This is Geometry Processing Made Easy!

No video

This is Geometry Processing Made Easy!

Рет қаралды 125,615

4 жыл бұрын

❤️ Check out Linode here and get $20 free credit on your account: www.linode.com...
📝 The paper "Monte Carlo Geometry Processing: A Grid-Free Approach to PDE-Based Methods on Volumetric Domains" is available here:
www.cs.cmu.edu...
Implementations:
- / 1258218688726962183
- / 1258237114958802944
- www.shadertoy....
Our mega video on Multiple Importance Sampling: • The Story of Light! ☀️
Koiava’s MIS implementation: www.shadertoy....
My course at the Vienna University of Technology on light transport is available here. It is completely free for everyone:
users.cg.tuwie...
🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Aleksandr Mashrabov, Alex Haro, Alex Paden, Andrew Melnychuk, Angelos Evripiotis, Benji Rabhan, Bruno Mikuš, Bryan Learn, Christian Ahlin, Daniel Hasegan, Eric Haddad, Eric Martel, Gordon Child, Javier Bustamante, Lorin Atzberger, Lukas Biewald, Michael Albrecht, Nikhil Velpanur, Owen Campbell-Moore, Owen Skarpness, Robin Graham, Steef, Sunil Kim, Taras Bobrovytsky, Thomas Krcmar, Torsten Reil, Tybie Fitzhugh.
More info if you would like to appear here: / twominutepapers
Károly Zsolnai-Fehér's links:
Instagram: / twominutepapers
Twitter: / twominutepapers
Web: cg.tuwien.ac.a...

Пікірлер: 231

@RodrigoTheHappyDog 4 жыл бұрын

I actually shouted "WHAT" at my screen when you showed the length of the source code. Absolutely mental how compact that is.

@azafreak 4 жыл бұрын

same vibe

@sadhlife 4 жыл бұрын

goes to show that algorithm design is and will always remain King

@thebeststooge 4 жыл бұрын

I see how little there is to the code and how simple it will be to implement I just know Nvidia will throw that into their hardware somehow and real time/on the fly will happen to go along with their real time raytracing. I can only imagine the uses for anything graphics related (including deepfakes).

@martiddy 4 жыл бұрын

I was expecting at least a couple of thousands lines of code. This is super compact indeed.

@thebeststooge 4 жыл бұрын

@@martiddy Extremely modular from the snippet I saw.

@TwoMinutePapers 4 жыл бұрын

We are about to go on a vacation (we have found a way to do it responsibly). The next video is expected to appear approximately July 17th. Apologies and please hold on to your papers in the meantime! Of course, we'll be back on the regular schedule then.

@shrammstorm 4 жыл бұрын

Have a nice rest, Károly!

@TwoMinutePapers 4 жыл бұрын

@@shrammstorm Thank you!

@sawyer-m 4 жыл бұрын

Hope you have a great time, you deserve a break.

@danielkrajnik3817 4 жыл бұрын

well-deserved. Stay safe!

@crimsonstrykr 4 жыл бұрын

Enjoy!

@leonardosaraceni4402 4 жыл бұрын

Just for clarification purposes, I implemented this in c++, the idea is to use monte carlo approach to obtain samples to use for the Random Walk on Sphere algorithm, that is very weel described in the paper. In this way we can find boundary functions able to assign to each sample (pixel) a color, depending on its position. Hope this is clearer now. 👋

@TheZenytram 4 жыл бұрын

It did not 😫

@leonardosaraceni4402 4 жыл бұрын

Consider there is a boundary function, that separates different regions in which there are different colors for the pixels. To decide whether a pixel has kne color or another, we use the WoS algorithm that just makes a progressive "walk" through random samples on spheres built around the current sample. When the distance between the sample and the boundary function is less than an infinitesimal value (epsilon) it stops and assign to the pixel the found color. Sorry, but in few words this is.

@ivandimitrov7994 4 жыл бұрын

I think you need to have a basic understanding of how these things work at all to get anything about this, but let me ask this more practical question. Would this translate into getting more geometry in videogames without a performance hit?

