NVIDIA Is Simulating 100,000 Hair Strands!
HTML-код
- Опубликовано: 16 май 2023
- ❤️ Check out Lambda here and sign up for their GPU Cloud: lambdalabs.com/papers
📝 The #NVIDIA paper "Interactive Hair Simulation on the GPU using ADMM" is available here:
research.nvidia.com/publicati...
My latest paper on simulations that look almost like reality is available for free here:
rdcu.be/cWPfD
Or this is the orig. Nature Physics link with clickable citations:
www.nature.com/articles/s4156...
🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Aleksandr Mashrabov, Alex Balfanz, Alex Haro, Andrew Melnychuk, Benji Rabhan, Bryan Learn, B Shang, Christian Ahlin, Eric Martel, Geronimo Moralez, Gordon Child, Jace O'Brien, Jack Lukic, John Le, Jonathan, Kenneth Davis, Klaus Busse, Kyle Davis, Lorin Atzberger, Lukas Biewald, Martin, Matthew Valle, Michael Albrecht, Michael Tedder, Nikhil Velpanur, Owen Campbell-Moore, Owen Skarpness, Rajarshi Nigam, Ramsey Elbasheer, Richard Sundvall, Steef, Taras Bobrovytsky, Ted Johnson, Thomas Krcmar, Timothy Sum Hon Mun, Torsten Reil, Tybie Fitzhugh, Ueli Gallizzi.
If you wish to appear here or pick up other perks, click here: / twominutepapers
Thumbnail background design: Felícia Zsolnai-Fehér - felicia.hu
Károly Zsolnai-Fehér's links:
Twitter: / twominutepapers
Web: cg.tuwien.ac.at/~zsolnai/ 
Наука
It took a single scientist to solve a problem that hasn't been solved for so many decades. Congrats to this man.
I can't fully agree with this statement, because it could not be possible without the prior tremendous work of the industry in general. Starting from hardware, ending with other researches in the field of maths and deep learning.
@UCuAgJUDCXcDIyEqGB0tYDSA it's always been like that. Every scientist stands on the shoulder of giants.
Even Einstein did. But we still give every single one of them the fully deserved credit.
The problem was solved a decade ago, this is just never gonna be put in any type of game, because the card would burn through the ground.
@@DarkCrux I think you are confusing. It's not a problem to solve the problem by emulating hair in terms of physics. But it's indeed an extremely heavy computational task. But that's what this paper is about: AI-based hair simulation, which is incomparable faster to direct physics simulation.
imagine how enlightening it must've felt to discover a solution to that problem
As I age, my hair gets "easier to simulate" while my skin gets "more photorealistic and detailed"
The Euphemism Award for today goes to you
Totally underated comment
Ai time to fix that soon hopefully.
@@warrenarnold what do you mean underated??? it was posted 2 hours ago.
@@neoruss3553 Looks like he got you dancing
Congrats to Gilles for this achievement, what a mad lad. He was also part of the team who made the LOKI physics simulation framework at Weta to create unified simulations and probably the most realistic photoreal water simulations all under once framework. Insane.
Gilles
Gilles Daviet
Gillet 😅
Iam simply amazed that the research articles for gaming technologies do exist
Yeah, the simulation is amazing and all but I never seen hair sooo well rendered in real time like with his demo. It's almost photoreal. I'm double jaw dropped.
Great... Now I'm going to have to brush my hair in video games.
@@jtjames79 Imagine living in a metaverse, and still needing to manually brush your hair
@@nagualdesign I thought I had only one, guess I was wrong. Now I'm second guessing. Who knows how many jaws I have?? Károly, keep doing this amazing job! I wonder if someday I'll find a third jaw hidden somewhere around here.
Likewise, wow.
Yeah it looks great, just needs an RTX 6090 to run it 💀
The guy that wrote this paper has been working on hair physics since 2012, especially for movies and his software was used in avatar 2 leading to him getting an oscar of technology from the oscar academy, must be nice to be intelligent...
Not just intelligence, it takes real man of culture to not waste time solving boob physics, but hair physics ( ͡° ͜ʖ ͡°)
Checkmate Japan.
