"Unlimited detail" pop up again

Karlos

Karlos

Supper Moderator
Moderator
Joined
Apr 1, 2005
Messages
585
Reaction score
189
Some of you may recall a small Australian company about a year or two ago claiming to have found a way to make 3D point cloud data (normally used for high resolution medical imaging etc) suitable for use in game engines.

Not long after the announcements were made, they vanished.

Well, it seems they are back and have been working on tools to make life easier for existing 3D artists to convert their assets over to their format.

http://www.thinq.co.uk/2011/8/2/euclideon-promises-graphics-detail-breakthrough/

Interesting stuff, though I noticed in their criticism of polygon graphics, they gave tessellation a wide berth ;)
 
  • Like
Reactions: adz
That is enormously cool. Showed it to a buddy of mine who is a 3d artist and he's excited.
 
a GIMP group I'm a member of has been talking about this......we need to see more evidence, but it sure looks interesting.......comments about Star Trek Holograms were fly fast and furiously :lol:
 
Actually there is a lot of stuff out there. If you take a look on youtube and search for sparse octree voxel there are quite a few people working on this sort of stuff. Quite impressive.
 
I do have some questions about this rendering technique. For static scenes, I can see that there is probably some algorithm that works well, but none of the video clips I've ever seen of this show any animation of any sort. I wouldn't be surprised to learn that it may not be so straightforward to animate objects in the scene.

Secondly, their rendering technique presently runs in software. If it can be rewritten for shader cores it will become much more useful, I'm sure. There are already point cloud rendering systems that are based on CUDA and OpenCL so I expect this isn't infeasible.
 
Now that I have my preferred GPGPU SDK all up and running again, I might have to look into some of these sparse tree rendering techniques to see if they can be rendered on the GPU. I would very much resent having spent money on a graphics card, only to have the next generation of 3D run on the CPU :lol:
 
Revisiting their demonstration, one feature that is missing that could add an enormous amount of realism to their rendering is ambient occlusion. This basically simulates the loss of overall illumination you get where objects block the overall amount of light. Compare the left and right hand sides of the sample rendering below, to get an idea:

Screen_Space_Ambient_Occlusion.jpg


There are well-established methods for achieving the latter in "screen space" on the GPU and best of all, theoretically it doesn't depend on polygon rendering since the technique actually operates directly on the framebuffer and uses only the depth buffer to estimate the effect. So, providing your funky new non-polygon-based rendering technique can write depth information for the scene whilst it's plotting the point cloud, you can run SSAO over the scene afterwards and improve the quality of the lighting significantly.
 
I was very impressed by the voxel engine and thought of it as the future of 3d gaming. I really love to see a new approach :D
 
Karlos said:
Now that I have my preferred GPGPU SDK all up and running again, I might have to look into some of these sparse tree rendering techniques to see if they can be rendered on the GPU. I would very much resent having spent money on a graphics card, only to have the next generation of 3D run on the CPU :lol:
Click to expand...
Think Amiga! Everything can be hardware accelerated! Some modern 3D cards support physics as well, while that always has been the CPU's job.
 
Yep, my card supports the PhysX API. What is really needed, however, is a Havok wrapper, since that's what most of the game engines use.

Of course, now that DirectCompute has been added to the DirectX repertoire and OpenCL is finally taking steps, I expect future titles could simply roll their own.
 
Speelgoedmannetje said:
I was very impressed by the voxel engine and thought of it as the future of 3d gaming. I really love to see a new approach :D
Click to expand...

Hehe, don't let Bruce hear you call it a voxel engine, you'll upset him. He's quite clear on the distinction between voxel techniques and point cloud data.

Strictly speaking, rendering point clouds is not actually new. This type of rendering has been used in scientific and medical imaging for quite a while. However, I've never seen it used for a game engine before.

Whilst it looks nice, to reiterate my concerns about it:

1) No as-yet demonstrated examples of animated scenes. If I were a game engine designer, I'd definitely want to see that it isn't limited to static scenery before investing in it.
2) No as-yet demonstrated GPU renderer. This could also be a deal-breaker since for all it seemingly runs well on a CPU, concerns about the scalability with 2D screen resolution and the amount of CPU time left to perform other tasks, eg AI etc. Add to this that having a CPU only renderer could complicate the use of established GPU effects such as HDR rendering tricks, SSAO, motion blur etc.
3) How relevant the whole technology is in the face of recent developments in polygon rendering. The main impetus for this technology was that polygons divide the world up into obvious flat-edged facets that are conspicuous. However, the presentation is a little disingenuous. Modern polygon-engines have been able to use procedural techniques for generating detail dynamically (just look at Crysis 1's lush jungle vegetation for an example). Then came geometry shaders which allow vertex generation on the fly and finally, displacement-mapped tessellation in DX11 / OpenGL 4 which allows you to increase the vertex count by silly amounts. Nothing in the video that was demonstrated thus far is beyond the latter in terms of visual quality.
 
this debate is certainly way more interesting than calling people republicans and democrats
:lol:
 
You must log in or register to reply here.
Back
Top