This is Geometry Processing Made Easy!
HTML-код
- Опубликовано: 2 окт 2024
- ❤️ 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...
![NOSFERATU - Official Trailer [HD] - Only In Theaters December 25](http://i.ytimg.com/vi/nulvWqYUM8k/mqdefault.jpg)
I actually shouted "WHAT" at my screen when you showed the length of the source code. Absolutely mental how compact that is.
same vibe
goes to show that algorithm design is and will always remain King
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).
I was expecting at least a couple of thousands lines of code. This is super compact indeed.
@@martiddy Extremely modular from the snippet I saw.
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.
Have a nice rest, Károly!
@@shrammstorm Thank you!
Hope you have a great time, you deserve a break.
well-deserved. Stay safe!
Enjoy!
This time I really have no clue what's going on
Just... Just HOLD ONTO YOUR PAPERS!!
"How are you doing, my fellow scholars?"
Yeah, I feel like we need those 9 other videos because this looks amazing but I understand very little about what it's doing.
Lighting and raytracing through unconventional methods. It takes a light source and casts a ton of random rays in order to make realistic lighting.
Didn't expect seeing you here!
I’d totally be down for like a 30 minute explanation of this paper, cause I want to know more
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. 👋
It did not 😫
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.
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?
@@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.
@@leonardosaraceni4402 well, i'm a negativeth year informatic engineering so you are the embodiment of wisdom to me.
Every time I see that purple environment in Blender for the missing map, I want to cry
I cry
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.)
Does this mean we’re getting spooky new renderers?
yes !!!
Ah yes, now I can finally render my tutorial-based, prestigious Blender Donut in 4k in no time
nice
0:18 - Károly, don't you realize that you're doing RUclips 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. 🤦
ya but it saves all the most important people time and headache, and learn faster! We want answers and wisdom.
And then maybe I could understand what this paper good for... I didn't get what is the novelty in this
I would be perfectly happy for this to be broken into 5 longer, detailed videos. Could call it "20 minute papers"
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.
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!
This goes so much high over my head that i didnt get what was the problem or the dificult nor the solution.
But it was entertaining nonetheless, right?
So people are supposed to read your name backwards?
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.
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.
@@HarryHeck2020 ohh, thanks that help understand a bit.
This channel never ceases to amaze me! Love the work and I never get bored of hearing the "Dr." now!
I still have no idea what the paper is about... something about generating mesh geometry from... volumetric data or something?
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
Read the paper's introduction section. It is very explanatory.
@@cayo3351 I don't doubt that
@@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.
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...
@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!
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
Actually missed the mandatory "What a time to be alive!" ;-)
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
Blender gets this implemented when? xD
Love it! Feel free to cover more of these topics in greater detail, very interesting indeed!
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!
it's just 1 small paper for a PhD student, but a great leap for humanity.
I come to this channel to reign in my sense of self importance
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?
Optimization, i guess...
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.
I'm really waiting for deformable meshes to become better and more widespread, so much potential!
@@Darth_Pro_x what is a better representation for geometry if we want rigid/soft body physics for complex models? SDFs? Triangle Meshes?
@@nilspin i have zero clue haha, sorry. not my field of knowledge, just a geek who enjoys learning about the progress in AI
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.
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.
That Rohan Sawhney guys has to be a genius damn
This is so awesome! Thanks for the video!
Epic shoutout to koiava on Shadertoy! I learned a lot from their shaders!
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)"
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?
Same, it's fun to recognize the Monte Carlo method.
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.
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.
Basically, imagine CSI going "Enhance!" and it actually working.
this man made my day
and you too, doc
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
praise be to Rohan Sawhney
What about GPT-3, i think it is impresive.
hmm, yes
this is good right?
yea?
ok, cool.
so how good is it?
wow
cool
Wow this is really cool! Professor Keenan Crane was my computer graphics professor at CMU
_Unreal_ . Yes this should be _Epic_
Couple versions down the line.
PS6 graphics will be awesome 😄
Get this into medical imaging devices ASAP
Proud to be an Indian
Rohan 🇮🇳
I appreciate the intent but I'm no fellow scholar I'm just one of many subscribed stoners
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.
