Polygons, voxels, SDFs... what will our geometry be made of in the future?

That’s Spanish for inertia. I interpreted it to mean that polygonal techniques have a huge amount of practitioners and tooling built up around them, and most people will improve or iterate on something they know before switching to something relatively unknown and unproven.
 
Yes, I know the meaning of the word, but I thought he maybe was referring to the name of a graphic engine which works with SDFs, or the name of a game built around it, or the name of a developer studio that is currently working on that... :D
 
That’s Spanish for inertia. I interpreted it to mean that polygonal techniques have a huge amount of practitioners and tooling built up around them, and most people will improve or iterate on something they know before switching to something relatively unknown and unproven.
Precisely. Only it wasnt spanish, but portuguese, unintentionally. I was a bit drunk...
 
I don't think that's voxels. Here's a demo from two years ago basiclaly showing the same thing -

The voxels are dynamically scaled, and you see them rotate on the wheels as well as move around the scene less than one voxel at a time. I think they're using a different representation such as SDF and layering a voxelised visualisation on top.
 
The voxels are dynamically scaled, and you see them rotate on the wheels as well as move around the scene less than one voxel at a time. I think they're using a different representation such as SDF and layering a voxelised visualisation on top.
ًWhy does it look low res and wobbly like that?
 
The voxels are dynamically scaled, and you see them rotate on the wheels as well as move around the scene less than one voxel at a time. I think they're using a different representation such as SDF and layering a voxelised visualisation on top.

What's keeping a voxel implementation to have multiple volumes, one for each individual moving part, just like they do with SDFs?
 
ًWhy does it look low res and wobbly like that?
Low res? They can draw the voxels at whatever resolution, but there'll be limits on memory and whatnot as you zoom in.
What's keeping a voxel implementation to have multiple volumes, one for each individual moving part, just like they do with SDFs?
Because you then break your voxel data structure which is super-optimised to allow fast drawing, in theory. Let's say the wheel is in its own voxel space, so it can rotate. And then the car is, so it can move freely and doesn't have to be grid aligned. And the leaves on a tree, and each rock...you end up with something that isn't a voxel representation of the world, and how do you accelerate the search structures with everything broken down into their own small spaces? The easiest, best way is to represent that space with a different form, like SDF, and then voxelise. Which would look exactly like they're doing here.

Notice their animation isn't combined with their 'infinite detail' demo. Why haven't they got an 'infitinite detail' scene and be animating that? A car in 'infinite detail', scanned in from a real life model, instead of something that looks like a game model? From a first demo, sure. Two years after the car and destruction was shown, they haven't added that to their 'future of graphics' hyping demos?
 
Low res? They can draw the voxels at whatever resolution, but there'll be limits on memory and whatnot as you zoom in.
Because you then break your voxel data structure which is super-optimised to allow fast drawing, in theory. Let's say the wheel is in its own voxel space, so it can rotate. And then the car is, so it can move freely and doesn't have to be grid aligned. And the leaves on a tree, and each rock...you end up with something that isn't a voxel representation of the world, and how do you accelerate the search structures with everything broken down into their own small spaces? The easiest, best way is to represent that space with a different form, like SDF, and then voxelise. Which would look exactly like they're doing here.

Notice their animation isn't combined with their 'infinite detail' demo. Why haven't they got an 'infitinite detail' scene and be animating that? A car in 'infinite detail', scanned in from a real life model, instead of something that looks like a game model? From a first demo, sure. Two years after the car and destruction was shown, they haven't added that to their 'future of graphics' hyping demos?
Their 'infinite detail' demo? What demo? Are you sure you're not mistaking them for Euclideon, with their 'unlimited detail' stuff?

Another question: now SDFs seem to be getting a lot of interest, do you think this non-polygonal solution with make people forget about voxels for geometry? It seems better.
 
Yes, having a different vixel volume for each moving part is not very efficient for highly dynamic scenes, thats why they have few dynamic objects in that demo. If you used SDF you'd still be doing the same thing. You'd be tracing against multiple volumes, they just encode the opacity info as distance to hit instead of binary solid/nonsolid. Either these volumes are full 3d textures (Unreal 4) or just analytical shapes (Dreams, Claybook).

This doesn't show infinite detail because i5 is not from that company. This is automontage, another, less hypie engine, which has been around for many years already.

EDIT: actually, Clayworld does stamp the many volumes into one big volume before tracing, more in line with what you described, shifty. Dreams also does something similar. But unreal engine's SDF AO, Soft shadows and the in-engine visualization of the SDF representation, all trace the many volumes directly.
 
Their 'infinite detail' demo? What demo? Are you sure you're not mistaking them for Euclideon, with their 'unlimited detail' stuff?
Indeed I am. It was the 'the future is volumetric' line, and the lack of visible progress (same demo content from two years earlier). that said, the show reel did incorporate realtime voxelised objects (the people) which is something. That'd be good for VR communications.
Another question: now SDFs seem to be getting a lot of interest, do you think this non-polygonal solution with make people forget about voxels for geometry? It seems better.
Yes. Voxels are a brute-force solution to volume representation. Representing volumes in a arbitrary form allows conversion to a visual 2D representation however you want, such as traced solids or Dreams' splats or sampling regular-grid voxels.
 
Last edited:
Further background and details on Atomontage.

I very much doubt the groundless speculation that this is SDF geometry.

pRZJJin.jpg

r0yGFhm.jpg
 
Last edited:
This is what the home page information on their website says:


THE INEVITABLE VOLUMETRIC FUTURE

MISSION
We are inducing an inevitable paradigm shift in interactive computer simulation and graphics. This allows us to build highly detailed, massively shared, and deeply editable volumetric worlds. There, millions of people will work, play, and enjoy each others' unrestrained creative output

PROBLEM
The currently dominant 3D graphics and simulation paradigm, based on triangle/polygon meshes, is fundamentally flawed

SOLUTION
Volumetric, sample-based geometry: "volume pixels" aka voxels solve these issues - if they can be processed fast enough for real-time simulations on common client hardware.
When scaling from millions of voxels to billions and beyond, one is presented with a difficult data management problem. Atomontage has focused fifteen years of R&D on this challenge, resulting in patent-pending breakthroughs that finally allow volumetric simulations to scale in fully interactive AR/VR applications.
Other features recently proven viable with our voxel tech include complex rigid-body physics, soft-body deformations, standard PBR shaders, and efficient playback of multiple volumetric video streams

atomontage.com
 
Back
Top