Topic: NURBS?

[Don'tHate742]

Looking around the internet, I came across something called nurbs. It has the ability to use far less "elements" (elements such as polygons) while rendering a beautiful 3D image.

Less power, yet more beautiful. Seems like Nintendo could use this.

I was wondering if anybody here *cough*KDR*cough* knew anything about this new developemental tool.

EDIT: Here's a link of NURB made renderings.. Make sure you click the other categories on the left.

[BlackNMild2k1]

NURBS are anything but new, and they have been specualated to death since b4 this generation was revealed.  Surprisingly I haven't heard it mentioned as a possibility for this generation till now.

[Don'tHate742]

Well then...

I'm no tech guru, but could cube-mapping coupled with NURBS produce images on par with Xbox360 and PS3 even with less raw power?

[ThePerm]

well  if nintendo were to  reveal thats how  their gettign  so much  powetr out  of their system  if it was  underpowered  then  th competitors  would  just do that  on  hteir systems  and  have even more pwoerful  graphics

[Don'tHate742]

Why would they have to directly state that they are using such a technology? Can't they just give out the specs and show off real-time vids?

[Pale]

Ok, I admit I could be wrong on all of this..but I really don't think I am.

NURBS are to 3D what vector graphics are to 2D.

What I mean is, they are a shapes created using a combination of control points and mathematics.  Because of this they are MORE computationally intensive....

NURBS are often used by 3D animators for more organic shapes.  For example, a soft pillow would be modeled in NURBS and a cardboard box would be modeled in Polys.

Think of it this way.  A game would store a polygon, a triangle for the sake of argument as (stripped a bit for common sense) 3 points in 3D space.  (3 sets of x,y,z coordinates)

This creates very hard edges where adjusting normals can be used to simulate curves.

A NURBS curve is different.  Keeping it 2 dimensional for the sake of understanding...  say you wanted to store a very smooth bevel.  Instead of storing hundreds of points to create this bevel, you instead store 2 points (the start and the end) as well as a mathematical formula that generates the cuve.  This seems like less space and it is, but each time that frame is rendered in real time that formula has to be run, thus bogging down the processor.

[Don'tHate742]

oh

[BlackNMild2k1]

ran across this on my travels through the interweb

Juexvideo.com

"First, remember Nintendo told us every pieces of the puzzle were in our hands
We (vent_d_est an I) have set some elements together it it seems like we understood what the revolution will finally be.
We found a spanish game developer speaking about the revolution and the surrealistic ideas of the gamers due to the lack of official informations.
First, he insists on the fact that the controller will be much more simple than what we all expect. Miyamoto already said that he found the actual pad designs too complex. The revolution pads should be simples and entirely set-able.
According to him, Nintendo will use the revolutionnary technology : Cube-mapping
This technology allows to generate incredibles and overdetailled graphics, more precise than any actualy known graphics. It's no more about a 2d Picture but as In Real Life, looking like real 3d (beware, we don't talk about stereoscopic 3d)"

So it won't be anymore about polygones but "points linked together by collision mails" named NURBS :
'the NURBS allow to show simple geometrical objects as lines, circles, spheres ... and more complex free-form geometrical objects as a car or a bodypart. The number of necessary informations to represent a NURB object is far less than the needed number for an approximative polygonal representation of the same object
Remember Nintendo said the graphics will be somehow different.

Maybe have you seen some movies of the ATI conferences were developpers presented the cube-mapping saying it would revolutionize the audiovisual industry. Nintendo also talked during this conference, saying they were going to explore a new path, a new way to develop.
To explore these possibilities, the architecture of the console has to be completely different of what ps3 and Xbox360 ones are.

-What follows is only speculation -
According to this developer, the GPU Hollywood should be specially made to emulate traditional 3d and make it a 'cube-maped' one. Nintendo said emulated games should be a little improved.
the CPU Broadway would be a single core with 1 Gb Ram

The developpement should then be easier for two reasons :
- a single core is much more simple than a multicore to program(ps3 & 360 use multicores), what John Carmak comfirmed last week
- Programing in cube maping is easier than traditional 3d, as its easy to simply improve gr'aphics with it"

-------------------------------

an answer to this post from a guy working in graphic industry :

"Well, if thats so, let me tell you that i'll take the revolution, i know these things well (nurbs, splines, cube mapping etc) I studied Informatic graphics and i can guaranty you that its a really great technology.
but i'd be really surprised to already see it in a videogame as it is brand new... even 3d animated movies still don't use it much.