Hmm something is not right about this paper. Radiosity is from a bygone era and hasn't been used in 3d for ages. The cycles renderer shown in blender there has been using Montecarlo or a method like that shooting rays with a denoiser to get fast results for years now. The days of rendering an image taking more than a minute or two or hours is also from a bygone era or from people who don't know what they are doing and I see a lot of them even to this day. Whatever this video was it is either not done correctly or just simply incorrect information/presentation. That explains all the confusion in the comments. Monte Carlo is an algorithm that has been used in rendering for years now that makes rendering easier so I just don't understand what this video is making a big deal out of there is nothing new here except perhaps the person has got it all written in a very simple script that is open source and perhaps the idea that using that method is so general purpose it looks like they are using it for things in real life like guessing the inside of how an ant looks under a microscope to get a good guess representation of how it would actually look and perhaps how the inside of frog looks....not really very useful or anything worth being labeled as the best paper of the year I'm afraid. Feel free Two Minute Papers to address this confusion.
still feel like this channel should be called 10 minute papers
no "what i time to be alive" my world is collapsing
I feel so bad that I'm too stupid to appreciate this
Sure the implementation might be short but the quality and efficiency is quite garbage compared to what you can achieve when you actually reconsider your concepts and integrate possible algorithms to shorten or improve it, so you will eventually end up
with conditionally different implementations based on the given environment and use case to lower the overall computation complexity.
That's also what happened to the ray tracing meta where directx devs started to use bvh trees for intersection checks, computed by morton code based splitting, and hardware ray intersection support - etc - not comprehensible for the average college pleb.
Not even talking about cache line alignment, SIMD and GPU compute pipelines. This is just a really simplified implementation and as a OG cpp news and project subscriber you surely can tell that the source clearly isn't the cherry of the cake.
Cpp shouldn't be written the same way as you write your shaders. This is a mixture where the coder seems to put GPU favoured code into cpp instead of writing a shader or making use of CUDA/OpenCL.. But I guess, sometimes it's better to stay focused on the mathematical part of things.
Anyhow someone will eventually come up with the 10k+ lines of magic for a more complete and qualitative implementation.
why can't we solve the rendering equation? whatahell is a rendering equation? so many questions!
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.
I'm guessing around 7 days
Trying to understand.. it’s raytracing without polygons? Hmm..
The scientific applications are more than expansive here
Honestly Two Minute Papers, I am just waiting for you to put out something on GPT-3 or Image GPT.
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.
You're gonna keep scrolling down more . . . right? You're just going to talk about the first part and then scan over the five other pages of it . . . right?
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?
You make everything seem so easy that i - a 12yo from Poland understands you
You are so underrated
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?
Unbelievable! I can't believe these are happening during my life time! What a time to be alive!
This thread from Dr. Keenan Crane goes into the problem, the approach and the motivation a bit more: twitter.com/keenanisalive/status/1258152669727899650?s=21
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?
ELI5 I'm lost
NOOOO MAKE 10 VIDEOS
I have absolutely no idea what the fuck is going on in this video.... but you get a thumbs up anyway.
Is that a geometry dash reference?
This must have been what Jensen was talking about when he mentioned a new path tracing algorithm implemented in the Marbles at Night demo
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?
Eh?
Monte Carlo was replaced widely in radiosity (bounced light) by Fprime solutions, perhaps this can be applied here?
Y using blender 2.7?
Can you make a few more videos talking about this? This is the only time I have been a little confused watching your videos! :)
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.
Can you do that type of video more technical about this paper? It would be really awesome!
Maybe you could also explain the code!
WHY DO I KEEP GETTING RECOMMENDED THESE VIDEOS AND WHY DO I KEEP WATCHING THEM
Las paper about Monte Carlo Geometry Prossesing
This is the first time I don't even understand the Problem lol.
I would be glad to watch 10 episodes designated for one paper
I'm going out to buy a few icecreams to enjoy this with 🤓
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.
monte carlo is one fancy way of saying just try numbers until it works fam
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.
60 lines of code
Can this be used for grid-free physical simulation models? Say, non-linear stress analysis or CFD? Both can be described by PDEs.
Nice
I laughed when it showed that it took 2 minutes for the render.
Is it posible that you are Hungarian? Love your work.
I have a question do you have a company?? Can anyone invest in it
so if a game's graphics were built this way from the ground up, it could run equally fast on older hardware just with "grainy" graphics
To be honest i don't understand this edition very well.
PS5 drops to its half of price
How is this different from nvidias ray tracing? I'm not sure I understand how either works exactly.
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?
More like seven and a half minute paper 😅
I wanna see an oobleck simulation with this lol
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!
Hello doctor! I would love a more in depth dive into this paper. It's very interesting.
Take care ;-)
my guess is that if you know how Monte Carlo simulations work and your programming and graphics knowledge is sharp, then it should not take more than one week to reimplement that based on the paper
Whats about rtx raytracing?
There's so much theory behind all the paper's applications. Cool visuals though!
Oh it's a Keenan Crane paper! I love pretty much ALL his works. And this one is no exception!
how does this will apply to commercial availability or any of the mentioned papers?
Is your voice generated by ai or you apply some filters on it or I lose my mind. Dear scholars is it only me or you hear some wobbly distorted voice as well ?