@leonardosaraceni4402 4 жыл бұрын

@@ivandimitrov7994 honestly i don't know. Take my words not too seriously, i am just a 4th year informatic engineering student. The field of Monte carlo methods for rendering is not well explored since other techniques (PDEs methods) are proved to be really efficient, but maybe in the future... let's say that this algorithm works best in some conditions, not for every.

@TheZenytram 4 жыл бұрын

@@leonardosaraceni4402 well, i'm a negativeth year informatic engineering so you are the embodiment of wisdom to me.

@generalfishcake 4 жыл бұрын

I love how you gave a shout out to Sawhney and Crane - we need more superstar scientists in this world, instead of superstar "influencers".

@abebuenodemesquita8111 2 жыл бұрын

how about both? I'm looking at you Two Minute Papers

@Golinth 4 жыл бұрын

I’d totally be down for like a 30 minute explanation of this paper, cause I want to know more

@jonathanxdoe 4 жыл бұрын

Every time I see that purple environment in Blender for the missing map, I want to cry

@enriktigasna 4 жыл бұрын

I cry

@KuraIthys 4 жыл бұрын

Is that where that comes from? I'm used to missing textures showing up either as the underlying geometry colours, or often as a placeholder texture. (one that, unsurprisingly, often has 'missing texture' written on it.)

@AviNewatia 4 жыл бұрын

This channel never ceases to amaze me! Love the work and I never get bored of hearing the "Dr." now!

@empireempire3545 4 жыл бұрын

I worked with monte-carlo during my MSc and PhD too, but in a completely different environment - evolution and molecular modelling. It is amazing how versatile and powerful this technique is, how it helps in all different areas of science.

@GrahamRomero 4 жыл бұрын

And I used a super basic version in Computational Physics for alphanumeric character recognition, and the impact of messing with it's "vision" or "brain". When learning a bit about its origins / potential uses, it's pretty interesting!

@WangleLine 4 жыл бұрын

This time I really have no clue what's going on

@NotASpyReally 4 жыл бұрын

Just... Just HOLD ONTO YOUR PAPERS!!

@GeneralKenobi69420 4 жыл бұрын

"How are you doing, my fellow scholars?"

@user-on6uf6om7s 4 жыл бұрын

Yeah, I feel like we need those 9 other videos because this looks amazing but I understand very little about what it's doing.

@VirtualConnor 4 жыл бұрын

Lighting and raytracing through unconventional methods. It takes a light source and casts a ton of random rays in order to make realistic lighting.

@sakurahertz 4 жыл бұрын

Didn't expect seeing you here!

@malolan98 4 жыл бұрын

@Two Minute Papers thank you for all the great videos! I love the way you summarise and present the papers! As I am not from the computer graphics domain, I found it hard to follow through this video easily, so I'd be really happy to see a second video explaining the concepts in more detail! However, do have a fun vacation and if you're unable to make such a video, no worries! I'll try my best to understand stuff by reading the paper!

@0dWHOHWb0 4 жыл бұрын

I still have no idea what the paper is about... something about generating mesh geometry from... volumetric data or something?

@Squiffel 4 жыл бұрын

I think what he is saying is just that the algorithm can be applied to problems beyond just raytracing, that it can be more efficient when applied to anything that has some aspect of random distribution within the problem, but I also struggled to follow it

@cayo3351 4 жыл бұрын

Read the paper's introduction section. It is very explanatory.

@0dWHOHWb0 4 жыл бұрын

@@cayo3351 I don't doubt that

@Squiffel 4 жыл бұрын

@@0dWHOHWb0 I just read it, basically what the first part of the video discussed, pre-cutting up meshes into smaller problems being the conventional approach to solving raytracing pre-monte carlo, or more specifically a problem that has a lot of random distribution. In the case of the frog in a CT scan, it is still cut up into smaller pieces before solving/rendering, but this proposes that pre-cutting process is not necessary and instead you can use the same Monte Carlo algorithm. Let's you skip a step and still be efficient.