But all that was said her is true, it uses less power than polygones, its less complex, more precise, and much more beautyfull.

ex : to draw a circle, a beautifull one ... hox many polygonal elements do you need? 50? 50 000?

and using spline ? ... just 1"

--------------------------------

Examples of NURBS being used

here is a Nurbs made human face

only twelve nurbs elements to render this penguin

-translated thanx to Pingumask, spelling corrected(?) by me    

[Polemistis]

OMG that face is amazing! Nice informative post too

[NinGurl69 *huggles]

That FACE is halfway to becoming DAVID LO PAN.  ROCK ON whatever this nurbs is!

[KDR_11k]

NURBS, Non-Uniformly Rendered Bezier Splines. They were used for movie CG before anyone thought up subD (using a plain polygonal mesh and tesselating/rounding it). AFAIK NURBS are very limited and a nightmare to use. They don't render any faster than polyons because they are simply converted to polies for rendering (probably slower because the conversion is more complex than he catmull-clarke subD algorythm). Black makes it sound like NURBS are some future technology that just hasn't caught on. No, they're past technology that has been largely replaced by superior alternatives.

Higher order surfaces like NURBS are complete idiocy for realtime systems. Their only advantage is that you can scale their detail level with the system power and that's rather useless with consoles. NURBS will still generate the same number of polygons so you could put polygons there right away. Since NURBS have very few splines (isolines or something, dunno the exact name, that tech was before my time) you have a lot less control over the mesh shape. Less control == more polies to define the same shape == waste of processing power.
The only thing that comes close to higher order surfaces in realtime is Ati's truform, which was designed to work around the AGP bottleneck and graphically update older games (provided someone patched the game to use the feature!). Don't get your hopes up, using that for "upscaling" N64 or Cube games wouldn't work, it distorts geometry and the game needs to apply it because the hardware can't tell characters (wwhich you'd apply it to) and world geometry (which you have to leave untouched) apart. Besides, geometry not meant for truform looks like ass when you tesselate it.

There may be a way to render untesselated NURBS but it's probably VERY processing intensive.

Pale: Mostly but with subdivision surfaces you don't need to model the bevel in much detail, either (no more complicated than the NURBS version). Actually you don't have to do any complex stuff for bevels even without any helpers.

According to him, Nintendo will use the revolutionnary technology : Cube-mapping

Whoa, dude, welcome to the year 2000! I think cubemapping was a big new feature of the GeForce graphics card. Or maybe the GeForce 2. It's merely a different kind of environment mapping that allows you to use six textures arranged like a cube instead of one texture that has to cover all angles of the reflection.

Maybe they mean that cube UVmapping which is completely pointless as it won't work with the traditional art pipeline (namely Photoshop) and is meant for machine-generated textures (like normalmaps) only. Any good art asset will include handpainted sections. Besides, since UVunwrapping isn't THAT hard it's not worth it to forfeit Photoshop just to gain autogenerated UVs. Sure, you could render from a UVmapped texture to the multicube map but since the texel densities don't match that'll be one ugly approach.

[Rancid Planet]

Oh man I LOVE NURBS! They are so good! I like how they come in a two sided box so you can get two flavors at once. But sometimes they get stuck in my teeth.

[Pale]

Even assuming for a second that NURBS are a brand new technology they were going to use, how would that make the Rev easy to develope for?  Yeah, you can develope games... you just need to use a completely different geometry system than before.

[Galford]

Is the ghost of Jason running around here somewhere???  
Didn't we have this debate back on the PlanetN2000 forums??

Doesn't the PSP support NURBS in some way?  I'm too tired to check right now...

PS - Why some many questions??

[KDR_11k]

I've heard the PSP can do realtime tesselation (NPatches) to conserve memory but noone seems to use it and the devs say the hardware can't push that many polygons.

[WuTangTurtle]

I don't think any game studio will be doing any NURBS modeling, from what I hear it's just not practical for the gaming systems.  The instructors at my school work at Neversoft and Rainbow Studios and they all say to stay the heck away from modeling with NURBS.

