Unlimited Detail technology. Yay or nay?

A place to discuss the science of computers and programs, from algorithms to computability.

Formal proofs preferred.

Moderators: phlip, Moderators General, Prelates

User avatar
MHD
Posts: 630
Joined: Fri Mar 20, 2009 8:21 pm UTC
Location: Denmark

Unlimited Detail technology. Yay or nay?

Postby MHD » Sun Jun 05, 2011 4:12 pm UTC

A more game-programming interested friend of mine recently introduced me to this new up-and-coming graphics technology called "Unlimited Detail."
(This write up is gonna be a bit short, my hands are sore...)

Whoa. Hold up. Did they just say 'Unlimited?' Bold claim, but they seem to be onto something.

Official websites and some demonstration videos:


What they claim to have found is something akin to a holy grail.
They have found a way to have voxel-esque graphics without screwing everything over.
Their technology uses some loosely-allocated, ordered point cloud scene (they are really sparse on the details) over which they iterate some search algorithm to render the scene. They claim that this algorithm is so powerful it can run over complex scenes, and only picking out the necessary points, render it, in CPU on semi-old laptops.

Now to me, if this is true, we are on the precipice of a breakthrough. This means that we are one step closer to moving our graphics algo's over to CPU space (which by my estimates and the current trend in GPU load would be in 10-15 years).
However, their brand new thing has some holes.

Now, I know that you don't just jump out and say, "hey guys, this is the next big thing!" and then publish all your research without first patenting it in twenty seven different ways, and classifying it in four more. ATI and NVidea probably want to get their hands on this and make a new GPU pipeline for it asap. But, still it gives my that iffy feeling of "we haven quite worked out the kinks, wait ten years."

First, they seem to be hating on polygons. A lot.
That's fine, polygons looks like hell when you use them without proper forethought, but they really hate them with a passion. They (probably for advertising purposes) very clearly state that Unlimited Detail is much better. Good use of polygons and the right tricky application of parallax and normal mapping can, presto, give you brilliant result for very little computing power. In their demo vids they show some real flaws in map design, instead of actually showing what UD can do for the world.

Second, it seems like it is going to be hell to make animations and map design in.
A voxel-esque graphics format is terribly difficult to animate.

What are your thoughts on the future of Graphics?
EvanED wrote:be aware that when most people say "regular expression" they really mean "something that is almost, but not quite, entirely unlike a regular expression"

_Axle_
Posts: 253
Joined: Fri Sep 24, 2010 7:33 pm UTC

Re: Unlimited Detail technology. Yay or nay?

Postby _Axle_ » Sun Jun 05, 2011 7:47 pm UTC

My thoughts on Unlimited Detail technology....

I am still watching one of their comparison video and read their "What is it?" page. I have some major issues with how this is all presented. They are real critical at the comparison and not even understanding some of the stuff of "Polygon system". Then when their system has issues, the commenter mentioned "Be nice to us, we are new". They had both shadow popping and artifacts around. They also keep mentioning their buzz words of technology, which is rather annoying.

A major issue they don't touch is the memory footprint. Part of the issue with modern games is they are only allowed so much art data in the game, since everything has to fit on 1 disc ( Well, maybe more discs, but the majority of all models and textures are repeated over multiple discs ). Not only does the data have to fit on the disc, it has to fit on a rather small amount of RAM. Their comparison also is just a graphics tech demo, so no other process is being run : input, AI, physics, sounds, networking, etc.

They really need to get some of the major engine players to adapt this, if it will have any kind of life in the industry. Re-doing the graphics pipeline and algorithms isn't something that I am sure most companies are going to risk ( when they already kind of have to with the switch of DX9 - DX10/11 and the new pipeline stages ). If no company shows interest, then ATI and NVidia are not going to waste R&D time in designing a card that won't be used.

My thoughts on the future of graphics....
I personally think that the only big upgrade for game graphics is going to be a switch to Ray traced scenes. We will need need hardware to streamline and speed up the rendering time of traditional ray tracing, but I could see it happening. People have achieved in small scale real time ray tracing using our current GPUs using CUDA.
Yakk wrote:Computer Science is to Programming as Materials Physics is to Structural Engineering.

User avatar
undecim
Posts: 289
Joined: Tue Jan 19, 2010 7:09 pm UTC
Contact:

Re: Unlimited Detail technology. Yay or nay?