@thebeststooge 4 жыл бұрын

I can't wait for Nvidia to put this at the hardware level and allow us access to it as I see a lot of uses for it beyond raytracing such as on the fly deep fakes, denoise samples, etc...

@vlaamscherp 4 жыл бұрын

Ah yes, now I can finally render my tutorial-based, prestigious Blender Donut in 4k in no time

@thefakepie1126 4 жыл бұрын

nice

@kishanmaharaj9745 4 жыл бұрын

Does this mean we’re getting spooky new renderers?

@canbabalik3092 4 жыл бұрын

yes !!!

@trymbruset3868 4 жыл бұрын

Love it! Feel free to cover more of these topics in greater detail, very interesting indeed!

@TheZenytram 4 жыл бұрын

This goes so much high over my head that i didnt get what was the problem or the dificult nor the solution.

@vladchira521 4 жыл бұрын

But it was entertaining nonetheless, right?

@LV-426... 4 жыл бұрын

So people are supposed to read your name backwards?

@HarryHeck2020 4 жыл бұрын

There is the hard way to solve a variable equation and there is an easier way. You could draw out a grid on your X,Y plot and solve for every spot on the grid. That could be in the hundreds of thousands of calculations depending on the resolution you chose. Or! you could pick random spots, after a few calculations try to link them together with a curve, if you solve enough of the random equations without being to far away from your curve, you can probably predict where every point in that section of the plot is with a fraction of the calculations, then fill in the rest of the grid with approximate answers. This paper just says that this method works on more stuff than just ray tracing.

@NorseGraphic 4 жыл бұрын

If they can reduce the rendering-speed of photorealistic images down from 14 hours to 2 minutes, think how it can be done real-time. Computer-graphics in video-games will get a major overhaul.

@TheZenytram 4 жыл бұрын

@@HarryHeck2020 ohh, thanks that help understand a bit.

@ygy4301 4 жыл бұрын

This reminded me of UE5's revealing trailer. Some insiders said this is the way they created Nanite. Until now I didn't know how it works. Thanks!

@user-vn7ce5ig1z 4 жыл бұрын

0:18 - Károly, don't you realize that you're doing KZfaq wrong? You're not supposed to try to cram 10 videos' worth of material into a single 7.5-minute video, you're supposed to stretch a 3-minute video into at least 5 10-minute-1-second videos. 🤦

@immortaldiscoveries3038 4 жыл бұрын

ya but it saves all the most important people time and headache, and learn faster! We want answers and wisdom.

@YYYValentine 4 жыл бұрын

And then maybe I could understand what this paper good for... I didn't get what is the novelty in this

@thomasreese2816 4 жыл бұрын

I would be perfectly happy for this to be broken into 5 longer, detailed videos. Could call it "20 minute papers"

@DiamondShocked 4 жыл бұрын

Wow this is really cool! Professor Keenan Crane was my computer graphics professor at CMU

@iskamag 4 жыл бұрын

this man made my day and you too, doc

@jordanscarrott3749 4 жыл бұрын

Wow this is amazing. I was expecting a month of new learning to implement this but it looks great. I can't wait to read the paper

@YYYValentine 4 жыл бұрын

I don't understand one point. What is the novelty? What is the purpose? There was already montecarlo with importance sampling for lighting. .....Does it generate a point cloud geometry with montecarle algorithm, or what?

@Nikowst 4 жыл бұрын

Optimization, i guess...

@aniksamiurrahman6365 4 жыл бұрын

Unbelievable! I can't believe these are happening during my life time! What a time to be alive!

@konignickerchen7265 4 жыл бұрын

Wonder if it can support animation. I know they're intended for other areas, but motion and especially deformable meshes is usually never seen in most point cloud renderers.

@Darth_Pro_x 4 жыл бұрын

I'm really waiting for deformable meshes to become better and more widespread, so much potential!

