Euclideon Island Demo 2011 "Unlimited Detail"

[gorobei]

The developer of Minecraft seems to think they are: http://notch.tumblr.com/post/8386977075/its-a-scam

yes and no. he hasnt watched and listened to the original videos.
http://www.youtube.com/user/UnlimitedDetail

they are using octrees, but not voxels. Point clouds are not voxels.
they do appear to be sorting the points using octrees to isolate only the relevant/visible-to-camera points.

as i mentioned in the locked thread, you can apply an octree to a 3d object and use it to store any kind of data(texture info in the case of the sigraph presentation i mentioned). if used to parse all the point cloud data, then it allows them to sort through down to whatever laser scan resolution they are using without actually loading all the points of the cloud or rendering voxels into polys.

animation demo
http://www.youtube.com/watch?v=cF8A4bsfKH8&feature=related
looks like they are rigid attaching the scanned feathers and other mini parts to a rigged deformer mesh. the cloud sub parts are attached to something moving, but the points in each cloud are not actually being bent by joint deformers. would probably work fine with robots and fur/scaled creatures and not so well with flesh(unless you decide to model down to individual skin cells).

 

[Mopetar]

The developer of Minecraft seems to think they are: http://notch.tumblr.com/post/8386977075/its-a-scam

His post makes terribly use of a math to try and prove a point. His calculation of 512 PB of data assumes that the entire volume of a game level is full of points. This fails on a basic level as most of a level will be open air, drastically reducing his figure. It's a matter of surface area vs. volume.

It also fails to assume that none of the data can be duplicated. Current designs reuse a lot of the same assets so why should this new method require that every single tree or blade of grass is unique down to every molecule? For that matter, parts of larger objects can be reused and mixed together. Once you start reusing parts, the amount of data necessary drops significantly.

I don't mind valid criticisms of something, but this seems to miss the mark.

 

[Mr. President]

His post makes terribly use of a math to try and prove a point. His calculation of 512 PB of data assumes that the entire volume of a game level is full of points. This fails on a basic level as most of a level will be open air, drastically reducing his figure. It's a matter of surface area vs. volume.

Not to mention the actual volume of each object. Those rocks and boulders, for example, aren't going to be filled with unseen point data.

I'm still quite skeptical of the whole thing but it's not because the maker of Minecraft decided to 'tell them how it is'.

 

[Deathray2K]

I see those comments from Notch posted everywhere and it's pretty annoying how people take his word as law. His math wouldn't even be right if they were voxels, since he doesn't account for empty space or hollow objects.

I think this is perfect for solid objects, and they even say they have animation working (I'm sure we'll see how well that works in a few months). I think a bigger problem will be integrating this seamlessly with traditional rasterisation (or raytracing) for stuff like transparency, volumetrics (real or faked), liquids, etc. Physics might also have some problems, and I suspect will have to use polygonal collision meshes on top of the point cloud geometry.

I'm excited to see where this goes.

 

[BFG10K]

He's actually responded: http://notch.tumblr.com/post/8423008802/but-notch-its-not-a-scam

Why I assume it’s voxels and not point clouds:
* Voxels store only the information about each point, and their positions are implicit in the location of where the voxel is stored. Point cloud data stores both the information about each point and the position of each point.
* They mention “64 atoms per cubic millimeter”, which is 4*4*4 points per mm^2. While it’s possible they only refer to the sampling frequency for turning polygonal structures into point data, the numbers are just too round for me to ignore as a programmer.
* All repeated structures in the world are all facing the same direction. To me, that means they aren’t able to easily rotate them arbitrarily.

About the size calculation:
* I was trying to show that there was no way there was that much UNIQUE data in the world, and that everything had to be made up of repeated chunks.
* One byte per voxel is way lower than the raw data you’d need. In reality, you’d probably want to track at least 24 bits of color and eight bits of normal vector data per voxel. That’s four times as much data. It’s quite possible you’d want to track even more data.
* If the data compresses down to 1%, it would still be 1 700 three-terrabyte hard drives of data at one byte of raw data per voxel.

To me this demo is trying to pass off that it's something new when it really isn't.

 

[gorobei]

He's actually responded: http://notch.tumblr.com/post/8423008802/but-notch-its-not-a-scam


To me this demo is trying to pass off that it's something new when it really isn't.

Did you watch the older videos? He explicitly states "not voxels".

Dell(the narrator) doesnt do himself any service by calling them atoms, as that is a term associated with voxels and mri scan renderings. I suspect he keeps trying to dumb down the analogy for lay people who have no experience with laser scans or point cloud data. The fact that they are converting poly mesh models (fictional models as he terms it) means they are simply freezing out subdivision surface models and stripping out the poly/edge data leaving only vertices/points.

