Unlimited Detail

[JamesE]

Pulled this from the modo forum.

[kay_Eva]

that's awesome. If it is developed I hope it isn't as unattainable for most people as procedural animations software is. But it probably will be.

[ifxs]

Whoa!! Thats either a ploy to get startup capital for another B.S. scheme, or a darn revolutionary "Voxel-CRUSHING" method for "3D" drawing on screen... wonder what creating content for that environment is like, starting with "atoms" perhaps... ;-) LOL

I've seen similar "sparse vox-tree" details released by John Carmack from ID, regarding their new realtime renderengine, and in fact I've brought it up in the forum suggesting this approach for 3DC, oddly enough my mention of this method was met with some apprehension from some of the members on the forum, but IMO this type of method would make 3DC absurdly fast for users even on mediocre systems.

to the best of my current understanding, these next gen voxel-like methods are using something like a raycast, or multiple raycasts to know the size limit for which "voxels" are needed to be drawn at the screen pixel dimension, I'm still studying it. hopefully 3DC catches onto this method for the betterment of all of our 3DC user experiences. AFAIK this produces MAJOR MAJOR advances over every single(afaik) rasterization techniques know currently

A current issue with 3DC IMO is the lack of amy optimization of the polygon skin(or even the existence of the poly skin) that is overlaid on the ENTIRE voxel mesh, I could go on at length regarding the ways in which this type of voxel representation could be optimized, hopefully soon we'll get a 3DC display revolution... OF course, I've never seen Andrew's code, so this is likely a FAR more complex process than that of a simple statement a user of 3DC can make.

[Cube]

I've seen similar "sparse vox-tree" details released by John Carmack from ID, regarding their new realtime renderengine, and in fact I've brought it up in the forum suggesting this approach for 3DC, oddly enough my mention of this method was met with some apprehension from some of the members on the forum, but IMO this type of method would make 3DC absurdly fast for users even on mediocre systems.

to the best of my current understanding, these next gen voxel-like methods are using something like a raycast, or multiple raycasts to know the size limit for which "voxels" are needed to be drawn at the screen pixel dimension, I'm still studying it. hopefully 3DC catches onto this method for the betterment of all of our 3DC user experiences. AFAIK this produces MAJOR MAJOR advances over every single(afaik) rasterization techniques know currently

Did you post about it on this forum? I ran several searches before i posted about ID Tech 6 a week or so ago, and i never came up with more than a handful of results for anything (ID Tech 6, Carmack, SVO etc...) and all of them were from 2008/2009.

About raycasting into a sparse voxel octree, it will hopefully be feasible for real time rendering in games etc.. in a few years time, but you can't edit a SVO on the fly so it's useless for 3d-Coat unfortunately, it's a compression format for streaming pre-defined volume data. IE you would make your model in 3D-Coat, then convert the finished item to SVO format to use as a static object in your game engine.

This Unlimited Detail thing looks interesting i have to admit, they need to make some better demo's though - they look terrible!

[ifxs]

but you can't edit a SVO on the fly so it's useless for 3d-Coat unfortunately, it's a compression format for streaming pre-defined volume data.

darn, well, good for pre-defined assets then.. and yes, I've made posts regarding sparse vox-trees on the 3DC forum. I think they were in the "3.2 updates" thread, not positive though.

[T.H. ROCK]

Hmmm, I feel I need to wash after that, but not sure why.

[splodge]

Interesting!

The only catch that I see is the storage requirements for a video game based on this technology, it's gonna be something like a 250GB installation size. It'll also require a whole lot of system memory too.

But the despite the storage problems it's definitely the future. It just means we're now gonna have to depend on Western Digital and Corsair instead of NVidia

[philnolan3d]

One question is "how easy is it to work with?". For example making a large smoothly curved surface is fairly easy with sub-Ds, but how easy would it be here?

