Why Are Cloth Simulations So Hard?

[@ayushkarnawat6817]

π minute papers

[@Fruitysfaction]

@Ethan Ansell Maths check out 3.141 * 60 seconds = 188.46 seconds = 3:08.46

[@Leibniz_28]

👍👍👍👍👍

[@johnsherfey3675]

@Ethan Ansell uhh π minutes isn't π seconds but, okay?

[@AmeshaSpentaArmaiti]

@@johnsherfey3675 it's 3.14 minutes, the math is correct. he used 60 seconds which is one minute. but, you know, okay?

[@johnsherfey3675]

@@AmeshaSpentaArmaiti confusing ending but what ever.

[@moominfella7614]

i really don't know why i love cloth physics so much, but i really REALLY like them, it's so satisfying when in a game you see your clothing move around that im genuenly upset when a game doesn't have them, am i weird?

[@nyramz2034]

Same lol, i always enthusias to see the cloth effect in fifa or pes

[@moominfella7614]

NYRAMZ mafia 2 remastered disappointed me big time cuz the original had really great physics.

[@tomshepard9050]

no u not buddy, but we're in the minority. Every Ue4 game that I play. If I can model swap the character, I'll rig up cloth sim, whether it'd be Sifu, Remnant, etc...

[@prismalglue]

one of the big reasons for that is cloth simulation connects the digital with the physical. when you see cloth move and react, your brain gives the objects weight, momentum,texture roughness etc which are very important to an animation looking real. its a similar thing with fluid and gas sims. but cloth is more in your face because its always on human bodies and we're so used to see and observe it.

[@FreeFireFull]

I'm impressed with how realistic these simulations look.

[@username4441]

same, i know nothing about it, always thought lighting/diffusion would be the hardest to simulate, turns out its all the collisions, makes sense I guess.

[@solaris5303]

​@@username4441 Lighting/diffusion is rendering, not physics simulation - it _may_ have been even more intensive than collisions, but it would be off-topic for this video.

[@Linck192]

looking forward for the day this is used in real time in videogames

[@OsedayCan]

It has been used for a very long time.

[@Linck192]

@@OsedayCan to this scale? I usually see capes and other more loose parts of the outfit being dynamically simulated but not the whole clothing.

[@pretzelboi64]

@@Linck192 Simulating entire outfits in games is kinda useless anyway. Is it going to look that much better?

[@Linck192]

well I think some people are more sensitive or care more about this than others. For me personally I would be very excited to see this same kind of effect that I see on this video being applied in a video game, in a controllable character etc

[@mayorc]

I would appreciate to have this in blender to start with.

[@ilovecodemonkeys]

It always feels like the clothes in simulations have super gravity Like they never stand awkwardly because of fabric issues I always wanna wear something that flowy lmao

[@nickpatella1525]

I hate when my pants put their cloth where the money isn't. It makes me wonder if clothing companies are conspiring against me: "let's make his date as awkward as possible!"

[@비부영양제]

Also cause they don't simulate the thickness of clothes but only imitate them They tend to be bent and deformed easily without the tensions and fractions needed inside the textile

[@MobyMotion]

I love that was still finding interesting ways to fully make use of GPUs, even though they've been around for so long. I wonder how long it'll take us to take full advantage of quantum computing if and when it ever becomes widely available

[@Kram1032]

More than Quantum Computers, for now I'm wondering more about TPUs (tensor processing units meant for AI applications) and the RTX cores Nvidia introduces for Ray-Tracing. - Both of those might have supplementary applications in tasks like the above as well. Potentially ones unrelated to their original purposes.

[@NextFuckingLevel]

@@Kram1032 yep i thought methods similar as ray tracing is still decades away, and boom NVIDIA has done it... Just wow

[@kelvinnnnnnnn]

Probably never, because you gotta cool that shit. But hey, how many times has the impossible been done by now?

[@Czeckie]

parallelism is hard

[@yellowblanka6058]

Quantum computers are still in a very primitive state that amounts to a handful of Qubits in a large enclosure at a very low temperature. You're not going to see Quantum CPUs for decades, and even then, because of the very different way they work there will likely be scenarios where traditional CPUs are more suitable.

[@alexkennedy4990]

I'm glad we have people working on these hard problems.

[@8923903910]

🤣🤣🤣

[@DanKind]

Happy to see a big english speaking hungarian youtuber. Also about topics I like :)

[@Darth_Pro_x]

Please make a video on hr algorithm used for the black hole image!

[@username4441]

It was a colorful spreadsheet, dont call it an image.

[@Darth_Pro_x]

@@username4441 hmmm... all images are colorful spreadsheats... what's your point?

[@username4441]

@@Darth_Pro_x Well, not all, but I assume you are referring to digital ones. And yes.. I suppose you are correct, but they were taken from a single sensor capturing light from a single vantage point, in the same scene at once. The black hole "picture" is about the most convoluted mess of confirmation bias ever created by man. But I do not think for a second that you have a good grasp on the procedure/whitepapers for the tech/method of garnering the data.