Notch states at the start he may be wrong. I'm pretty certain the hummingbird demo video indicates some animation is possible.

The new part isnt the octree sorting, it is the switch to point data without eventual polygon rasterization. Even voxels get rasterized at some point.

 

[sm625]

So does this mean that intel could buy these guys and tune their gpu specifically for this type of rendering? I hope AMD and nvidia are watching these guys very closely.

 

[pcm81]

http://www.youtube.com/watch?v=00gAbgBu8R4&feature=channel_video_title

Atoms > Polygons?

This company is claiming 100,000 more detail than using Polygon technology. I'm not knowledgeable enough to doubt these claims, so what do you guys think?

They claim that island is made of 21 trillion "atoms".
So does that mean you need a video card or computer with 21 TerraBytes of RAM? Actually allot more than 21TerraBytes, since you need to store 3d location and color for each "atom"

 

[BFG10K]

Did you watch the older videos? He explicitly states "not voxels".

I'm not arguing one way or another, I'm just presenting evidence for alternative view-points.

 

[Piotrsama]

I am not that surprised. Before when computers where slow (486) there was games that used block-like graphics to handle more detailed landscapes, because handling that many polygons was to slow at that detail level (some helicopter games etc. that I dont remember the name to anymore).

Probably Comanche (1992) with their VoxelSpace engine.
http://www.youtube.com/watch?v=u1u3EoN-6V0&feature=related

 

[happy medium]

I think he answed most of your questions. These videos were posted today 8/10/2011 on youtube. Real impressive stuff.
Thanks, Killercheese for posting these in the locked thread..

http://www.youtube.com/watch?v=JVB1ayT6Fdc

http://www.youtube.com/watch?v=JVB1ayT6Fdc&feature=player_detailpage#t=1326s

 

[AnandThenMan]

One thing completely missing from the demo is movement. Not camera movement, but nothing in the environment deforms, moves, changes shape. It's all static. I am far from an expert in this field, but what this looks like to me is scanned objects rendered as fractals.

 

[happy medium]

One thing completely missing from the demo is movement. Not camera movement, but nothing in the environment deforms, moves, changes shape. It's all static. I am far from an expert in this field, but what this looks like to me is scanned objects rendered as fractals.

I watched the entire video and he says , the working movement models are not quite ready. He also said that the demo was put together in 2 or 3 weeks. I for one think this is real and comming soon.

 

[ViRGE]

I watched the entire video and he says , the working movement models are not quite ready. He also said that the demo was put together in 2 or 3 weeks. I for one think this is real and comming soon.

I expect you'll be disappointed. It's not that the technology is a scam, but it's extremely, massively resource intensive. This would be akin to going back to the 1970s and giving Digital Equipment Corporation a demo of a GTX 580 in action. It's so far beyond what the technology is capable of that it won't be viable for many years.

Not that it's the only such technology either - there are a number of technologies in any given field of CS that are fundamentally resource limited. Some eventually come to pass when the hardware catches up, others never see the light of day because something better comes along first.

 

[jimhsu]

I still don't get the rotation (and yes, I did watch the entire video). Do simple transformations (scale/rotate) require a new "voxel/pixel/point cloud/atom/whatever" object? If not, can't you do a demo to randomize every object in the scene by a small magnitude just to show that you can?

 

[AnandThenMan]

I expect you'll be disappointed. It's not that the technology is a scam, but it's extremely, massively resource intensive.

This is what I'm thinking, the processor time required to move and deform the objects will make this rendering technique useless on today's hardware. I fully admit that is just a guess however. But as they say, no free lunch, I just can't understand how doing "unlimited" detail in a dynamic environment is feasible, I don't care how clever and efficient the routines are. Even if you didn't render a single pixel you didn't need to, there is still just too much to deal with in a dynamic setting.

 

[AnandThenMan]

If not, can't you do a demo to randomize every object in the scene by a small magnitude just to show that you can?

I'm sure you can. But that is still a static environment. Now toss in collision, dynamic lighting, shadows, degrading objects, fire, smoke, particles, destructible objects etc. and suddenly your CPU and GPU is choking.

Yes I'm extremely skeptical.

 

[happy medium]

I expect you'll be disappointed. It's not that the technology is a scam, but it's extremely, massively resource intensive. This would be akin to going back to the 1970s and giving Digital Equipment Corporation a demo of a GTX 580 in action. It's so far beyond what the technology is capable of that it won't be viable for many years.