@nilspin 4 жыл бұрын

@@Darth_Pro_x what is a better representation for geometry if we want rigid/soft body physics for complex models? SDFs? Triangle Meshes?

@Darth_Pro_x 4 жыл бұрын

@@nilspin i have zero clue haha, sorry. not my field of knowledge, just a geek who enjoys learning about the progress in AI

@tooneepoo 4 жыл бұрын

Actually missed the mandatory "What a time to be alive!" ;-)

@mikaelengstrom6639 4 жыл бұрын

I was really glad that it was missing. The one and only thing that I don't like about all these videos, as it is quite cringy =D

@juliuszeusz3828 4 жыл бұрын

You make everything seem so easy that i - a 12yo from Poland understands you You are so underrated

@Kram1032 4 жыл бұрын

Oh it's a Keenan Crane paper! I love pretty much ALL his works. And this one is no exception!

@Towik 4 жыл бұрын

Blender gets this implemented when? xD

@msmith2961 4 жыл бұрын

Nope. Sorry, Imma gonna need those other 9 videos to wrap my little brain around how awesome this is and why you're so excited.

@luislindgren5615 4 жыл бұрын

I recently started learning FEM for physics/engineering problems, and I feel like that field would also benefit from this paper. I'm a total newbie, though... Any thoughts on that?

@onlyeveryone2253 4 жыл бұрын

Same, it's fun to recognize the Monte Carlo method.

@kylerichardson514 4 жыл бұрын

I would be interested to. I'm currently studying FEM as a Ph.D. student. I found the paper interesting as it did talk a lot about FEM and the very real challenges that it faces. I'm always a little skeptical of new techniques that claim to compete with FEM. The paper really only considers linear problems which is boring. The big question for me is how well can it handle nonlinear problems? The paper claims that the method can be extended to nonlinear problems, but my experience has been that many can claim that, but few actually can deliver. It is always much harder than they think it is. That being said, I do think that there is some possibility that this work could be used augment the power of FEM. I'm not entirely sure yet on what form that might take as there are some new methods in FEM on the way down that could greatly alleviate many of the problems with FEM that the paper talked about.

@cosmotect 3 жыл бұрын

I come to this channel to reign in my sense of self importance

@thefastsnake3015 4 жыл бұрын

praise be to Rohan Sawhney

@thatchessguy7072 4 жыл бұрын

Do the light rays need to be treated as a collection of vectors? Could they be treated as a smaller collection of cones, would that result in a blurry or distorted image? Would using cones be computationally expensive?

@BalancedEarth 4 жыл бұрын

so wait, is it calculating the light bouncing within the ray that is let into the other part of the box? and as it does that it can predict and create a working formula for calculating a quicker rendering with geometry? Isn't it just guessing and being more precise?

@romanberdichevskii5141 4 жыл бұрын

I would be glad to watch 10 episodes designated for one paper

@abhijanwasti7991 4 жыл бұрын

This is so awesome! Thanks for the video!

@miguelpereira9859 4 жыл бұрын

That Rohan Sawhney guys has to be a genius damn

@sistemsylar 4 жыл бұрын

it's just 1 small paper for a PhD student, but a great leap for humanity.

@dwindeyer 3 жыл бұрын

This must have been what Jensen was talking about when he mentioned a new path tracing algorithm implemented in the Marbles at Night demo

@ali32bit42 4 жыл бұрын

i kid you not i was working on a digital sculpt and blender was taking hours to remesh it. and now this paper shows up. strange coincidence.

@diegosolis9681 4 жыл бұрын

At first i was like: "Holy hell this is actually extremely clever!" Then i was like: "Broooother, this is amazing! What genious!" Then: "Man, this stuff must be like a million lines of code" Then i saw the code and i screamed, i shit you not: "Awwwwww, f************ck off! (in my native language), How? Hoooooowwww?" Then my flatmate came to see what was going on and i explained him why this dude is simple amazing, just as the trying to explain meme.