Imagine working your hardest for 10+ years just to get your accomplishments dismissed with "must be nice to be intelligent" lol
@@lucasfranke5161 I’m sure he feels dismissed after getting that Oscar
Without being genious, probably he could not achieve solving those questions. Many people is hard working but being genious and being hard working is being in another level. Also in this centuary, without not being hard working, only being genious you can gain lots of things for your wealth. First, to be able to solve that kind of questions you must be genious, second you may not work so hard :)) You could be hard working too but you most probably cannot solve those problems, trust me :)
Is this really important? why dying so much to have realistic hair physics, just hire a real person if you want realistic visuals. This only puts ridiculous and very expensive pc requirements in future games. 💀🙄 graphics cvltist are out of control nowadays!
This looks pretty good. Imagine this with a proper fluid simulation so the hair can get wet and dries over time.
I believe it's totally not a problem technically to already expand this technology to different degrees of hair humidity.
Can't see it being used in games like that because we usually don't see the head/hair too close.
But to create cgi videos, cutscenes... Wow!!!
wouldnt "wetness" just be a parameter here?! 🤔
i mean, if the hairs check for collisions anyways, then having them "magnetic" sticking to eachother when close, seems like a way without opening the fluic-dynamics can of whiggly worms 😆
@@cesarkopp2 it can most games disable cloth physics from a distance and active it when close like LOD
Two Minute is kind to give a shoutout to unknown corporations like Nvidia
Looking out for the little guy
Research is research no matter who finds it
To be fair, the scientist probably feels wonderful about being recognized for his work.
Who are you and why are you everywhere? You some AI?
@@erithax I bet he has an AI automatically grab the transcripts of new videos from popular channels and generate comments for him
Thank you! This is why i subscribe to two minute papers, it’s great to see really vital ai updates, but this is the type of content i long for!
Every video like this just hypnotizes me into watching it, often multiple times. I have always been obsessed with stuff like this, and every video like this gives me a bit more hope that we might see simulated worlds on a smaller scale one day...
Hey Karoly, I really enjoyed this video on simulations, and would love love more content on it! LLMs and Text to image generators are cool, but I'd love if more videos like these popped up every once in a while. Thank you so much for your amazing content, you're always able to update me on the forefront on computer research, keep up the amazing work.
Absolutely fascinating as always 🙌
wow Big grats on this one Gilles! Looking forward to a beautiful and flowing future!
You should mention what gpus are these timings on. For the curious like me, here's an excerpt from the paper:
All simulations were performed on a workstation equipped with an
Intel Core i9-10980XE CPU and two NVIDIA GeForce 3080 Ti GPUs
with 12 GB of memory each (unless otherwise mentioned, only one
GPU was used for the simulations presented below).
Scientifically accurate visualizations! This is everything I'm about! My current work revolves around simulation science, though I'm more on the molecular physics level.
I'm really impressed by the classical models that have come out recently. The effects of static electricity being taken into account are genuinely impressive, especially the fact that the accuracy is nearly perfect! Though one thing I'm particularly interested in is the simulation with respect to air flow. Computational fluid dynamics, particularly with regards to turbulent flow, are ridiculously difficult, to the point that it's easier to simulate physically accurate quantum mechanical models with 100 atoms than it is to perform a macroscale turbulent flow simulation.
I'll be eyeing the progress on this one closely, because if they can somehow figure out how to run a turbulent flow model.
Damn I would pay for an accurate "hair cutting" simulator or sandbox. That would be insane!
But even more insane would be that method applied to fur or human hairs in videogames 😍 (maybe with a lower vertex number-)
A new era of AI is not just coming, but is already starting. Applications are innumerable. Simulating a hair cut for real-life purposes (studying to be a haircutter, chosing a new haircut for yourself, etc) could be just one tiny piece of this breakthrough.
I was just thinking of how tedious it would be to model unique braids or hairstyles in a typical modeling software this way; it'd probably be quicker to jump into a VR environment and just do the braiding/cutting yourself!
