Lets talk about possible unconventional methods....

[Nesh]

We know that developers have been using more or less standard methods in producing visuals.

Now I ve been thinking, "how about developing other methods and techniques to produce high detail surfaces or surfaces in general?"

The general idea so far was to use polygons and add textures and normal maps.

When I see surfaces in real life though there are no polygons. There are smooth surfaces, independent of the excistence of digital polygon surfaces.

It would be interesting if a company develops a different technique that doesnt involve the use of polygons.

Like for example the use of images changing angle and geometry in a similar manner normal maps do but without resting on a polygon surface. Or a creation of a surface that isnt made out of polygons but uses a different technique.

These are just some ideas that may sound stupid but you get the picture. The first organization (developer, hardware producer etc )who manages to create a new teqhnique that requires less hassle and less resources will bring out HUGE possibilities.

Now I dont want to shoot down the possibilities, consider this thread useless and get it locked.

Lets think out of the box.

 

[Squeak]

 

[Laa-Yosh]

Polygons and textures have sticked around for decades for the simple reason that they're the best all-round solution. All that's left is to increase the fidelity and get rid of the low-res edges.

Flames, smoke, and some other stuff are obviously not suited to this technology though, and the current method of transparent quads could really get a replacement - however we need a lot more processing power to do that.

There's a lot of past discussion about this topic on the B3D forums though, so if you're really interested, try using the search function.

 

[Acert93]

Come on Laa-Yosh, you know you wanna work with Voxels and some freaky HOS

 

[DieH@rd]

Voxels for the win.

 

[Neb]

I remember Outcast, the voxels made it beatiful and asweel as the effects and detail, and AI, insane for a game from spring '99. Ran great to on a 500MHz Celeron. I thought I remembered terrain deformation for fotsteps (snow), hmm?

Crysis do have some voxel rendering.

 

[Nesh]

Polygons and textures have sticked around for decades for the simple reason that they're the best all-round solution. All that's left is to increase the fidelity and get rid of the low-res edges.

Flames, smoke, and some other stuff are obviously not suited to this technology though, and the current method of transparent quads could really get a replacement - however we need a lot more processing power to do that.

There's a lot of past discussion about this topic on the B3D forums though, so if you're really interested, try using the search function.

It is probably the best solution because there isnt any other better existing method. But how about trying to think of/design a different new solution?

You are thinking about what we have now. Which isnt the point of the thread

 

[game-hunter]

Nvidia's first 3D accelerator anybody?

The NV1 http://en.wikipedia.org/wiki/NV1
I don't know how many of you remember this chip but it used quadratic curved surfaces. I wonder how this technology would have progressed had they not switched over to polygons.

 

[BoardBonobo]

Didn't the original Comanche (novalogic) use voxels for terrain and cloud rendering? I vaguely remember playing it on a 386 with a pVGA card.

 

[Shifty Geezer]

It is probably the best solution because there isnt any other better existing method. But how about trying to think of/design a different new solution?

You are thinking about what we have now. Which isnt the point of the thread

No, he's covering all the known quantities, just as the 3D engineers who work on creating fast 3D renderers have. Do you really think we're going to hit upon something here that no-one has thought of before?!

A quick summary. 3D tessellated triangle renders are damned fast and flexible. Perfect curves either need

1. super-duper high resolution models
2. rendering without tessellation
3. tessellation down to smooth contours
4. some jiggery-pokery to remove the polygonal edges

1) Not possible due to memory, bandwidth, and processing power constraints.

2) Requires HOS model representation which has issues, and a renderer that doesn't tessellate. Ray-Tracing can successfully render HOS without tessellation but is an order of magnitude (or two. Or three) slower than current rasterized graphics. Some people are working on Ray Tracing accelerators but they're not seeing amazing results that can compete with rasterized graphics. See the Ray Tracing thread for more discussion. Now if 3D accelerator manufacturers really targeted this technology, perhaps they'd make headway? But the technology isn't being progressed, and with the hardware we have to use, it's not viable.