@terner1234 4 жыл бұрын

why do you use blender 2.7X instead of newer 2.8/2.9? in 2.9 there is viewport denoising, and since 2.8, the UI is much better

@z-beeblebrox 4 жыл бұрын

As someone who struggles at modelling, I'm curious: is this something that's going to help me or is it more of for scientific purposes? Like, if I take a volumetric photo of an object, and I have this code handy...am I good to go?

@1DInciner 4 жыл бұрын

Gorgeous solution!

@skyr8449 4 жыл бұрын

Can you make a few more videos talking about this? This is the only time I have been a little confused watching your videos! :)

@elultimoescriba892 4 жыл бұрын

What about GPT-3, i think it is impresive.

@kylebowles9820 4 жыл бұрын

Epic shoutout to koiava on Shadertoy! I learned a lot from their shaders!

@davidm.johnston8994 4 жыл бұрын

Hello doctor! I would love a more in depth dive into this paper. It's very interesting. Take care ;-)

@bernardobritto8352 4 жыл бұрын

Can you do that type of video more technical about this paper? It would be really awesome!

@bernardobritto8352 4 жыл бұрын

Maybe you could also explain the code!

@tune_m 4 жыл бұрын

There's so much theory behind all the paper's applications. Cool visuals though!

@Xizilqou 3 жыл бұрын

This is the content that should be on the front page, not "(OMG) CALLING BALDI'S BASICS AT 3AM (NOT CLICKBAIT) (ALMOST DIED) (GONE WRONG)"

@martylawson1638 4 жыл бұрын

Can this be used for grid-free physical simulation models? Say, non-linear stress analysis or CFD? Both can be described by PDEs.

@nilspin 4 жыл бұрын

I still don't get it. I lost track from around 2:20 about how geometry processing comes into standard ray-tracing process (did you mean the process of dividing on-screen geometry for better radiosity calculations?). Regardless, I cou;dn't understand how monte-carlo came into the picture to solve this problem of splicing geometry.

@PrinceWesterburg 4 жыл бұрын

Monte Carlo was replaced widely in radiosity (bounced light) by Fprime solutions, perhaps this can be applied here?

@Raj.madge078 4 жыл бұрын

Proud to be an Indian Rohan 🇮🇳

@vivekmittal1454 4 жыл бұрын

This video made my day.

@dfghj241 4 жыл бұрын

why can't we solve the rendering equation? whatahell is a rendering equation? so many questions!

@sychuan3729 4 жыл бұрын

It is integral equation which describes all light going from sphere around viewer eye or camera. As many integral and differential equations you usually can't find for it analytic solution (function which gives answer) but you could solve it numerically (aproximate). This is what light transport is.

@orraviv1536 4 жыл бұрын

do you think we can create an AI that both learns how to walk and change the muscles or bone shapes and positions to be more efficient?

@TC-rv6sz 4 жыл бұрын

I know you were joking when you mentioned it, but please consider making the 10-video series on this paper/topic! In-depth coverage would be amazing.

@q23main 4 жыл бұрын

_Unreal_ . Yes this should be _Epic_ Couple versions down the line. PS6 graphics will be awesome 😄

@grrsa 4 жыл бұрын

I've loved your content for years, Karoly. I wish I could do similar work. Could you offer any advice, courses, or readings that would help someone to a career in computer graphics research?

@alechart462 4 жыл бұрын

Wow! I have NO other words!

@sunimod1895 4 жыл бұрын

How is this different from nvidias ray tracing? I'm not sure I understand how either works exactly.

@Goten1010 4 жыл бұрын

Is this about creating geo from some sort of volumetric scan?I understand the concept of monte carlo sims, but i really dont understand whats going on here? Sure, you can use these randomness methods for other things other than rendering, but why? I am a bit lost!

@PissMasterPlus 4 жыл бұрын

Can Monte Carlo Integration be used, for say, sound reverberation?

@TheGraphiteCovenant 4 жыл бұрын

how does this will apply to commercial availability or any of the mentioned papers?

