PDA

View Full Version : T-Splines, have you heard about it?



caesar
11-11-2004, 08:06 AM
I found this site http://cagd.cs.byu.edu/~tspline/index.html, that claims its has a Subd and Nurbs future substitute (but is backward compatible) with .It was published in SIGGRAPH 04.
Cool! Take a look, and what you say?

Karmacop
11-11-2004, 08:40 AM
Look as if it achives the same thing ngon subd surfaces do, only they act as NURBS ... although I thought catmull-clark sds was based on NURBS ... maybe it was another curve.

caesar
11-11-2004, 09:17 AM
This image helps...SDS/nurbs needs to "cut" the entire model (all the way), but with this tspline, cuts are "terminated", so when you add points/geometry, it wont affect other parts of the model.

Mylenium
11-11-2004, 10:11 AM
Well, the methodology has been around for ages with NURBS it just seems it has now been coded into a smart algorithm so the user doesn't need to worry. Since it is based on real NURBS/ SDS, I'm afraid it won't be anything for LW's sub patches.

Mylenium

blabberlicious
11-11-2004, 02:34 PM
Ceasar:

How'd that face deform then?

I can't quite see how those abrupt terminationas would be that useful.

Particualrly for good facial muscle flow.

It not hard to terminte/localise cut with LW - so what's the advantage?

I'm confused.

:-)

Architook
11-11-2004, 03:20 PM
Ceasar:

How'd that face deform then?

I can't quite see how those abrupt terminations would be that useful.


With a NURBS model, if you need extra control to deform a specific part, you basically need to split NURBS patches all the way around the object in two directions. So to get one extra control point to reshape a jutting chin, it's totally possible that you'd have to split 40 patches all the way around. This makes the model bigger and actually harder to deal with since now for example your eyebrow is split when you don't need it to be, making it harder to adjust!

T splines are interesting that you don't need to split entire belts of NURBS, you just need to split around the local area. So changing the chin to add an extra patch split may only need a few extra changes like on the cheek or lip, and that's all, you won't be affecting the back of the head or anything. And you'll end up adding just a few extra patches, not entire belts of 40 or 400.

So that head model shows it well. If that were a NURBS patch, it'd have exactly the same shape but probably 5X as many patches, if not more.

Mylenium
11-12-2004, 03:25 AM
The head displayed is a poor example, but if done properly, that should not interfere with any animation issues. People have often requested to implement local multi-level subdiv, which is almost impossible for any company without having some money for licensing since Pixar and Alias have a lot of patents on this (even though technically it would not be too difficult), but T-Splines in a way are a similar solution. This would be pretty useful for slick and not so polygon heavy geometry.

Mylenium

blabberlicious
11-12-2004, 05:40 AM
The head displayed is a poor example, but if done properly, that should not interfere with any animation issues. People have often requested to implement local multi-level subdiv, which is almost impossible for any company
Mylenium

I totally get what your say, and terminating cuts like that would be great for localising detail - The model does a good job of showing that ;-)

Yog
11-12-2004, 06:35 AM
I'm probably missing something, but how is this different from programs that support N-gon Sub-D's like MAX, XSI, Modo, etc ?

Nemoid
11-12-2004, 09:25 AM
uhm... maybe it produces meshes that deform better desppite ngons?

i see no other reason than this one. ;)

Mylenium
11-12-2004, 10:34 AM
It's all about control... Even with a n-gon capable SDS algorithm you would want to be able to explicitly control the geometric behavior and density. That's important for character stuff to mimic the flow of living tissue such as muscles. N-gons cannot offer that way of control (due to the way the smoothing is calculated) and are avoided whereever possible or moved to areas where no deformation occurs. the downside to this is that you need lots of quads to maintain flow even in areas that don't move.

Mylenium

Nemoid
11-13-2004, 01:11 PM
incidentally, every part of the human body moves. in the head movements are also in parts that u couldn't suspect great movemennts. temporal area for example. top of the head maybe its the main area of it that doesn't move or deform.

so actually i see no reason for these t splines to be better than current sub- d algorithms we have leet's say in maya, xsi or Modo. catmull clark compliant, or doo sabin or other.

Mylenium
11-14-2004, 05:09 AM
I didn't say that this method would be better - it's just another alternative. As already mentioned, there are serious copyright issues to consider with e.g. Catmull-Clark compliant algorithms and T-Splines might be a way to circumvent this. Also this religious beating about n-gons is getting you nowhere. A good modeller/ rigger will always try to avoid them - at some point even the smartest algorithm may choose the wrong tangents for smoothing and that is bad with animation. You don't have this problem with quads.

Mylenium

Nemoid
11-14-2004, 06:36 AM
I agree with u about quads they are the best always. even in Maya manual it's stressed to avoid ngons for organics the more u can. meshes deform better with all quads.

i saw alot of models with ngons too, coming from Maya or max users , though.

reagarding t splines if they are a new algorithm is surely welcome.especially with the copyright issues u talk of.

Even if they are not a huge technology progress, maybe Nt could adopt them for less money, keeping Lw price down and having ngons and more. :)