3) This again needs HOS surfaces which aren't friendly to artists. Straight tessellation throws up the same sort of problems as number 1). Clever tessellation hasn't been developed as a software technology yet. Try a search for HOS techniques.

4) We hit on this in a recent discussion and nAo pointed me to a paper on rendering a silhouette map for polygon models. It was quite old and I'm a bit surprised the technology hasn't made an appearance anywhere, but you can be sure developers keep an eye on these sorts of papers and play around, so I'm sure there are legitimate reasons for it's absence.

You then have things like Voxels, which all have their own issues. Real Life has infinite storage capacity to store 3D models at perfect resolution. Computers don't have that luxury, and storing objects by volume is incredibly expensive - one problem that triangle meshes has got around.

So if you're interested, as Laa-Yosh says go do some searching. But don't think developers are ignoring some obvious out-of-the-box solutions or haven't done lots of brainstorming to try and do something different. There are incredibly clever folk out there working on 3D, and have done for decades, both for realtime and offline rendering where speed is always sort along with flexibility in the pursuit of attaining realism. The maths they come up with for shortcuts to approximating GI and other things boggles the mind. The odds of them missing a trick are vanishingly small. It may happen, but in my following of 3D rendering, I've not yet come across an idea that hasn't been explored a good 10 years earlier than anyone would expect, and instead have been bowled over by the ingenuity of developers in speeding things up.

 

[Nesh]

No, he's covering all the known quantities, just as the 3D engineers who work on creating fast 3D renderers have. Do you really think we're going to hit upon something here that no-one has thought of before?!

A quick summary. 3D tessellated triangle renders are damned fast and flexible. Perfect curves either need

1. super-duper high resolution models
2. rendering without tessellation
3. tessellation down to smooth contours
4. some jiggery-pokery to remove the polygonal edges

1) Not possible due to memory, bandwidth, and processing power constraints.

2) Requires HOS model representation which has issues, and a renderer that doesn't tessellate. Ray-Tracing can successfully render HOS without tessellation but is an order of magnitude (or two. Or three) slower than current rasterized graphics. Some people are working on Ray Tracing accelerators but they're not seeing amazing results that can compete with rasterized graphics. See the Ray Tracing thread for more discussion. Now if 3D accelerator manufacturers really targeted this technology, perhaps they'd make headway? But the technology isn't being progressed, and with the hardware we have to use, it's not viable.

3) This again needs HOS surfaces which aren't friendly to artists. Straight tessellation throws up the same sort of problems as number 1). Clever tessellation hasn't been developed as a software technology yet. Try a search for HOS techniques.

4) We hit on this in a recent discussion and nAo pointed me to a paper on rendering a silhouette map for polygon models. It was quite old and I'm a bit surprised the technology hasn't made an appearance anywhere, but you can be sure developers keep an eye on these sorts of papers and play around, so I'm sure there are legitimate reasons for it's absence.

You then have things like Voxels, which all have their own issues. Real Life has infinite storage capacity to store 3D models at perfect resolution. Computers don't have that luxury, and storing objects by volume is incredibly expensive - one problem that triangle meshes has got around.

So if you're interested, as Laa-Yosh says go do some searching. But don't think developers are ignoring some obvious out-of-the-box solutions or haven't done lots of brainstorming to try and do something different. There are incredibly clever folk out there working on 3D, and have done for decades, both for realtime and offline rendering where speed is always sort along with flexibility in the pursuit of attaining realism. The maths they come up with for shortcuts to approximating GI and other things boggles the mind. The odds of them missing a trick are vanishingly small. It may happen, but in my following of 3D rendering, I've not yet come across an idea that hasn't been explored a good 10 years earlier than anyone would expect, and instead have been bowled over by the ingenuity of developers in speeding things up.

I dont think you are following me. Then again I am not a tech specialist to express thing sbetter

edit: lets assume computer technology and techniques evolved differently. What route should have things taken?

 

[Shifty Geezer]

The only truly viable alternative is ray traced HOS, but I don't know that the technology would be comparably if that route had been pursued instead of rasterizing. Again though, see the RT thread for discussion on that.