Postby undecim » Mon Jun 06, 2011 2:12 am UTC

I'm not convinced... They aren't releasing any information about the algo, really... Other than saying it's a "search algorithm"

And the word "Unlimited" very much angers me. Computers cannot be unlimited, and so there cannot be unlimited data. You can have structures that can have unlimited detail, perhaps, like fractals, or 3d-functions, but that's not unlimited data.
Blue, blue, blue

Moose Hole
Posts: 398
Joined: Fri Jul 09, 2010 1:34 pm UTC

Re: Unlimited Detail technology. Yay or nay?

Postby Moose Hole » Mon Jun 06, 2011 2:23 pm UTC

Sounds like perlin noise?

User avatar
WarDaft
Posts: 1583
Joined: Thu Jul 30, 2009 3:16 pm UTC

Re: Unlimited Detail technology. Yay or nay?

Postby WarDaft » Mon Jun 06, 2011 5:24 pm UTC

The point count they talk about makes any worthwhile animation not difficult, but completely infeasible. Sure, you can move it around. Try deforming it, like say a hand when it closes. It's either going to look terrible or be too computationally intensive to do.

And their search algorithm must be something fantastic - to render 1600 by 1200 at 60 hz requires identifying at least 1 billion points per second (scaling up rapidly with any form of anti-aliasing) from something that cannot possibly be stored in the CPU's cache, and would be almost impossible to store in memory without high powered compression algorithms - which would then take up more CPU time.

If they could actually do this practically, all they'd need to do is to show an interactive example to either ATI or nVidia (so that you could see that it was in fact not using a graphics card and still rendering as much detail as they claim in real time) and they would have more research funding than they could have ever dreamed of.


There's nothing at all wrong with the polygons + depth mapping concept. You're not storing data you don't need, not calculating things you don't have to, and we already have a vast amount of research already dedicated to it. To switch to something completely new requires that we redo all that for the new model to get to the same point.
All Shadow priest spells that deal Fire damage now appear green.
Big freaky cereal boxes of death.

kmatzen
Posts: 214
Joined: Thu Nov 15, 2007 2:55 pm UTC
Location: Ithaca, NY

Re: Unlimited Detail technology. Yay or nay?

Postby kmatzen » Mon Jun 06, 2011 5:52 pm UTC

WarDaft wrote:Try deforming it, like say a hand when it closes. It's either going to look terrible or be too computationally intensive to do.


Well, I don't know about that. Last month, I implemented point-based physics and rendering for my computational motion course. If you compute the internal forces on the GPU, then point-based deformation works pretty well and is easy to implement. The memory footprint just sucks. I get what you mean by the count, but I would think they would deform it in some reduced coordinate space. Even for detailed polygonal models, deformation is too expensive if done directly to material coordinates. These concepts are years old though.

My guess is that their search algorithm assumes the camera motion is coherent. Then the most expensive frame would be the first frame which they could just precompute offline. I'm also wondering if the examples in the YouTube video render each object once as points and then generates many instances of that single rendering or if they render every instance as points. They had some pyramid made of animals. If only one was rendered as points, that's not very impressive. You can render with about 120,000 surfels and get something like that in just a few milliseconds, for example. The difficulty with a surfel approach though is clipping intersections to make them appear sharp without CSG surfel operations and raytracing. The nature scenes were much more impressive, but YouTube compression makes it difficult to evaluate.

I might have missed it, but did they ever say they don't use the GPU? These sorts of techniques they talk about usually make heavy use of fragment shaders. The vertex shader ends up being something trivial and just there for the sake of being in the pipeline.

Oh, and as for unlimited, I think what they mean is that you give it N points and it will in some way do something reasonable with those N points. It might cull many or downsample them based on the viewport size, but if they are there, it will render them and not die. At least that's what I took away. For all I know, maybe they just implemented the entire Point-Based Graphics book and turned it into a marketable product.

Edit: Oh, is this not an interactive demo? I don't understand what is novel then. Pixar's RenderMan already supports point-based rendering so I'm not sure if we need another offline solution. I guess I just assumed the demo was interactive since they compared it to video games.

User avatar
WarDaft
Posts: 1583
Joined: Thu Jul 30, 2009 3:16 pm UTC

Re: Unlimited Detail technology. Yay or nay?

Postby WarDaft » Mon Jun 06, 2011 7:15 pm UTC

They say it's interactive and that it runs on just the CPU. No one else actually knows if it is or not.
All Shadow priest spells that deal Fire damage now appear green.
Big freaky cereal boxes of death.

kmatzen
Posts: 214
Joined: Thu Nov 15, 2007 2:55 pm UTC
Location: Ithaca, NY

Re: Unlimited Detail technology. Yay or nay?