[Pale]

Yeah.  I've taken classes in both animation/film level modeling, and game modeling.  NURBS are used a lot in feature films, but never in games.  At least as of now.

[Rancid Planet]

I think you're all a bunch of nurbs.

[KDR_11k]

Pale: Um, when was that? NURBS are antiquated, these days everyone uses meshsmooth/subdivision/whatever your app calls it. Watch the making-ofs of various animated movies, they'll often say "this thing has that many polygons", a metric not applicable to NURBS. Bay Raitt's demonstration of the Gollum model  from LOTR showed subD, any game using normalmaps will need polygons for generating them (though you could probably use NURBS and convert to poly but it's not worth it). I don't think any of the major packages are still improving their NURBS support, as opposed to their polygonal mesh editing. Besides, I think RARE used NURBS to create the graphics for DKC, at least the wireframes in the making-of video I got with Club Nintendo looked like NURBS. So NURBS were used for games once but no longer are .

[Kazeneko]

Funny, but NURBS (Non-Uniform Rational B-Splines) are in NO WAY antiquated.  NURBS are the de facto standard for the geometric modelling of everything from a mechanical pencil to an entire skyscraper, and are used probably far more often than you may think in the CGI industry.

A quick list of modeling apps that extensively support NURBS:
Maya
SoftImage
Amapi
TrueSpace
Universe
Rhino3D/Rhinoceros
PowerSolids
FreeForm
Paraform
Cinema 4D
form*Z
3D Studio Max

I'm sure plenty of game and movie artists are using NURBS when using those apps and MANY others to create their characters, worlds, vehicles, etc..

-What the model is going to be used for, determines how it will be 'rendered'
-The analytical representation of trimmed NURBS cannot be used directly on current graphics pipelines, therefore every NURBS surface of a complex object must be tesselated before it can be sent to the graphics hardware.
-The chances of Revolution rendering NURBS in realtime, are VERY slim, unless ATI has developed some very interesting technology over the last few years, and have decided to keep it completely secret form everyone.

In conclusion, NURBS can be used to create models to be used in games for any of the existing consoles (though, it'd be pretty damn pointless to create a NURBS model to be used in an Atari 2600 game! ).

[KDR_11k]

I'm sure plenty of game and movie artists are using NURBS when using those apps and MANY others to create their characters, worlds, vehicles, etc..

Can you name a single one? Because I have not heard of a single person using NURBS in the last ten years. NURBS are still supported, yes but it's mostly because there's no point in removing support for a feature. I mean, hell, MAX still includes edit mesh and physique, both are deprecated (mention either at a CG board and people will look at you funny and ask if you have a problematic exporter or something to use those) but required in order to open some older files. Any Pixar movie is built with subD, the CG in LotR was built with subD, the FMV characters in any modern game were built with subD. I have not heard of a single professional who still uses NURBS. And hell, why should you? SubD produces as least as good a result and it's a LOT more controllable. And the tools are gettin better and better.

Okay, so NURBS may still be part of CAD but CAD and CG are vastly different. CAD is used when you want to design something you will produce in real life but not for computer graphics.

Game modeling uses exactly four different applications: 3d Studio MAX (most common), Maya (pretty common), Lightwave (rare and never used alone) and Softimage|XSI (rare). There are isolated cases of Wings 3d and Silo (I think only GWOT uses that...) out there. For movies apps like Mirai (I think that's what it's called, the app Wings 3d is based on) and perhaps Rhino enter the scene. Apparently Blender was used for the pre-vis in Spiderman 2 but I don't think it was part of the final material. But e.g. Pixar uses Maya exclusively. More apps just complicate the pipeline. Oh, before I forget, for painting it's Photoshop and sometimes Deep Paint 3d but the latter is rare.

[pudu]

Alright I'll give my 2 cents on this whole NURBS thing.  I spent a few years in school for 3D modeling/animation and know all about NURBS, polys, and subD surfaces.

The difficulties of modeling and animating with NURBS

As Pale and KDR_11k pointed out, NURBS are complicated surfaces created using mathematical equations to interperate a surface between points/curves.  It's far more complicated then people might think.  In fact, ALL NURBS surfaces are 4-sided.  These surfaces are split up two dirrections, referred to as "U" and "V".  If a modeler wants to add more detail he/she must insert more rows/columns.  By doing this there will be more points on these rows and columbs to move to manipulate the surface.  What must be known is that these surfaces are always trying to be perfectly smooth.