[@parthi2929]

1:20 That is bharatha nattiyam, a form of traditional dance from India

[@punalmanalan2346]

yeah man i too was thinking that the dance looked familiar!

[@Dirtfire]

The show "Love, Death+Robots" has some really good clothing animation in it.

[@michaeladams4999]

Meta what is your favorite episode?

[@matthiaswust1647]

@@michaeladams4999 Zima Blue

[@byama21]

When the yogurt took over...idk why I like it so much lol

[@kekchanbiggestfan]

Thanks for sharing, Borat. Very nice!

[@TheZindarod]

I long for a time when open source hardware will be as accessible as open source software is now. Imagine the innovations happening when researchers won't be limited to using the hardware manufactured by a couple of companies holding monopoly over the industry.

[@Telhias]

The problem with hardware is not whether it is open source or not. Of course companies like Intel or Nvidia are demanding ridiculous prices compared to the manufacturing cost of a given piece of hardware however the prices would continue to rise regardless. Even if open sourced hardware will only continue to rise in prices. That is because newer hardware requires smaller and smaller manufacturing errors. They require increasingly purer wafers and produce increasingly higher proportion of subpar products. The reason why Intel keeps on introducing more high end and expensive lines of products is because they produce less and less actual perfect wafers. Intel does not produce i5s and i7s separately. They are the exact same product just that i5s have some of the cores not working for one reason or another. Rather than throw it away as a failed product they simply rebrand it as a cheaper line of products. You have to realize that a dozen atoms off may result in a failed product. By making their architecture smaller they make the margin of error smaller as well and as such instead of 10 atoms breaking the product 5 will be enough. This in turn creates a higher proportion of wafers not perfect enough and throttles the supply of top of the line components. Unless we fundamentally change how we process data we will soon run into a wall of diminishing returns.

[@NortheastGamer]

@@Telhias Are those numbers accurate? Is it really two digit numbers of atoms that make or break a product?

[@Telhias]

@@NortheastGamer An atom size is between 0.1-0.5 nanometers, with silicon atom being 0.21nm. According to wikipedia 5 nanometer technology is already produced as of 2019. That would be around 24 atoms. It is not only tens of atoms but also steadily approaching single digits. Next technology scale is planned to be 3nm which would mean around 14 atoms.

[@NortheastGamer]

@@Telhias Neat! Thank you!

[@tylerdurden3722]

Tesla is basically a company with your idea of "open source hardware". Yet, that open source Tesla Model 3 will still cost you $40K.

[@VenomRaven]

Hey man, love your channel and how you make really complicated findings into bitesized pieces. I really appreciate the new quality settings. But are the video clips (and font) used in high res too? Because if not its somewhat counter productive.

[@boonleng]

Thanks for another great video. Great choice of paper. I’m always amazed by these types of computer vision work.

[@suri4Musiq]

I am happy to always listen to these kind of evergreen videos...Don't go AI savage...

[@lohphat]

Still in 2019 fabric and hair still have challenges. It too often looks like the material is too dense or it's underwater -- there's too much residual momentum is not appropriate for the input -- the dampening of momentum needs work.

[@Lurker1979]

Marvelous Designer needs this.

[@grendelum]

The *Andy* benchmark was gorgeous...

[@Kakerate2]

2:31 Whats the speedup rate on the 2nd and 3rd videos? Certainly lower, but how much does lowering the time help on frames?

[@tatzecom]

it doesnt help in term of render times, but it does help in terms of... well, how fluid the motion is afterwards If you have a scene that renders at 5 fps and you speed it up 12 times so you get to 60fps, you could also lower the quality of the scene so much that it renders at 60fps. It would just look worse

[@mibli2935]

What is the software package that allows these human figures to be developed and animated? Thanks!

[@thefacethatstares]

can't wait to see this being used in future equivalents of MMD

[@deivisony]

Watch the Ted Talk about something like "hearing audio from videos" in wich by looking at the pixels he can recreate áudio and by knowing the movement pattern of áudios he knows the movement patterns of Wind collisions and probably others. It is still in Alpha (at least the colission part) I only hope it will evolve in time to help with this video's cloth simulation

[@Kram1032]

Cloth simulation is becoming insane

[@dannywhitaker9008]

Dude imagine all this in a game! It would be real life at that point!

[@TheNewton]

Probably because simulation is having to constantly compute created state instead of evaluating state. Does a timestep of 1/200s really need to run constantly or would it be possible to varistep using 1/10s and 1/200s is just used for evaluation to smooth low timesteps.

[@Guztav1337]

Dunno. But the problem still stands, it is too easy to miss collisions, and end up on a state that is impossible to get out of. But this new method is great, since you could use x4 times larger time steps.

[@bcreativemedia2428]

Please upload the detailed tshirt cloth simulation..

[@tmlen845]

do you have to represent it as a triangle mesh?

[@tailsfan465]

I'm really hoping that we can get a really really immersive VR simulation in the future using stuff like this, and I guess it might happen in like... 20 years where we'll be able to do all this stuff in VR? i'm not sure, but i'm talking high framerates like 144fps or so. Heck it could also happen in 10 years, it's crazy to think about.