@@CyJackO imagine also adding AR to this for virtually trying a new style on someone in front of you in real-time.
@@amegatron07 what's the link between AI and simulating hair / fur physics? Where/how AI is used in your example?
@@cesarkopp2 Not sure I got your question, but the link between AI and simulating hair is just in this video and corresponding paper.
This is amazing, so much like Light transport research, hair is always another common pain point in the 3d graphics world, I can't wait to see this technique make it into modern video games and other interactive 3d media!
My mind boggles at how this is possible. I couldn't even dream of wrapping my head around it I think. Amazing!
I'd like to see a recap of different LLM architectures at the end of the year, now there are RNNs, CNNs, transformers, and diffusion models being used for text generation
Oh? People are using diffusion for text? Interesting! I wondered if something like that could be done. I’m curious as to how they did that. Do you have a name for it handy?
@@drdca8263 AR-Diffusion: Auto-Regressive Diffusion Model for Text Generation
@@JorgetePanete Thanks!
In 10 years I can see video games having hair physics like this, but probably only running with the best GPU
more like 2 years bro
@@shirowolff9147 to run this you need a high end pc and it is only running one character, video games need multiple characters. Maybe 7 year minimum
@@chrisbampton5225 yeah, but pc parts will evolve much faster each year, 10 years is too much, in 10 we'll have real androids, let alone this, thats why l said 2
Absolutely mind-blowing video! The advancements NVIDIA is making in hair simulation technology are nothing short of revolutionary. The level of detail and realism achieved in real-time is truly a testament to the power of modern graphics processing. It's fascinating to see how these complex simulations are evolving and becoming more accessible for real-world applications. Can't wait to see this technology implemented in future video games and virtual worlds. Keep up the great work, Two Minute Papers! Your content is always enlightening and a joy to watch.
Can’t wait for the new hair simulation to be exclusive to Nvidia’s newest overpriced GPUs with lackluster performance gains and low VRAM.
If the hair is basically mimicking real-life, how well does it handle tangling/matting?
Very cool and beautiful results! I have to ask though, 17ms on what system (on the GPU only I presume, on a high-end like the 4090?)? If so, it's definitely for _next-gen_ and potentially a great usecase/timesaver for film and TV. I'm sure it's faster than most other solutions out there and the first law of papers applies as always!
Real time hair simulation is pretty exciting, I kinda want to try it out myself
very cool!
Thanks for sharing
Very well done! I had hoped to see the avatar stand in a wind tunnel on full speed, or be affected by static electricity :D
Yess/finally. Physics simulation content!!
One thing that always bothered me, and I'm not sure if they've fixed it now:
A lot of soft-body/springy physics simulations really have a lot of trouble with persistent tension. You can tell with the horse's tail at 1:04, for instance, that they are re-computing even constant forces every time , so any time there is a constant force, there will be all kinds of jittering based on the physics sample time. Even though the force is constant and steady. In this case, the horse's tail should just be "down" and relatively stiff. Not jumbling around like it's being blown in the wind or is made of live worms.
Like this happens ALL THE TIME with simulated rubber bands, for instance. You pull things apart, and it just loses its mind updating its shape very quickly for a force that is changing very slowly.
I've only done this for motion controls but not physics simulations, but the idea in my case was to basically pull out constant values into an averaging accumulator, then have this accumulator respond more slowly. Then you can respond to rapid changes compared to this average. That is, you've created a new "rest" state, and you just have to worry about how things compare to this "rest" state. You aren't recalculating gravity every frame and causing a constant force to produce very non-constant motion.
Instead you'd have basically just modified the hair to point downwards constantly, then modify it based on other forces besides gravity.
In my case, I used an "accumulator" initially. But then I switched to cascaded EMAs to pull out frequency components. The lowest frequency was gravity, and could be sampled/interacted with at a slower rate. Then you had "Tilt" as medium frequency, and "Jerk/Bump" as the high frequency components. The calculations and decisions for these were done at different rates according to the frequencies. You weren't recalculating gravity at the same rate ask jerk/bump.