Now, remembering that they are 4-sided objects and are only made of varying amounts of Rows/Columns being interpereted as one, smooth cohesive whole, imagine trying to model complete objects and characters.  Here are some of the big problems I ran into time and time again while modeling with NURBS:

-In order to make part of the same NURBS surface to look angular one must add in many rows or columns next to each other...therefore making the object overly complicated (will use MANY polygons close together to interperet a corner rather then just using a few perpendicular to each other).

-If you want to add detail to a small part of a model surface (say a nose) and you don't have enough points to properly define it you must add a complete row or column to the model to move perhaps only a portion of all the new points that it would make.

-If you want to create and combine different surfaces smoothly it can be a pain in the arse to get them to match up at the edges (sometimes next to impossible).  The only way to flawlessly do this is to not work with multiple surfaces in the first place

-Animation!  If you have a brain in your head and want to animate something that has organic/smooth moving parts don't use NURBS!  Any connecting surfaces will go all wacky and split, the surface will look all rubbery, it will be deathly slow, etc. and so forth (basically it's a terrible option)

Now, even with all these limatations and annoyances, as Kazeneko said, "NURBS are the de facto standard for the geometric modelling of everything from a mechanical...".  I don't know who KDR_11k talks to about it, but I know for a fact that NURBS are still used a lot in the CGI industry.  They still have a purpose.  Apart from all engeniers using it to model anything from cars to skyscrapers, people use them in the Media Arts and Animation industries aswell.  My teacher, who just completed his short, "The Journey", taught me in class where NURBS still have their place.  It's not like people making animated films never need to make a nice and precise model of something like engineers do.  For instance, my teacher made a barbeque  commercial and needed an exact-looking model in it.  By far the best way to do it was using splines to make a NURBS surface.  Also, for models that are to look nice and organic and aren't to be animated (at least not precisely) there is often no reason not to use a nice, smooth NURBS surface.

KDR_11k, if you're experience is mainly with 3dsMax then I know why you hate NURBS...  I hated them in Max too.  When I started 3d modeling I used Rhino (all NURBS) and did some pretty neat stuff (no characters though).  Then went to Max and used almost no NURBS.  BUT, when I starting using Maya I was reintroduced.  Maya can use NURBS far easier then Max and it's more intuitive.  You can also build something or parts of something in NURBS and convert it to Polys and tweak it there.  Actually, there are more options and tools in the NURBS menu pulldown then the SubD one in Maya.

Why NURBS are more power-hungry and don't make sense to use directly in games

As has been said by KDR_11k, "They don't render any faster than polyons because they are simply converted to polies for rendering."  All NURBS are is a way to model the surface.  The secret behind them is the algorithms used to convert their curves into a smooth surface.  The purpose behind NURBS is not that of making efficient models.  They often have higher polygon counts then a subD model because of the way you model with them.  With SubD you can model with more polygons the only the parts that require it and leave less detailed areas with less polygons, then smooth the model.

If they could have NURBS support for the Revolution it would still be rather taxing.  As KDR_11k said, NURBS surfaces are always converted to Polygons anyway.  Why have the CPU have to convert them so the graphics card can understand them when this can be done before the models are placed into the game?  There is absolutely no point it appears.

BUT, KDR_11k got me thinking.  He mentioned that, "Their only advantage is that you can scale their detail level with the system power and that's rather useless with consoles."  How is that useless?  LOD (level of detail) is a big deal in games.  If developers could include NURBS surfaces in environments that could easily control their tesselation in regards to how far away the camera is there may be some use to it.  The quesiton is, would it be better then current tech?  I know developers already use different quality models that they can swap out depending on how far away the camera is.  I guess if they could, the ultimate thing would be to include real-time subD support for transitions.  Or maybe using NURBS LOD tricks for environments and subD for animated objects like characters?  Anyway I'm ranting so I'll stop.

[KDR_11k]

In order to make part of the same NURBS surface to look angular one must add in many rows or columns next to each other.

Depends on the implementation, at least Blender allows you to modify all four tangents of a point (I think they're called "vertex" even with NURBS in some programs) individually so you can create hard edges without the triple line method. Of course you can just use edge hardness with subD and avoid complicating your cage as well.

