Unlimited Detail

[Krej]

http://unlimiteddetailtechnology.com/home.html
vidays
http://www.youtube.com/user/UnlimitedDetail

well forums is named randomness..
idk I saw this shit somewhere and I thought of wolfire blog how its always intressting(almost) and theres sometimes news similar to this, so im just postin here to link it yo.

This stuff just recenrly had some updates with videos, its hard reading the info and watching the videos without being doubtfull but I hope its for real. would be sweet.

peace

[TadMod]

I've seen this before, but it's based on voxel volumes and point clouds, and is therefore completely useless for anything you might want to animate... which is a fair amount.

Still a cool concept, if they can overcome the animation issues

[Krej]

well unless shits fake
animation test
http://www.youtube.com/watch?v=cF8A4bsfKH8

[TadMod]

Yes, but that is a single bird. The point is, that despite using what they call a "Search Algorithm", the point-cloud as a whole must be deformed to suit the animation.

Unless they have created some radical new method of animation, it won't work in today's industry. I only say this because I have a friend who worked in computer graphics programming, and she tried using point-cloud data for animation. Not only did the animation lag, it also had issues such as warping texture masses. I'm not sure whether her voxel display method was significantly more convoluted than these guys, but I imagine that it isn't a truly viable method (at this time) to start using voxels and point-clouds...

Just my 2 cents.

[SamW]

They might be able to animate a 3d point cloud, not by doing animation, but by making a compressed point cloud 'movie'. Meaning each animation frame is just a different set of points that combine together into a 4D point cloud movie. If their search algorithm is good they should be able to go from frame to frame quite fast, but I don't know how much more memory such a 'movie' would take compared to the single object. That would depend on how well they can compress it.

But I believe compression is the basis of infinite detail, so I wouldn't be surprised if they could do point cloud movies with immense compression.

The thing with point clouds, is you don't have to make and render each from scratch. Instead every thing can be built up of something else. That is the magic of using an octree, and having your sub divisions being recursively defined. A model would consist of a set of pointers or search parameters to eight other models, and you can recurse down the tree till you hit a leaf, or till you determine that the space you are searching is already pixel sized anyways. You run a search for each pixel, set some sort of cutoff for search depth and that is it, you have a maximum limit to how many calculations you need per frame, but your data structure can be infinitely complex as long as it fits into memory (and if your model is recursive and recurses to itself, then an infinitely 'complex' if not repetitive model could fit into memory).

The big issue is when you get to procedural animations, and if you want to move a model inside of your octree. That kind of thing could get very messy, but I don't think it is impossible.

[Ragdollmaster]

Besides the practicality problems behind implementing this technology, I don't really see how this is really too useful. It doesn't seem very, uh... aesthetically pleasing. Obviously this is just a technical demonstration and, theoretically, it should look fucking beautiful, but really polygons are pretty useful as it is. After all, we already have three dimensional games that use polygons for this, and they run fine.

And no, I don't mean 3D games as in first person shooters hurfdurf, I mean this bananas, which is actually old news but which many people don't know about for some odd reason.

EDIT: And Inb4 "Omeggle 3D TVs are expensive", they're really not. Not exorbitantly, at least. Plus the prices will inevitably drop as the technology ages- remember how PS3s started at like 600 and dropped to 300? How computers with like 256MB of RAM and a 1 GHZ processor could easily cost over 1500 dollars and now don't even cost a tenth of that amount? Yeah. Be a little patient.



tldr this sounds too good to be true, but if they can apply it and use it in gaming, good for them. Awesome.

[rudel_ic]

Just to clarify, this is basically a topic about voxels.

They might be able to animate a 3d point cloud, not by doing animation, but by making a compressed point cloud 'movie'. Meaning each animation frame is just a different set of points that combine together into a 4D point cloud movie. If their search algorithm is good they should be able to go from frame to frame quite fast, but I don't know how much more memory such a 'movie' would take compared to the single object. That would depend on how well they can compress it.

Another approach is defining each voxel in spacetime as an RGBAt-tuple. So you basically have multiple RGBA-values per voxel, each is then set from a specific time on (t).
Ultimately, static voxel clouds are 3D matrices with one RGBA value per cell. Here, we have 3D matrices with multiple RGBAt-values per cell (at least 1 with t=0).
The advantage would be that there is no structural difference between static and animated clouds. They are formally the same thing, static clouds just have only one RGBAt-value per cell.

This doesn't transfer well to stuff like dynamic gradients and detail level junk though.

Besides the practicality problems behind implementing this technology, I don't really see how this is really too useful.

It's easier to make art for, for starters. You can just literally deform matter instead of deforming matter and then looking whether a polygonal equivalent looks cool. Or you can just laserscan actual things and directly use them as-is.
It's easier to process mathematically. Collision detection is rather trivial, for instance. Not the case for poly stuff, as any 3D programmer surely knows.
You can do a lot of awesome shit since you're actually talking about volumes and not surfaces. As an example, you could handle literal bulletholes instead of just placing awkwardly oriented decals on a surface. Terrains could be way more complex.

Most importantly: If voxels are dense enough, they are indistinguishable from reality with really simple lighting terms; but this goes a little beyond realtime code.

It doesn't seem very, uh... aesthetically pleasing. Obviously this is just a technical demonstration and, theoretically, it should look fucking beautiful, but really polygons are pretty useful as it is. After all, we already have three dimensional games that use polygons for this, and they run fine.

They don't actually run fine. We just sort of can deal with them. But it's really easy to get into messed-up situations with polygonal data, or to stumble onto hard problems.

Simply put, voxels are to polygonal surfaces in 3D what pixels are to vectors in 2D. So there's advantages to each side, but if you want to represent something as complex as reality, you would choose pixels in 2D and voxels in 3D, given infinite resolution.

[Krej]

Just to clarify, this is basically a topic about voxels.

end of the first video tho, about 7:24, he says it v different from voxels.
http://www.youtube.com/unlimiteddetail# ... Ima3Oy18IE

so i guess the topic is more about** is this shit true or fake as they dont explain how it really works, since its new?

Im just hoping its for real it would be cool to experiance some kind of revolutionary tech for games, fo sho

** talk bout wtf u want lOl

[rudel_ic]

end of the first video tho, about 7:24, he says it v different from voxels.
http://www.youtube.com/unlimiteddetail# ... Ima3Oy18IE

Well, as long as he doesn't define a particular point's properties, I fail to see the difference as per his description. Just looks like a specific implementation of voxels.

Sure, I could be wrong there since I've got the same intel as everyone else on the matter.

[SamW]

The difference is that voxels are sampled chunks (usually cube) of space describing the average density of the quantized portion of space. (binary density is able to describe being empty or not empty. But things like MRIs are not binary.)

Point clouds are samples (which can be irregular) of a continuous surface.
When used as a rendering primative they are called surfels

[rudel_ic]

You're guessing that though; he doesn't specify that his points are not evenly distributed, oriented primitives or surface-only. They could as well be evenly distributed, uniformly sized and filling the volumes completely; neither the screenshots nor his description are telling me anything different.

So we're both guessing.