[@Enceos]

Cloud computing is the best invention of this decade.

[@diefe6515]

we really should investing more on this tech than gimmicky feature like Raytracing, HDR, etc

[@yy9w382]

if I use 4*gtx1080, can it run 32fps? Does it means if GPU have 5 times power in the future, cloth simulation can be achieved?

[@lucaspedrajas5622]

doesn't houdini vellum do that?? I think it's gpu accelerated

[@marcozolo3536]

Finally realistic upskirts in games

[@merxellus1456]

Now we know how Good Japanese Animation Artist are

[@cozzykhan7305]

Wish video games could use this

[@FullFledged2010]

So is this a modifier plugin for 3dsmax?

[@deltam5194]

Maybe you could have a price of cloth with a lower level of detail for collision and a higher quality for visuals

[@FooFooANIMATIONS]

Now we can simulate nooses

[@OgatRamastef]

amazing!

[@JoshFlorii]

Is cloth more resource intensive than fluid or smoke?

[@Hasnain1F]

It isn't about cloth or fluid or smoke. If you want something to look like real life, you have to include a significantly high amount of shapes and a load of calculations in order to figure out where they are and what are they doing. Thus it really depends. You have have fluid that are resource intensive, or smoke that is resource intensive and so on.

[@DutchmanDavid]

Well, well Mr Black, how the table cloth turns!

[@movax20h]

Collision detection with big timesteps can be easily solved using phase or minkowsky space and using concepts similar to raytracing in phase space. I did elastic ball collision simulator that was exactly , independent of the timesteps value. And doing it for triangles or few other shapes should be equally easy. Then you can plug it into adaptive implicit is over and simulation will be stable and accurate.

[@abowden556]

Maybe you should tell the authors that.

[@pretzelboi64]

How is Minkowski space going to help with that exactly?

[@FunkyPrince]

Well, I guess we will be wearing clothes with the help of ANNs a day.

[@shivani26v]

the test subject is doing bharatnatyam :) Nice

[@frankx8739]

Wind resistance?

[@konstantingeist3587]

no time to be alive this time, or did I miss it

[@freenomon2466]

thank you for your awesome post.. this is really great. is this possible to run on UNITY? with a 2080ti this would run pretty fast!

[@qiratvohra4436]

I am doing my final year project and the part of our project is based on cloth simulation Any one can guide me I am using blender for models and python for application But I don't know how these two work together to get the whole simulation done

[@aswath1144]

Has anyone tried out this simulation API?

[@iochair136]

I have seen many such cloth simulation papers, all with beautiful videos and no open source. What happened to these papers, are these implementations commercially available? Why is the cloth simulation in today's 3D software still so bad.

[@Jone952]

Maybe parallelization will keep Moor's law going!

[@Soraphis91]

> Moore's law is the observation that the number of transistors in a dense integrated circuit doubles about every two years (wikipedia) Moores law does not care for parallelization. its purely about hardware

[@Jone952]

@@Soraphis91 parallelization let's a program use more cores and hence more transistors

[@tylerdurden3722]

@@Jone952 more cores does not equal higher transistor density. Moore's Law actually refers to transistor density doubling...and the cost halfing...every two years. If you double the number of cores, you double the price of your product....and you wouldn't be doubling transistor density either.

[@npmmalayalamvfxtutorials1717]

Not just cloth every simulation is hard

[@chrisjansen1943]

Cloth simulations are probably the easiest thing to do in Unreal.

[@ruchitmicro8409]

I want GTA 6 to have this

[@dijon_]

it wont

[@tomshepard9050]

I'm still disappointed in today's option in cloth. Unreal has cloth sim, but it is limited to character collision, and sub-par barely acceptable collision. It always clips through the body doesn't matter how well you rig the capsules. No collision with ground and random objects, no environment wind, no volumetric air pockets to make it looks like in the movies. I'm still holding my breath.

[@todayschef1734]

I love you

[@pavithragomis9696]

what is a soft?

[@husnainanwaar1992]

Cloth Simulations is a 4D repenzantation

[@zekion.]

more polygons = less fps

[@Macho98]

Bro indian dance very nice to see

[@Vagolyk]

Researchers are desperate to flip some skirts.

[@nynra6584]

Don't worry guys our quantum cpus are coming. 300 qbits of which will be enough to simulate the atoms of the entire universe, so this is piece of 🍰

[@Ponlets]

why cant we use this in blender

[@theenglishman1312]

Not what a great time to be alive?////

[@theSpicyHam]

with those 'humans teach'es would bei 'slimenes also ' some 'deprivednes, 'off with those 'humans of

[@Marzaries]

I can tell you the solution, "AI"

[@virtualmartini]

Cloth is soft, which is why it's hard.

[@willinton06]

I bet he promises his wife that he’ll be done in 2 minutes too

[@hiddenmaster6534]

ugh that voice

[@qqqqqqqqqqqqqqqqqqqqqqqqqqw]

1:21 holy shit this is creepy

[@Jone952]

Maybe parallelization will keep Moor's law going!

[@BobbyRobby1000]

Concurrency has been the aim since the mid 00s. That why we have multiple cores