Benefits of modeling in Tris
I want this to be a collaborative thread and I want to hear what you have found and I hope we will all learn more about this topic.
For the past few months I have been researching about modeling in tri's compared to modeling in quads. The questions I asked myself were: Why model in tri's? Is there benefits? I have come to some interesting answers.
First off, the question why the hell does it matter? all models need to be triangulated before being brought into a game engine, for animation the model is triangulated at render time.
Fun fact: tri's are left alone when triangulation happens. a quad is always triangulated into 2 tris.
knowing these facts we can see that tris can be very useful.
I would like to show a example that you have probably seen before
Both these spheres have the same radius and both share the same silhouette and the most important of all is that the sphere built in tris is 192 polys cheaper.
AND FINISHED.................................................But what about Animatable topology?
For us modelers lower poly counts isnt everything we strive for. WE need models to deform correctly and without problems.
I want to acknowledge Vadim (http://slipgatecentral.livejournal.com/) for his excellent job on during Dominance war. I would also like to use an image of his here because it shows a game character with animatable flow built in tris.
[IMG]http://tuckercool.com/images/loops1 (1).jpg[/IMG]
The method Vadim uses in his modeling videos of this character I believe to be very inspiring and the way I will be modeling in from now on.
here are links to his videos:
Part 1: http://vimeo.com/5738519
Part 2: http://vimeo.com/5734031
I have to go now but I will be continuing to post later on.
For the past few months I have been researching about modeling in tri's compared to modeling in quads. The questions I asked myself were: Why model in tri's? Is there benefits? I have come to some interesting answers.
First off, the question why the hell does it matter? all models need to be triangulated before being brought into a game engine, for animation the model is triangulated at render time.
Fun fact: tri's are left alone when triangulation happens. a quad is always triangulated into 2 tris.
knowing these facts we can see that tris can be very useful.
I would like to show a example that you have probably seen before
Both these spheres have the same radius and both share the same silhouette and the most important of all is that the sphere built in tris is 192 polys cheaper.
AND FINISHED.................................................But what about Animatable topology?
For us modelers lower poly counts isnt everything we strive for. WE need models to deform correctly and without problems.
I want to acknowledge Vadim (http://slipgatecentral.livejournal.com/) for his excellent job on during Dominance war. I would also like to use an image of his here because it shows a game character with animatable flow built in tris.
[IMG]http://tuckercool.com/images/loops1 (1).jpg[/IMG]
The method Vadim uses in his modeling videos of this character I believe to be very inspiring and the way I will be modeling in from now on.
here are links to his videos:
Part 1: http://vimeo.com/5738519
Part 2: http://vimeo.com/5734031
I have to go now but I will be continuing to post later on.
Replies
if you watched Vadim videos you can see he models his low poly by extruding an edge and then collapsing it into a tri and positioning it. triangulating a mesh will not give you the same results as building in tris.
I modeled these chairs, the left chair is a maya triangulated model the middle chair is the chair built in all quads and the right chair was built by building in the tris over the quad mesh. all have the same silhouette
With out trying to hit a lower poly count I was able to shave off 400+ polys out of the mesh using this method.
"Triangulation" is part of optimization, but you use them wisely and as needed. It's not like modeling everything with triangles is magically saving anything, there's just such a thing as modeling effectively.
Theres more to gameart than "OMG too many polygonz!!!"
You need to consider easy rigging, UV splits, all that.
Also the couch example is irrelevant since you removed details aswell... Plus imagine you have some kind of very clean cut you need to make happen where the fabric is stitched. With your tri version you're fucked.
you're trying to compare selecting all verts, right clicking, and hitting connect. with actually optimising the mesh. something entirely different, and is not only faster, but easier to do when the base mesh you're using is the one you had for your high poly too. which means it'll start in quads, and of course will end up almost entirely in triangles.
what you're suggesting, is that you build a sculpt/smoothing ready mesh, in quads.
then doing your high poly alterations/sculpting.
then building an entirely new mesh from scratch, in triangles.
whereas the more usual workflow (i guess, or it is for me anyway);
build quad only base mesh
sculpt
optimise basemesh and adjust for baking. <--- half the work is already done compared to your final step.
pior: No worries, I am discussing modeling in tris not really uv splits, shading split or material split. Optimizing the materials in a game engine will boost the games FPS way more then optimizing a few million polys. The detail for the couch would be added from the zbrushing and as long as both examples have the same UVs the detail will remain.
