PDA

View Full Version : Freezing Geometry...Why do they do it?



jamball
02-26-2003, 03:31 PM
Freezing Geometry...Why do they do it?

I see that many modeler/animators tend to freeze their geometry at the end of their modeling development pipeline. Why do they choose to do this rather than keeping it as a metanurbs object throughout?

What are the considerations regarding rigging a frozen vs nurbs based object?

What are the considerations for surfacing?

I see that Leigh van der Byl in his surfacing for dummies tutorial says that when UV texturing a model some will prefer to unwrap the model once it is frozen, while others prefer to do it before freezing.

Does everyone keep a nurbs based version of every model so that they can version it for polygon density?

Thanks.

Elmar Moelzer
02-26-2003, 03:49 PM
Hey Jamball!
For e it depends on the object. I usually freeze non- deforming Objects like cars etc. This usually makes shorter render times.
I also tend to freeze at level 0 or 1 during modeling just to get dsome additional smoothing (you can control that well with subpatchweights).
Never call Leigh a "him", she wont like that ;)
I think she meant that by unwrapping after freezing you can easier get rid of the strecthing problems UVs can produce with subpatches.
I usually keep most of my modeling- steps as backups so i can go back and resue parts or make another model out of an earlier modeling- stage.
Hope that helps.
CU
Elmar

riki
02-26-2003, 05:11 PM
Freezing greatly reduces render times. But you would only do this with certain kind of objects, like a vase etc, things that don't move. I wouldn't freeze a character that I intended to animate. There may be some benefit for print work?? Not sure. Some people also freeze to increase the resolution in certain areas. I have some more info on this here. http://www.suture.net/tutorials/modeling/page06.htm

See 'Organic Modeling - Freezing To Add Detail'

jin choung
02-26-2003, 05:45 PM
hiya jamball,

it's a very good question. there are MANY very compelling and important reasons to convert your SDS models to POLYS at the end of the production pipeline in lw.

1) import/export to other APPS! if you are using multiple apps like maya or max along with lw at your production facility, the ONLY way to move models across all 3 is by way of polys. there is no STANDARDIZED IMPLEMENTATION OF SDS like there is for NURBS and so if you're using SDS in any program, you are STUCK IN THAT PROGRAM.

also, if you want to use a 3rd party 3D PAINTING APP, you have no choice but to convert to polys. these paint apps don't support SDS either.


2) LW's UV MAPPING distorts when used with its SDS! there used to be two reasons for this distortion. newtek addressed one, now there is only one reason left - that is, a single straight EDGE in uv space remains a STRAIGHT EDGE, even though in 3d space, it has resolved into many edges forming a CURVE.

recently, a very ingenious WORK AROUND has been devised that will allow you to uv map sds models in lw without distortion. but essentially, it ends up COUNTER DISTORTING your bitmap image.

a significant reason why i think right now, SDS is still not ready for prime - time in lw yet. there are still a strong vestigial remnant of the idea that SDS is still a modeling tool....

3) NO EDGES HARDENING/SOFTENING in lw sds yet. you CAN modify the curvature at VERTICES but this ends up sharpening EVERY EDGE going into that vertex. this cannot for instance be used to create a CREASE.

you can create a hard/soft edge effect by DOUBLING EDGES where you want sharpness but that results in inelegant, waste geometry.

so, if polycount is an issue (and it always is), you have to freeze and then optimize by hand.

or simply sculpt in detail on the frozen, hires mesh.

this is my other reason to feel that SDS in lw is still INCOMPLETE.

4) EXTREME DETAIL - sometimes, you have to add such extreme levels of detail to a model that you end up having to subdivide what is supposed to be your LOW POLY CAGE to an extremely high degree anyway.

you get to a point where you run into diminishing returns by keeping the model SDS. might as well just freeze into polys and then work with the ample detail level of polys to sculpt in more.

also, because of the nature of SDS in general and SDS IN LW in particular, every time you add detail in your model, you do end up propagating polys unnecessarily. it is more controllable than with NURBS but it is not as controllable as polys (because of inability to handle NGONS as well trying to avoid TRIs in your sds model).

4footnote) there is a method that's gained great acclaim recently that will allow you to create a VERY HIGH DETAIL poly model and then extract that detail as a DISPLACEMENT and/or BUMP MAP and then apply that map to your SDS surface. but they really need to fix UV MAPPING on SDS PROPER in order to get this to work in lw.

5) INABILITY TO TRIM, BOOLEAN and BLEND SDS - this is possible with NURBS and still be able to work with the higher order surface. in SDS, it is NOT POSSIBLE. and so, sometimes it is more efficient (sometimes downright necessary) to freeze, then do your boolean or whatnot, and then re-sew what is probably a mess of a seam back in polys.

this is true of all SDS but it is an inherent limitation. but all surfaces have drawbacks - NURBS can do the neat trimming and such but their surfaces are limited to the topology of a single quad sheet.

(however, being able to HARDEN/SOFTEN EDGES would mitigate the need for trimming/blending/booleaning in a great many circumstances)

because of point 5, i highly recommend doing industrial design surfaces and car chassis' and such with NURBS. there are many many people who do really excellent cars with lw sds. but in my opinion, it is just more straightforward to do it with NURBS - the best tool for that particular job. on the other hand, i don't know why anyone would make a character or animal in NURBS as opposed to SDS.

luck

jin

Skonk
02-26-2003, 05:46 PM
When lightwave renders it has to calculate a highres poly/tri mesh of the subpatched objects and this takes a bit of extra time, for stills it doesn't really matter cos the added time isn't much for 1 frame but over a large animation it could add several hours, the benefit though to useing subd's on none deforming objects is that u can have the display level set quite low to help speed up layout while your animating but the render level nice and high for smooth renders.