In the case of hair, you might even be able to have a low-pass filter that gets weaker the further you get from the root of the hair. That way you're mostly doing fast calculations on the tips, and the areas further up the hair are less jittery.
I was thinking about that. Interesting read thanks.
I also agree that your superhero alter ego should be called Dr. Papers
Tremendous, as per usual 💥😁
This channel indeed makes me feel alive!
I love how the doctors voice sound like an AI generated voice. Every time he talks in his videos lol. I love you man.
Awesome! Nice to see how far software can improve once the hardware is up to snuff.
you have blown my mind once more
Gunna be able to give headpats and haircuts in VR now!
That is amazing... now my question is... how can we implement this kind of tech to voxel worlds like "Teardown" to reduce lag and maybe even reduce the need for voxel
Can't wait for their new Rogaine gpu lineup!
What a time to be alive! It never gets old!
I missed you talking about non-AI topics! This is amazing!
Literally everything about that hair cutting demo is amazing to look at
Finally! I really miss these 3D CGI videos. I get the appeal of all the AI novelty, but I love the ones that are explicitly germain to video games
What a time to be a live!
Man, I never get tired of these. Now if we could get this as a grooming tool in Unreal and combine it with the upcoming face animation tool for their MetaHumans then life would be perfect!
this could be used for many purposes - to train barbers and hair stylists in VR. Or, I could even see a use case for scanning a person's own hair type and simulating it in virtual reality to see what different hair cuts or styles would look like on you and the projections would be 100% accurate.
You could also use this technology to develop virtual haircutting and styling robots and A.I.s - train them on virtualized machinery that can be produced in real life - that give you a consistent haircut exactly as you want every single time, so every time you get your hair cut it really is what you are picturing in your mind and not what your barber comes up with, even if you just want to take a half an inch off the top, it's not gonna get chopped up anymore.
My question now is, does it work for different hair types for example ethic groups that have - very kinky and curly hair, twists, dreds, animal hair, etc?
Looks good, but does this have any practical uses/applications other than video games and animated films?
This is super cool, but I really want to see it in action. Is there more samples of the new method in more “videogame-esque” environments? The simulation of them manually moving the hair and cutting it is already breathtaking, but I want to see more
I think it would be a good idea to contact some other tech-bloggers to implement this paper in some custom environments 🤔 We see it in action, they get the audience.
My question now is how well does it scale. One charter is cool but if you are standing on a busy street does the render time fall apart?
This truly looks great! Dealing with hair is always a pain.
NVIDIA's hair simulation is a captivating blend of precision and passion. Your channel's sober exuberance shines through, elevating the discussion above the hype and delivering insightful content that leaves me craving more. Your intelligent approach truly sets you apart. Keep enlightening us with your thought-provoking videos!
wow, we need this
The old minutes paper is back what a time to be alive
Holy ship! They actually did it! WOW
I like how the hair gets messy after they move it around. You can see how the hair was straight at 4:24 but then is messy at 5:07
It's always that way. Teams huge teams get incremental improvements and the leaps are made by one nerd in his basement, then the huge teams are credited anyway. Really makes you think about the actual potential of everything knowing that just multiple one guys make most of the progress in anything. What if even just a fraction of people worked with the same efficiency? That's why sometimes we get a Year Worth of progress in one week, it's just a motivation hits a field for one reason or another, and that's enough.
Inaccurate.
He is using corporate equipment to do research during working hours
Can you explain what is the breakthrough that allowed this to be so fast?
the author's galaxy brain /s
From what I understood, the primary contribution here is that the authors created an algorithm that can be efficiently parallelized across the GPU cores hence the performance gains. Since GPUs can have thousands of cores you can potentially gain huge benefits from parallelization since each core can execute computations independently. However, programming for GPUs can be quite tricky due to memory management and having to balance the load between the cores evenly.
@@Mega4est Ahh that makes a lot of sense! I know the power of GPU programming just from messing around with shaders in unity. Thank you for the explanation!