almighty_gir: Im trying to say that this method is for everyone I believe that with this method you have more freedom to break away from you low polys silhouette
add detail in zbrush ^^
I am doing this research for an independent study class. What I am really after here is to find if it's definately more beneficial to model a char in tris then allow a program to triangulate the mesh for us.
you should optimise the low poly appropriately.
resources/uncomfortable_loveseat.obj
OP: it's all relative to what's going to be seen in the game. If the cushions aren't removable via physics play then they should probably be welded to the frame, saving triangles. If they are removable, they might benefit from more geometry so as to use bendy-bones (benefiting the construction of pillow forts). If the bottom of the couch is never seen, it shouldn't exist in the low, same for the bottoms of the legs (or even the back legs - all depends on how and where the prop needs to be usable).
this made me laugh
As far as Vadims piece goes, it looks fantastic as an end result but some parts are painful, especially on the heavily triangle regions. Dont take examples as a rule and just try things out man.
Huh? So hmm what's the point then.
You mention perdictabillity, if you have the loops for deformation does it really matter whats imbetween them? Isnt that just what Stahlberg does.
This plus the increasing number of tools for quad modelling out today, greatly increase the speed of an artists work flow. In this industry, hitting deadlines is the key to you getting that check from your investor.
That's why they have left tri modelling behind.
I remember a coder trying to explain it to me once before. Tri stripping happens at run time, so speed is really important, if you have a model made 100% out of clean quads and loops it will strip the fastest. the more you break up the rings with tris it gets a tiny bit slower. Though I cannot remember if he said fully triangulated models were okay. But I think the animation team would have flayed me alive if I supplied them with a fully triangulated model.
this is what I am concerned about perna. same silhouette but -20 polys
/runs for cover
but thats only when i can't build stuff straight out of triangles.
I believe that I have gained alot of knowledge about modeling by working in tris. I dont feel that the efforts I put forth here were in vain.
I will most likely continue modeling in quads, but i still would like to know more about the technical side of tris and rendering.
It's like a big triangle made of quads
(anyways, joke aside .... just show us your work man. "Theory" wont get you far, just practice practice practice)
tuckercool.com
Or maybe tucker's just doing some viral ads for these guys:
You can do that same trick and keep it all quads (see C2 & D2), by removing every other edge from your 40 tri cylinder and rotating the top ring of edges 20 degrees then flipping the hidden edges (if you're in Maya you're forced to triangulate it).
OR
Take the top ring of edges in your triangulated example rotate it 20 degrees and remove the diagonal edges (see D2).
The problem is, this process is longer and more involved than just removing every other edge (see D). As others have pointed out it also makes it harder to work with on almost every level. Unwrapping, vert weighting, deformation ect.
Ahh but the silhouette is more round! Not really, not even in your example. It looks to be leaning in your example...
In your 20 tri example the top ring is twisted and misaligned from the bottom this isn't actually going to round it out any better (see B4 & C4) its just a bit of a trick that works in the screenshot when viewing it from a particular angle.The method doesn't hold up that much better than a quad strip when viewed at other angles. It really starts to break down the longer the cylinder is.
Lets say you need to up-res your triangulated example, back to 40 or maybe 80, or take it down to 10? That's a lot more work on a triangulated mesh than it is on a quad mesh.
Another things to keep in mind
- Rigging and animation will probably kick your model back (I know I would) if they get a triangulated ugly mess where quads could have been used (see C & D).
- Other artists might kick the model back to you instead of fixing or tweaking issues because its not worth the time and effort to try and work with it.
- If you freelance and turn in a triangulated nightmare they won't bother explaining why they assume you're a time traveler from 1995 and aren't going to use you again.
If you're not convinced why people abandoned triangle modeling stick with it and do a few projects from beginning to end. Take it from just theory to concrete finished projects. Fully evaluate both and use which is better. A lot of people have and I'll give you 2 guesses as to which they pick.
I can only speak for myself, I started out modeling in tris and still use them from time to time in specific locations but I'm a quad convert.
Let me just try to put it as simply as possible:
1. Use quads to build your initial lowpoly mesh - this means you can use all the awesome tools relating to edge loops, edge rings, everything that saves you time and does not work on triangulated meshes.
2. Triangulate and optimise where necessary as a "finishing step".
Working purely in triangles is slow and unnecessary, all optimisation should be done at the end of modelling the lowpoly, and your couch images are a terrible example and tell me nothing more than you don't know how to effectively model in quads.
If you think those two have the "same silhouette" then you're smoking some seriously heavy substances... like, really, do your eyes work in a different way to everyone else's?