[ifxs]

One question is "how easy is it to work with?". For example making a large smoothly curved surface is fairly easy with sub-Ds, but how easy would it be here?

Excellent point Phil. I've been pondering the same thing... creating an object with SubD surfaces is indeed easier(for now) than starting to create a multimillion point cloud... Imagine the creative workflow for creating transparent objects that need to refract(still need to raycast in form form likely), and also have an internal structure that has non-uniform specific color/"textures", etc...

I've had some ideas on this on how a developer could approach this, pretty tricky in many ways... That'll be the element that adds to the adoption of this tech, the toolset to create with such a display method/volume structure, and an agreed upon "model" format for display in various realtime-engines of this type, amongst others

[splodge]

One question is "how easy is it to work with?". For example making a large smoothly curved surface is fairly easy with sub-Ds, but how easy would it be here?

It's only a rendering engine, not a modeling application. Modelers will just use their usual methods/tools (NURBS, voxels or polys) and then convert them to point data later. Although I'm sure somebody will eventually create a modeling application thats built upon the rendering engine.

[GED]

wonder if they can get animation/ shaders in there? it does seem like a crazy big step forward if this is true

[splodge]

wonder if they can get animation/ shaders in there? it does seem like a crazy big step forward if this is true

I've always thought that this kind of tech would come eventually and yet I've been concerned over the animation aspects of it. Animating polygons is easy as you only need to move a relatively small number of vertices around, and yet with voxels or point clouds it's going to take a considerable amount of processing time to shift all those points or voxels around.

And I noticed in their video that their animated characters seem to be made up of separate moving parts, as opposed to being organic figures with seamless limbs. Animating separate parts in their system will obviously be easier as they don't need to move the actual points but rather they just adjust an offset value for each part. But if they needed to animate something organic like the tentacle of an octopus then it's gonna require some fairly heavy processing.

So maybe animation is the catch?

[Nemoid_]

well, if it is a last step of the process i see no real prob. We could model, texture, animate, then some app could transform

all that in point clouds for game engine?

[GED]

seems like there are a few horses in this race http://atomontage.com/?id=gallery

[ifxs]

seems like there are a few horses in this race http://atomontage.com/?id=gallery

very cool demo, actually interaction of elements is much better for a proof of concept engine, even if it's secretly pre-rendered. Thanks for pointing this one out!

[Digital777]

Interesting, in the video they say they gave a big company a tech demo and they got turned down. If that's true then i guess the real reason they stay with polygons is so people need to keep buying graphics cards to keep up with the demand as more polygons were needed in games but this would avoid the need for that.

In general though when new technology like this appears it will spread quickly so as soon as they release the SDKs it will probably be used by some popular game company's right away because they just want things to look as good as possible. One of the comments says it's made by engineers and not artists so it would be interesting to see what a good games company could make with this, basically it seems things are on the way to what people are doing now with rendering apps but displayed realtime in games.

[spacepainter]

http://artis.imag.fr/Publications/2009/CNLSE09/GigaVoxels_Siggraph09_Slides.pdf

[Cube]

The most frustrating thing about the proposed models for using SVO/Point Cloud/whatever methods for real-time 3D is that ultimately all of it will be determined by what Sony/Microsoft decide to do with their next consoles. I think it's safe to say they they will definately be an order of magnitude faster than the Xbox 360/PS3, and hopefully very programmable (since ATI and Nvidia have both gone that route, maybe even Larrabee will be used), but i am worried that they will save costs on the consoles with limited HDD space which will make all these methods far more difficult (or impossible) to impliment. Even with heavy compression it looks like games made with volumetric data will be easily 10x plus the size of current releases, and you will probably need a very fast HDD (maybe SSD) to stream all that data.

I know gaming is only one market, but if this stuff really does pan out then it will be a very exciting time. Since 3D-Coat is already quite far advanced in Voxel modeling i could easily see it becoming the #1 software to use in 2-3 years, at least for making world assets.