Postby kmatzen » Mon Jun 06, 2011 7:24 pm UTC

WarDaft wrote:They say it's interactive and that it runs on just the CPU. No one else actually knows if it is or not.


That's what I thought too. The thing is, the company's exit strategy seems like "get acquired by Intel, AMD, or NVIDIA", so if it doesn't work, then they are going to have a hard time fooling the people they need to fool. I know that NVIDIA has key players in point-based graphics and physics from ETH Zurich, so they might be the hardest to fool. I know nothing of the experts AMD and Intel have.

User avatar
Amnesiasoft
Posts: 2573
Joined: Tue May 15, 2007 4:28 am UTC
Location: Colorado
Contact:

Re: Unlimited Detail technology. Yay or nay?

Postby Amnesiasoft » Mon Jun 06, 2011 11:30 pm UTC

I swear I saw this once... two years ago, and when I saw it then, I could have sworn I'd seen it two years ago too.

And here it is from over two years ago on Jan 30, 2009: http://www.somedude.net/gamemonkey/foru ... f=12&t=419

I'm having difficulty finding it from before that, but this seems to be completely bogus. It never changes every time it comes up, and there has never been a public demo to demonstrate that their claims are, well, real.

User avatar
WarDaft
Posts: 1583
Joined: Thu Jul 30, 2009 3:16 pm UTC

Re: Unlimited Detail technology. Yay or nay?

Postby WarDaft » Tue Jun 07, 2011 12:45 am UTC

I have seen the system in action and I have seen the C++ source code of his inner loop. What is more impressive is that the core algorithm does not need complex math instructions like square root or trig, in fact it does not use floating point instructions or do multiplies and divides!
Okay, this is nonsense. With modern graphics cards, you don't even want to avoid things like that because they're implemented as a single hardware operation! Working around it will slow you down.

Now yes, you can do ray tracing occlusion in an octree with just bitshifts and comparisons, but it's still going to take a lot of operations per pixel. And you have to do it for every light source. It's also not 'point cloud data' but rather an octree voxel.

You can also apply all the same principles to produce depth mapping on a polygon, in fact that's basically what the most advanced bump mapping types already do.
All Shadow priest spells that deal Fire damage now appear green.
Big freaky cereal boxes of death.

Parsifal
Posts: 113
Joined: Thu Feb 28, 2008 1:35 am UTC

Re: Unlimited Detail technology. Yay or nay?

Postby Parsifal » Wed Jun 08, 2011 10:45 am UTC

Tech demo or it didn't happen!

Site is either a pipe-dream, mad grab for VC cash or some sort of hoax.

If anyone can assist with directing us to some immensely high detailed 3d models we would be most grateful.


Ah, there we go. They want free models to run through what is apparently some sort of ray tracer to generate hits for their site.

kmatzen
Posts: 214
Joined: Thu Nov 15, 2007 2:55 pm UTC
Location: Ithaca, NY

Re: Unlimited Detail technology. Yay or nay?

Postby kmatzen » Thu Jun 09, 2011 5:02 am UTC

Parsifal wrote:Ah, there we go. They want free models to run through what is apparently some sort of ray tracer to generate hits for their site.


Yeah, it's not really clear to me how they would acquire models in a cost effective manner. The Digital Michelangelo project was hard enough the way it was and here they are just acting like all you need to do is somehow scan a tree.

Bruce Dell's linkedin profile says he's been the CEO of Unlimited Detail since 1995. Apparently I'm three hops away from one of his software developers via my video game professor from when I did my bachelors. I'm conflicted in my speculation. He's getting real people who supposedly had real jobs beforehand to work for him, but his technology still remains as vaporware 16 years after he claims that he started. How doe something last that long without compromising and releasing a project just to keep the company afloat? If he's been surviving on VC money for 16 years, who is he fooling, how is he fooling them, and how does he have any share left in his company? He has zero verifiable credentials in his profile to suggest that he knows what he's doing. Maybe he does, but he's not conveying it online.

othello
Posts: 1
Joined: Tue Aug 02, 2011 5:56 pm UTC

Re: Unlimited Detail technology. Yay or nay?

Postby othello » Tue Aug 02, 2011 5:57 pm UTC

This isn't really innovative. We have the technology to this this quite easily but it isn't practical for games.

User avatar
Kirby
Posts: 199
Joined: Tue Mar 30, 2010 11:08 pm UTC

Re: Unlimited Detail technology. Yay or nay?

Postby Kirby » Thu Aug 04, 2011 7:43 pm UTC

Notch (of Minecraft fame) made a couple of posts about this on his blog.


Return to “Computer Science”

Who is online

Users browsing this forum: No registered users and 5 guests