James..

doc_roswell
02-26-2003, 08:38 PM
Jin:

Do you ever use splines to do things like modelling cars? I realize that they are NOT NURBS, but are somewhat similar. Though putting points on splines isn't exactly as seamless as putting 'em on NURBS curves (as i recall with my limited Maya experience.)

I've also been told that trims, particularly in Maya, are often not used by some shops (as a rule) due to instability in the surface at render time. Have you run into any trouble with that? This was about the time of Maya 3. Might've been fixed by then.

I've also heard that someone holds the patent on edge-sharpening along one axis and that's why even Maya's sharpening tools were semi-crippled. Again, this might be old news, but might explain why LW's are only semi-implemented.

No, i don't use LW's sharpening tools either. Generally i'm forced to make more geometry to get the shapes i want, and it's gettin' kinda old.

-Matt

jin choung
02-26-2003, 10:51 PM
hey doc,

yup, i use splines all the time for industrial design, machine kinda things in lw. BUT note that when you use splines to generate geometry, what you get is POLYS. lw is unlike other apps that support RT, higher order patch surfaces - it directly generates polys... everything seems to end in polys. which is kinda what i was saying above.

nothing is as robust in lw as polys. i don't mind that AT ALL, but if we're gonna say that SDS is no longer just a modeling tool circa 5.0, we should go balls out to make it as robust as polys. either that or just fess up and say that we're the best goshdarn poly only app in the world.

splines are plenty cool and useful, but just like their RT, higher order surface cousins, the inability to trim and project curves and such hampers the workflow. and note i say hamper. nothing is impossible in lw. have a look in our gallery. but again, i think for things like chassis' and ID surfaces, nurbs is the best tool for the job.

btw, probably one of the best things you can do with splines is use them to sketch appropriate LOW POLY cages that you can SDS later....

As for NURBS in a production environment, i don't end up using them a lot at work since i do games. but in the experimentations i've done in maya 4.5, there does seem to be an issue with getting perfectly seamless trim joins... though if you want to use them, a studio could probably afford to just photoshop over the seams.
this problem is exascerbated tremendously when your surface is not just an organic hard body like a modern car chassis but if it's the skin of a whale that's DEFORMING!

but i wouldn't even push for keeping NURBS surfaces as NURBS. they are an excellent modeling tool and the right one for doing ID and car chassis and things. but i would not at all be averse to freezing the nurbs ultimately into a poly surface for welding and render. it's just that getting the complex compound curvatures in smooth ID shapes just work out better and faster when modeling in nurbs.

as for CHARACTER NURBS models, again, i can't imagine why someone would want to do it! BUT THEY DO!!! for a nurbs model, they do indeed try to avoid trims and active blends so they end up modeling TWICE! once to get the form using whatever method you need - trims, blends, etc.
THEN, after you're done, you manually go in and start DICING UP YOUR MODEL into QUAD PATCHES!!! in this way, you can generate complex models without having trims in the final version. UGH! there are lots of books even in borders and barnes and noble that go over this proces but i can't, for the life of me, imagine why someone would do this!
(btw, the DICING mentioned above is essentially the way that you're modeling if you use lw splines (consisting of quad patches [ok, you can have tri patches too]).... only we don't have the option of generating a trimmed model first and go backward)

PIXAR does indeed own the patent on their CATMULL CLARK SDS algorithm. i would imagine that this is the reason why the SDS format is non standardized and as fractured and UN-interoperable as it is.

if you don't want to pay pixar, you've got to roll your own - with varying results.

BUT pixar cannot possibly own the patent on directional edge sharpening - that would be like owning the CONCEPT. so just like SDS itself, if you roll your own, it's fine.

that's why CINEMA4D has really excellent edge sharpening tools - enough to make a man envious!

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

the only app that i know of that uses PIXAR SDS is maya itself. but as you said, there is speculation on whether this is why maya sds only has THREE degrees of edge sharpening - SHARP, SEMI SHARP, SMOOTH - and in my mind, this has to be a legal thing cuz it's a DUMBASS way implementing edge sharpening in my book.

why the heck WOULDN'T you just give 0-1 fp value of sharpening? actually, i do believe that C4D does have this so there's no reason why we shouldn't.

ALTHOUGH - with NURBS curvature, there are only 3 degrees of CP sharpening so i wonder if maya's sds derive the limitation from that....?

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

yes, we need to FINISH SDS in lw!!!

jin

doc_roswell
02-26-2003, 11:16 PM
Right. LW is totally poly-dependent. There are just times that we forget that, since we can render SDS now. And yep, using splines to make the low-poly cage is a great way to start your models. Sometimes i find it easier to work strictly with boxes, though (particulary since previewing spline cages is dicy at best).

I agree totally with your view that SDS are about 90% there in terms of being a fully-operational toolset. Just need to put in a few more features.

Yep, trims on organic models would be a living, screaming nightmare. Makes me wonder how folks use NURBS at all for organic stuff, but then i think my mind has been warped by Modeller. And folks actually work the models twice? Augh! Are they that wedded to that toolset? Yeesh!

I suspect that the problem with edge-weighting functions in SDS is that Pixar doesn't own the concept, but they do own the equations involved. Apparently those algorithms can be copyrighted (see also, Pixar vs. Exluna, Macromedia vs. Adobe, Adobe vs. Everyone). That's probably what's holding things up on that end. It might explain the crippled implementation that Maya's using, too.

Thanks for the info.

-Matt

jamball
02-27-2003, 07:34 AM
Now this is exactly what I had hoped for. :)

I'm astonded by the response.

Now to print this off and go read it.

James

jamball