 

[Neb]

Didn't the original Comanche (novalogic) use voxels for terrain and cloud rendering? I vaguely remember playing it on a 386 with a pVGA card.

I am shure Comanche 3 did use voxels for atleast terrain.

 

[Laa-Yosh]

First Comanche was voxel terrain and sprites for helicopters and tanks. It actually required a 486 to run smooth at full detail, but it did look amazing at that time. We didn't even had textured polygons for terrain until Strike Commander, which had low res textures, with no perspective correction, and only on some surfaces. And it was sloooooow...

Outcast was very nice too, but it would look butt ugly in any higher resolution. The amount of memory required for good looking voxel terrain at HD resolutions is just far too much. Voxels are pretty much a dead end for interactive 3D worlds IMHO.

HOS isn't that hard in every possible implementation though. NURBS is a bit problematic, but subdivision surfaces have pretty much become the industry standard and most game studios already have to use them to build highres assets for normal mapping. Combined with displacement mapping, we're going to have all the tools from the movie VFX box. However, current hardware isn't really suited to handle subdivs either - but I think it'd be easier to optimize for this, than to rework and reinvent everything to support voxels, CSG or other abandoned concepts.

 

[Squeak]

Something I always thought would be a fun experiment, is doing analog interactive 3D graphics, as in making a game with the tech.

I don't think people fully realize how far hybrid computers (analog hardware with a digital computer handing it tasks and running the overall program) had come, before digital computers became ubiquitous in the late 70s early 80s.
A few examples:
Some of the last (and most expensive) vector scanning displays, used analog hardware for drawing 2D primitives such as lines, circles and squares with far greater speed than any digital hardware at the time (and probably also of today).

The guys who did the analog Scanimate effects generator, that was used heavily in the 70s early 80s did a very advanced animation system before that called the Animac. It had animation bones, that was raster scanned individually while applying a transformed texture (feed from a live cam) to the individual bones.

I also remember reading an article in a book called Pertinent Topics in Computergraphics (I think), where some guy had done a real wireframe 3d modelling system for proteins, in analog hardware. It had a digital computer feed it the spatial coordinates via a modem connection and then transformation was taken care of by the analog hardware.

Also: http://scitation.aip.org/getabs/ser...00018000009000819000001&idtype=cvips&gifs=yes

With the advent of analog FPGAs (FPAA) the main hurdle with analog computers, expense of hardware and lack of flexibility, is more or less history.

For a very simple example look at this: http://www.vaxman.de/analog_computing/spiral/spiral.html

 

[Shifty Geezer]

I presume the context of the thread is what can software developers do differently with the console hardwares they have to work with. A Resistance:Fall of Man type alternate history hasn't really got any practical applications.

 

[Simon F]

We know that developers have been using more or less standard methods in producing visuals.

Now I ve been thinking, "how about developing other methods and techniques to produce high detail surfaces or surfaces in general?"

There has been some interest of late in point-based systems (e.g. see recent SIGGRAPH conferences). You might want to look at those.

 

[Squeak]

I presume the context of the thread is what can software developers do differently with the console hardwares they have to work with. A Resistance:Fall of Man type alternate history hasn't really got any practical applications.

You could use CELL to render some of the scene. It can do just about anything you can come up with (although to varying degrees)
But that would require an uncharacteristic willingness of developers and publishers to use a bit of resources on something that isn't strictly necessary.

 

[Shifty Geezer]

That's what I'm hoping to see from homebrew Cell development - some experimental rendering techniques. It just seems so quite on that front sadly. And given the jib my mate's had trying to get Linux up and running okay, I'm not too surprised! Entry into Cell development should be kept as easy as possible to encourage more people to give it a go.

 

[Frank]

What you really want is bound volumes, that are elastic and deform. Any material can be described that way. And it's about the only way to make things really realistic.

But it would require most of the computing power to go into calculating those volumes, with the visible outside more like an afterthought, or the result of the process.

And, of course, shadows are stupid, from that POV. Things should be lighted, not shaded. Which also could be done when you see light as a volume, instead of a material property or rays.