I don't understand this example at all. One example has half the amount of quads as the other, but is triangulated, and they don't really have the same silhouette.
If you didn't triangulate the one on the right, it would have half the amount of quads as the example on the left, and the same silhouette as before it was triangulated.
I'm a complete nub as far as modelling goes, I also don't know why quads are better to model in than triangles (well actually I do have some idea after reading this thread, and even some idea from constructing primitives with code) - but your arguments make no sense, especially this example.
Anyway, slipgatecentral is on this forum, he should read this thread and reply to clear this matter up!
Can people show me tricks they use to optimize their low poly mesh but still hold the silhouette?
In Slipgate's case, I have absolutely no idea why he'd choose to work with triangles. Sure, he's fairly generous with triangles and terminating loops anyway, but there are definitely some areas where it wouldn't have made a difference.
Is there any effect of triangles on animation, aside from it being a bit more difficult to paint the weights? I've been hearing that a lot, but it doesn't make any sense to me.
It's true that if you take 4 triangles connected to each other by their edge, you can get 4 angle changes in space along that thin strip. Whereas, when modeling in quads, its always better to 'planarize' quads so that the two tris making them up stay on the same plane (it looks better in my opinion, but it's also a very important thing to do in maya since the apps gives no control whatsoever on the hidden edge splitting the quads in two - stupid maya!!!).
Still, making a theory off it and calling it a better practice is, hmmm, wrong because the disadvantages are just too much, and OMG ITS LESS POLGONZ is an irrelevant question anyways esp when the difference is so little.
Make more assets and you'll see very quickly what we mean. Seriously, it should be so obvious after just a few days man. Even on the most simple shapes :
http://vimeo.com/7831268
Anyways, less talk more art man.
But i do not remember where...
And if someone want to do that in one click in Maya he can use this command :
global proc SAMA_RegularTwistedCylinder ()
{
string $sl[] = `ls -sl -l`;
polyPoke -ws 1 -tx 0 -ty 0 -tz 0 -ch 1 $sl;
delete $sl;
}
Each sphere is about 340 tris.
When I skin I frequently use the grow selection button to blend and smooth weighting. The bottom left sphere is "quaded" and it behaves predictably when I press the grow button. However, the geosphere and the triangulated sphere...well they're harder to work with.
Now, imagine instead of a few spheres it's a 10 thousand poly head model.
If the entire character model is less than say 1500 total triangles, it doesn't bother me, I can skin that in my sleep.
Of course now polygons are hardware accelerated but so are patches and even NURBS in some cases (PSP and even PS3 i think but i have to check with our guys)
What MoP said is correct.
Perna,
"A quad-base mesh has the predictability and stability of a grid. It has direction. You can tell a piece of code to identify elements relative to the U or V directions of a quad selection. That, in short, is the technical part."
The same stands for triangles i am afraid. Of a Geodesic grid for that matter. People who have worked with triangles in the past (not many are there but still some ole geezers like me are around) can easily identify the same grid on it and that is how we used to work. in fact many of us used to hide the lines in between and initially what polygons was for us was a way to save time from hiding the in between lines for clarity
If you actually use proOptimiser properly in 3dsmax on a very complex model you will see how the triangle theory stands.
Is it easier and faster to work than Polygons? Most probably and as far as my experience goes NO.
After i saw and worked on nendo and mirai, i immediately fell in love with polygons. BUT
The thing is, triangles, apart from that unfortunate example with the ring, can be properly tweaked and tuned on a complex organic model to match what you are trying to do with less geometry and that is a fact. The geodesic model in some cases is more adaptable structurally sound and efficient.
I was showing to one of our new artists some time ago how his model after using proOptimizer reduced to 70% of his original tricount without loosing significant detail AND without loosing any functionality in terms of animation. You are even able to take models down to 60 - 50% in some cases without major loss depending on the forms. You might argue that the modeler did a bad job. Well yes and no
All this if you tune your process properly ofc.
Many times when i manually optimize a model I find myself turning the edges to achieve the same outline with less triangles as with max (and maybe other tools as well) you have that flexibility. (don't take it again as if i am saying max is the best please...)
Still, nothing beats the clean solid polygon modeling workflow. Perhaps because that is what everyone is working on improving the last years and is bustling with tools. But definitely I can't see myself working with editable mesh, but perhaps if it offered all the modeling features that editable poly offers I wouldn't be thinking the same. After all that is how we who were not using NURBS (like softimage and maya users) used to model in the first place.
I think polygons brought the clarity of nurbs and the absolute control of triangles together. And it is a VERY good thing