@@LightVelox Honestly if you are researching this sort of stuff you probably have a galaxy brain
looks very nice indeed … cannot wait to see these hairsims in games … probably only on $15000 24+ GB VRAM multiGPU setups fir 20 fps but hey we can dream .
Can't wait for the new "Haircut Simulator: Barber Adventures"
Same. And the character select would be amazing: Edward scissorhands or Sweeney Todd, which one do you choose?
Where's the wind simulation? 🤔 But pretty neat, hope it's as good as it looks!
Super incredible result!
This is awesome kind of the like the first time ive seen water look realistic on the n64 for the first time on a video game.
Any word on when tools like this would be available for users? Im working on an animated film that could greatly benefit from this, Blender's hair dynamics haven't been updated in years
so much room for innovation in computer graphics and physics rendering
This technique is super great for realistic story type video games like The Quarry, and Man of Medan
Hi, does the Nvidia program also do like wet hair to dry hair or dry hair to wet hair transition?
Knowing the "Hair problem" since the Unreal2 tek demos, this is really impressive & i truly hope someone compensates that man for this discovery!
The publication date at the link to the paper is a little too forward looking.
Congratulations fellow scholar. 😀
Amazing.. now they can solve game character walk inside brickwall or cloth inside their body problem
Can't wait for the new Hair-Tracing Technology
omg! i dropped my papers on this one. i'm baffled
Can't wait to play barbershop simulator in the future
I love this guys accent. He is like an educated Borat very soothing
I hope we may see this implemented in 3d tools not too far in the future
Question, I'm trying to create in Unity a global illumination in C# but I haven't experience with C# and Unity, where can I learn how to?
This will lead to an AI haircutting/styling robot like in sci-fi shows and movies
So this could be used to simulate hair styles/cuts of people before they get their actual hair cut to an extreme level of accuracy? This would make a lot of money if it was developed into a usable app.
Thanks Stimpy! ;)
I hope Nvidea gives their hard working scientists big bonuses and pay hike. Researchers deserve it more than executives who make overwhelmingly more
Million dollar app idea: haircut and beard practice .
Breath of fresh air to see non-AI Content in your videos...
but this is also ai
What about groom physics in UnrealEngine powered by Chaos Physics?
Next generation hair simulation!!! What a time to be alive!!! although I also wish I wasn't
Thank you doctor papers
100,000 hairs! Truly, we are living in the future.
I wonder if the concept of this breakthrough could be applied to somethninng important, like biomolecular simulations? They are full of point-point interactions ...
I want this to come up with my own hair style. I could never land on a good design for myself.
Barbershop simulator 2024 is gonna be insane!
I got my hair cut an hour ago…what a time to be alive!
We might finally get the Super-Saiyans we deserve.
Game grade realtime is a little bit different that what people call realtime. it have to take fraction ms max on lowest target device, or have a scale back strategy so it wouldn't noticeable if it's not present. 17 ms isn't even close to game grade realtime as it has to coexist with numerous of other systems that have their own budget.
Would still be perfectly fine in many scenarios imo. I think you could have the "simulation" run in a different framerate than the rest of the scene. Some games did that already in the past with things like moving capes and such.
Hope to see them on my metahumans soon
can this be hardware accelerated like with raytracing , as in NVidia using tenser cores or a dedicated engine inside the gpu die
Looking forward to all the video games becoming a lot more *hairy*.
AND what make this really impressive is not single AI is mention developing it.
I've said it before and I'll say it again, you really should hire someone for voice-overs.
At least as important as this is the Micro-Mesh Construction paper research.
one man can change everything
Amazing ! And now the good question : when will this be available in Unreal Engine ?!?! What a time to be alive !
The future of 3D simulations are going to be nuts. I can't help but wonder how much this research will pave the road to holographic technology kinda like Star Trek holodecks, 1000 papers down the line.
I think in the future they will make GPU as SOC where each core have different purpose so when the hair failed to compute the FPS still not dropped too much its just the Hair will have less quality
congrats now game devs can make a proper barber simulator
So oder so, ist das eine haarige Angelegenheit.
2 minute papers becomes, 6 mins and 21 secs papers