for AI developers, researchers, and business leaders. Yes, that is you, fellow scholars. And here's the best part. You can get tickets at 20% off. Just click the link in the video description and use the code to claim your discount

Другие видео автора — Two Minute Papers

Ctrl+V

Экстракт Знаний в Telegram

Экстракты и дистилляты из лучших YouTube-каналов — сразу после публикации.

Подписаться

Дайджест Экстрактов

Лучшие методички за неделю — каждый понедельник