|
|
Tris and Quads
About Truespace Archives
These pages are a copy of the official truespace forums prior to their removal somewhere around 2011.
They are retained here for archive purposes only.
Tris and Quads // Archive: Tech Forum
Post by Chester Desmond // Mar 2, 2007, 10:40am
|
Chester Desmond
Total Posts: 323
|
Somewhere along the way of learning 3d modelling I got the idea that tri sided faces are generally to be avoided and I can see why in the case of an object that is going to be subdivided. The idea of modelling in all quads is an absolute golden rule in my mind and it is making things more difficult than they have to be because I obsess over it- believing the model to be structurally flawed if there are tris in it; sometimes it is just a lot easier to allow the odd triangle to creep in. Can anyone tell me the main reason behind this rule? Are there cases where it really doesn't matter that much? Am I applying a guideline too strictly?
I fully realize that there are many paths to a result but I always try to avoid reworking down the road by making things as "sound" as possible right from the start. Any comments or chemical inhibitor suggestions welcome. |
Post by TomG // Mar 2, 2007, 6:17pm
|
TomG
Total Posts: 3397
|
Tris are not so good for SDS as you say - they result in a "tight" part of the mesh, with too many points pulling in too many directions and giving an uneven result.
On the other hand, many render engines - real-time and otherwise - absolutely require tris. That is because a triangullar face is always totally flat and there can be no confusion as to the surface the three points define.
This is the thing about a face of course - you define the points and edges, and then the infinite number of points in between can be calculated by the render engine using mathematical means. If you have any more than three points, and they do not all lie in a flat plane, then you actually have many possible surfaces that combination of points could define. Your mathematical formula doesn't know what to do! Does the surface in between those vertices curve this way, or that?
Many game engines require triangulated models. Many other 3D applications only work with triangulated models and reject so called "n-gons" (faces with four, five, six, etc edges). Many formats only accept tris, so if you export to them from tS you find your model gets triangulated.
In fact, for getting my model for sale through Viewpoint I had to triangulate it manually in many places, to avoid some of the nasty things that automatic triangulation can do.
Personally I find tris very hard to work with, and avoid them - unless I am planning on having the model available for other formats (ie going up for sale), in which case I always make sure I end up with a model made with tris. I may work with quads on the way, but I keep in mind that at the end, its going to have to be triangulated!
In general, over the last decade, based on what most modeling packages do, I would say that tris have been the golden rule. For a long long time, trueSpace was a big exception in being able to work with "n-gons", and an exception I was glad tS had!
Of course if you never plan on leaving tS, then quads are just fine. Interestingly, I believe the new real-time renderer in tS7 works with tris - the model is triangualted before being rendered, no matter how you have built it. That might have been true with the earlier real-time renderer, but I don't think it was, I think it rendered n-gons without triangulating them (hence some weird things could happen if faces had non-coplanar vertices, you saw "tears" and "holes" in the real-time mesh).
So, I would say, work with quads if you like. Do nothing else if you dont plan on leaving tS. Give careful thought to the construction of the mesh, and ensure you control the triangulation, if you plan on going to a game engine or other application or 3D format.
HTH!
Tom |
Post by Jack Edwards // Mar 2, 2007, 7:48pm
|
Jack Edwards
Total Posts: 4062
|
Hi Chester,
I've seen some interesting discussion on this and it seems that the answer is use what produces the best result for you. Which seems like a classic non-answer! :rolleyes:
From what I've been able to glean from the more modern debates on this subject is that the edge loops that describe the object are more important than whether quads or tris are used -- especially for models that are planned to be animated. I've seen some modelers go through contortions to maintain quad geometry but end up creating a mess of horribly non-planar faces (myself included!). Also be aware that tri's are not the only thing that can cause "pinching" in your SDS model. If you have more than 4 edges coming out of a vertex then you have what's called a "pole". That will also cause a pinching or creasing. Normally an undesired trait, but some modelers will create those on purpose to describe a certain feature on the mesh.