Personally I have no real experience with MAX, most of my understanding of professional workflows come from discussions with industry veterans on various forums.

Scalable detail can be done with Auto-LODs well enough, the LOD switches don't happen until the character is very small on the screen anyway. Messiah had fluidly scalable LODs but noone used them afterwards. My guess is that recalculating LODs takes more time than just drawing that many triangles. With scaling NURBS I was mostly thinking about upscaling, not downscaling since the latter often involves removing details completely and I don't think that's doable with NURBS. Since you always need a lot of control over highest-LOD meshes you cannot use an upscaling algorythm for generating them from a lower LOD, subD would be useless here.

Terrain often has dynamic tesselation already, I don't know which model they adhere to more closely (subD or NURBS) but they're usually heightmap based and I'm not sure hopw many engines allow modifying vertex positions for terrain beyond what the heightmap generates (i.e. move them out of their grid positions) without having the dynamic LOD terrain converted to plain "BSP" geometry.

BTW, splines seem different from the NURBS you describe, Michael Comet, for example, has a tutorial modeling a face with splines and that clearly doesn't adhere to any "all quad" rules or anything like that. Some people use splines with the lowest tesselation level to sketch out the shapes even for lowpoly models.

[pudu]

Quote
Originally posted by: KDR_11k
In order to make part of the same NURBS surface to look angular one must add in many rows or columns next to each other.

Depends on the implementation, at least Blender allows you to modify all four tangents of a point (I think they're called "vertex" even with NURBS in some programs) individually so you can create hard edges without the triple line method. Of course you can just use edge hardness with subD and avoid complicating your cage as well.

You're absolutely right about that.  I was mainly thinking back to the frustrating days of Rhino because with that program (or the version I used at the time) once you made your surface and began editing it you had to add points to make it look like an edge in the surface.  With other programs like Maya allow you to edit the points more freely and this isn't really that big of a problem.

As far as implimenting LOD with NURBS or SubD I think that if they were to use either that SubD would be the better choice actually.  You mentioned how NURBS wouldn't be able to remove details completely.  They actually could but because they rely on higher tesselations to look as they are intended a lot of the time setting the lowest possible detail doesn't have good results.  With SubD, though, you model it differently, with more detail where it's needed and less where it isn't.  The model you use to smooth and get the final product would be the one that is used at the lowest detail when LOD is in effect and it would look just fine.  The problem is (which I believe you were reffering to in you last post) is the cost on the processor of smoothing and unsmoothing these models on the fly when loading the different poly models in memory would be far easier on the CPU.  

Quote
Originally posted by: KDR_11k
In order to make part of the same NURBS surface to look angular one must add in many rows or columns next to each other.

BTW, splines seem different from the NURBS you describe, Michael Comet, for example, has a tutorial modeling a face with splines and that clearly doesn't adhere to any "all quad" rules or anything like that. Some people use splines with the lowest tesselation level to sketch out the shapes even for lowpoly models.

Actually splines can be used to model SubD, NURBS, and polygonal surfaces.  They are simply lines you can draw, adjust, connect, etc. that, when coupled with tools such as extrude and loft, will create your desired surface.  That tutorial you're talking about may using SubD surfaces or NURBS merged together or something.  Most programs will make a surface out of 3 curves (3 sides) but I can assure you that the surface is actually 4-sided in the fact that all of the cross sections of either "U" or "V" meet at a point on one of the corners.

I've also used splines with the lower tesselation levels to sketch out polygonal objects.  In Max the surface tool can be used even for low poly characters.

[KDR_11k]

get the final product would be the one that is used at the lowest detail

Lowest LODs are maybe 100 triangles. Even for next-gen games. Hipoly cages are at least a couple ten thousand. Loads more if you're going for Epic's detail levels. At ingame resolution, speed and distance you can't tell the lowpoly cage from the subdivided version except for the lighting and the lighting is handled by the normalmaps, anyway. And besides, these days everybody is using ZBrush, I don't think you can get an unsubdivided mesh after the ZBrush pass. The best method for games is normalmapping. At 25k per character I don't think the difference will be visible.

The demo models for the UE3 have a couple million polies in the hipoly version and 10k in the lowpoly one, apparently they were part of Gears of War. Even next gen hardware can't handle millions of polies per character but with normalmaps it looks good enough.