@fionnbarrett1030 4 жыл бұрын

Y using blender 2.7?

@Forjugadname 4 жыл бұрын

Wow, those shadertoy links nearly destroyed my PC, 99 percent cpu usage 80 degree temps and climbing(I think I need a better cooler). This is really interesting I barely understand it but I'm wondering if it can be converted to use in Unity.

@seamusoblainn4603 4 жыл бұрын

I'm going out to buy a few icecreams to enjoy this with 🤓

@nxtler7513 4 жыл бұрын

Whats about rtx raytracing?

@HAWXLEADER 4 жыл бұрын

Wasn't rendering already done in a random seed monte-carlo manner?

@jurekmc 4 жыл бұрын

monte carlo is one fancy way of saying just try numbers until it works fam

@sean2val 4 жыл бұрын

pretty amazing stuff

@scottbigbrain3944 4 жыл бұрын

Honestly Two Minute Papers, I am just waiting for you to put out something on GPT-3 or Image GPT.

@PriNcEoFSpAcE009 4 жыл бұрын

I have absolutely no idea what the fuck is going on in this video.... but you get a thumbs up anyway.

@demp11 4 жыл бұрын

I actually had that thought of putting ai inside a render engine couple days ago

@theshadow6583 4 жыл бұрын

the real question is when are we going to see its applications in real life and not as a research paper??? by the way a really good job from them and a good video from u

@anilaxsus6376 4 жыл бұрын

erm, am i the only one who still dont know what exactly this paper has created or improved ? If you understand pls explain, for the mean time, i will rewatch the video.

@michaelleue7594 4 жыл бұрын

Basically, imagine CSI going "Enhance!" and it actually working.

@petermarshall1634 3 жыл бұрын

I'm guessing around 7 days

@glitchp 4 жыл бұрын

Get this into medical imaging devices ASAP

@ye-ln1hx 4 жыл бұрын

Nice

@johnromanas 4 жыл бұрын

Is there any relation with this and Ray Tracing?

@JamesPound 4 жыл бұрын

Trying to understand.. it’s raytracing without polygons? Hmm..

@ThatOneIrishFurry 4 жыл бұрын

still feel like this channel should be called 10 minute papers

@ThatOneIrishFurry 4 жыл бұрын

no "what i time to be alive" my world is collapsing

@BradenEliason 4 жыл бұрын

Check out Crane's videos on KZfaq. He has many incredible lectures available. If you're a fan of TMP, you'll enjoy those lectures.

@skilbhumen2875 4 жыл бұрын

can we use this in blender?

@katokianimation 4 жыл бұрын

Is it posible that you are Hungarian? Love your work.

@CosmiaNebula 4 жыл бұрын

You can make a 20 minutes video, instead of 10 2-minute videos! Just make a warning for us to use a steel clamp to hold on to our papers.

@estehbread 4 жыл бұрын

I wanna see an oobleck simulation with this lol

@korrdavl 4 жыл бұрын

The scientific applications are more than expansive here

@thatJoseph 4 жыл бұрын

I feel so bad that I'm too stupid to appreciate this

@aiyushgupta8595 4 жыл бұрын

Yay! 50th Like- love this channel BTW

@troooooper100 4 жыл бұрын

ELI5 I'm lost

@OpreanMircea 4 жыл бұрын

5:20 one line of code

@felixinit 4 жыл бұрын

Las paper about Monte Carlo Geometry Prossesing

@rudragaitonde6197 4 жыл бұрын

Yesss

@notlegal99 4 жыл бұрын

wow

@imsorryforanythingweirdisa2767 4 жыл бұрын

WHY DO I KEEP GETTING RECOMMENDED THESE VIDEOS AND WHY DO I KEEP WATCHING THEM

@leonpreiler278 4 жыл бұрын

I have a question do you have a company?? Can anyone invest in it

@hermannpaschulke1583 4 жыл бұрын

This is the first time I don't even understand the Problem lol.

@josiane8516 4 жыл бұрын

One day