What works best is probably going to be different between software packages depending on the method it uses for SDS. So I'd recommend trying both techniques and seeing which one produces better results. Another trick that I've seen recommended is putting your n-gon (defining as 5 or more sided faces) and tris in less obvious places (ear canal, nostril, inseam, crotch, armpit, eye socket, inside of the mouth, back of the head where covered with hair, etc.).
If you're not using SDS, then it doesn't matter whether you use tris or quads as long as your model reads well to the render engine.
-Jack. |
Post by Ambrose // Mar 2, 2007, 9:41pm
|
Ambrose
Total Posts: 261
|
Correct me if I'm wrong ;)
But I thought the golden rule not to use tri's were because of the shadow artifacts they sometimes created, this can offcourse have changed but I doubt it.
What happen were, I think, if you had just a flat square but some tri's along, hard to explain, you could see black shadows when the engine couldn't decide if it were up or down something and therefor setting shadow were it should be the up side instead.
I know Tom can explain this better ;) and maybe I'm thinking of something else...
and Tom when exactly is the release ;););)
Off to pay bills.
SeYa/Ambrose... |
Post by Tiles // Mar 2, 2007, 10:28pm
|
Tiles
Total Posts: 1037
|
Talking about creating meshes for stills or animations.
A high poly mesh needs a proper edgeloop flow so that the topology works. Which is ways easier to model with quads.
It is also much harder to animate a triangulated mesh. The edgeloops can make fun then. Deforming a triangulated mesh can result in strange results. Especially when the exge that divides the quad goes in the wrong direction. And it always does at some points. What should bend comes out with a sharp edge then ... :D
That's a few reasons for the golden quad rule. Well, it may be a golden rule. But there are topology situations where i don't care about avoiding tris. When it's at a topology area that doesn't deform this much i definitely don't mind :)
The golden quad rule is a completely different chapter when it comes to low poly meshes for games though. Most game engines just can handle tris. And so here you have to avoid quads and N-gons ;) |
Post by GraySho // Mar 3, 2007, 2:02am
|
GraySho
Total Posts: 695
|
Non-quads can result in ugly results when using sds. The surface doesn't look smooth and you probably see some shadowing where there shouldn't be. Another thing is the edge loop selection. With quads you'll never have problems. As soon as you have a tri in the way, the software can't decide where the loop continues. The most important thing of course, is always the result. If the result looks good, goal's achieved, with or without tri's ;) |
Post by iBrent // Mar 3, 2007, 2:44pm
|
iBrent
Total Posts: 102
|
So knowing that I want to develop for a game engine that works best with tris, how do I set that up in ts7.11? :D And am I limited in anyway to the types of primitives I can use. Obviously I won't be doing any sds, so keeping that in mind, any other suggestions when working with tris?
The idea is to generate low poly models for a game engine.
Thanks,
iBrent |
Post by Jack Edwards // Mar 3, 2007, 7:19pm
|
Jack Edwards
Total Posts: 4062
|
I agree with GraySho about the end result being the most important thing.
This thread is a bit dated, but discusses the quad vs. tri thing a fair bit with regards to SDS:
http://forums.cgsociety.org/showthread.php?f=25&t=38469
Steven Stahlberg in particular seems to feel that the topology produced is more important than whether quad or tri or n-gon is used and if you compare w/ his results it's hard to argue his point:
http://www.androidblues.com/
@iBrent:
It's a lot easier to work with models that are mainly quads since it's easier to see the topology flow. I would think it would work better to work with quads until you finished the modeling. Then you can triangulate. The beauty with low poly models is that it's not so much work to go and flip edges that are going the wrong way since you don't have as many polys to deal with as you would in a high res mesh.
-Jack. |
Post by nowherebrain // Mar 6, 2007, 4:23pm
|
nowherebrain
Total Posts: 1062
|
it's a mater of the right tool; for the right job, there are some areas of a mesh that handle tris well(almost required)..even on catmul clark divided faces..don't be too rigid on your modeling habits, unless you're developing for a game/RT 3d environment...as it's already been discussed they require triangles. I work with both environments and eventually you get to understand where they are or are not acceptable....there is also the option of using 4 sided tris[not as wierd as it sounds(for quad meshes, btw)].I wouldn't recomend it, unless you really know what you're doing. |
|