Not that it's the only such technology either - there are a number of technologies in any given field of CS that are fundamentally resource limited. Some eventually come to pass when the hardware catches up, others never see the light of day because something better comes along first.

Well the demo he was running was on a laptop with no use of the gpu, i7 cpu, and 8gb of memory, and I would guess a 1tb harddrive at most.
Hardly what I consider massive amounts of power. He said it was unoptimized and running on just cpu power. Is he lying?

Did you guys watch the videos I posted in #36 or the OP?

 

[ViRGE]

Well the demo he was running was on a laptop with no use of the gpu, i7 cpu, and 8gb of memory, and I would guess a 1tb harddrive at most.
Hardly what I consider massive amounts of power. He said it was unoptimized and running on just cpu power. Is he lying?

Did you guys watch the videos I posted in #36 or the OP?

Is he lying? No, not really. But he's clearly maxing out an i7 system with only the most barebone of demos. Unique objects in place of replicated objects would bring the whole thing crashing to its knees.

 

[Red Hawk]

The demo had only very basic lighting, no animation, no post-processing effects, no demonstration of physics, etc. It still has a very long way to go before it becomes usable on modern PC hardware, even more so for the current generation of consoles (they've claimed the technology can work on modern consoles).

 

[happy medium]

Is he lying? No, not really. But he's clearly maxing out an i7 system with only the most barebone of demos. Unique objects in place of replicated objects would bring the whole thing crashing to its knees.

Oh, and I forgot to add only 1 core of the cpu. I would imagine a gpu would run it 10x faster.

 

[ViRGE]

Oh, and I forgot to add only 1 core of the cpu. I would imagine a gpu would run it 10x faster.

The problem isn't computational with respect to rendering - it's storage and I/O. If he's using Octrees, then he's significantly reducing the load by reusing objects. A dozen shrubs takes up the same amount of space as 1 shrub if they're all identical. Replace them with unique shrubs and you aren't rendering a significantly different number of points, but now you have to store and process 12 times the data.

 

[AtenRa]

They claim that island is made of 21 trillion "atoms".
So does that mean you need a video card or computer with 21 TerraBytes of RAM? Actually allot more than 21TerraBytes, since you need to store 3d location and color for each "atom"

http://www.youtube.com/watch?v=JVB1ayT6Fdc

4:55 onwards,

We have a search algorithm and it grabs one atom for every pixel on the screen. (1280x768 = 983040 pixels or atoms)

If you do it that way, you end up been able to have unlimited geometry but not been wasteful on how we presented on the screen.

I believe that’s the secret, they project only 1280x768 pixels/atoms every time no matter how big the island is.

One thing to consider is the view distance detail

Watch the video at 23:50

The island is all the way zoomed out, at that view distance the island looks like pixels and it’s clearer as the guy zoom in at 23:57. (Edit: it is clearer at 26:30)

If I understand it right, because the algorithm only grabs 1280x768 pixels every time, when you zoom out that far away the detail is getting smaller because the engine keeps projecting only 1280x768 pixels/atoms.

Zooming in will get you higher detail of the objects and if you zoom in close to the elephant the engine still projects 1280x768 pixels but you only see two three objects like the Elephant and a tree.

You really need to zoom in very close to the elephant in order to see the higher detail of the model. So if you zoom in very close that you only can see the elephant in the screen, that will be the higher detail of the model, the engine will again only project 1280x768 pixels/atoms.


I could be wrong though

Edit: @27:10 LOD (level of distance)

@28:02

So it’s not a trick, like building the objects numerous times, it really does just keeps tracks of sorting the points and grabs only the ones we need.

 

[happy medium]

The problem isn't computational with respect to rendering - it's storage and I/O. If he's using Octrees, then he's significantly reducing the load by reusing objects. A dozen shrubs takes up the same amount of space as 1 shrub if they're all identical. Replace them with unique shrubs and you aren't rendering a significantly different number of points, but now you have to store and process 12 times the data.

What about the times in the video when he was rendering 20 different objects at once? WHat do you mean by I/O , memory?

 

[Arkadrel]

The problem isn't computational with respect to rendering - it's storage and I/O. If he's using Octrees, then he's significantly reducing the load by reusing objects. A dozen shrubs takes up the same amount of space as 1 shrub if they're all identical. Replace them with unique shrubs and you aren't rendering a significantly different number of points, but now you have to store and process 12 times the data.

That is a very small issue.... most people dont mind seeing only 12 tree types in a game.

Its the movement thing thats a biggy.

If they can show leafs blowing in the wind, or a character model (that detailed) walking, they ll be a long way towards convinceing me that this is the future of gameing.

Then they ll slowly just add in all the otherstuff people have mentioned.