Land impact and Triangulation?

[Mizana Imari]

Hi there. I've been wondering about these two things for a while. I have read many things about land impact, but don't understand it fully.

For Land Impact, say we have a 400 polygon build with a land impact of 10. (This is completely made up for example reasons).

This 400 build is maybe a simple shed like this:

So this shed has a land impact of 10.

So, we make a closed shed without an inside just like this that is still 400 polygons, and it has a land impact of 2.

[Hehe, some of ya'll are probably going '@_@ I wanna make LI 2 stuff...' Remember these figures are entirely made up.]

The shed is 400 poly's either way, but one is less LI than the other. Why? I know it has something to do with the mesh being more complex, but it's still the same polycount, so that's what is confusing.

What is the secret to low Land Impact...?

 

 

Onto my second subject; Triangulation. When uploading to sl, I find SL automatically triangulates quads, which is a problem since I model in quads. Usually I find SL has broken up each quad into 4 triangles or some such, quadrupling my polycount. As you may guess, this isn't good.

Is it better to triangulate your build yourself in your 3D modeling program? Or am I missing something and it doesn't really make any difference?

That's about it, I guess.

[If it helps at all for any of you to know, I use Cinema 4D for my modeling.]

                                  Thank you greatly in advance for your answers!

 

[Kwakkelde Kwak]

---

Mizana Imari wrote:

 

So, we make a closed shed without an inside just like this that is still 400 polygons, and it has a land impact of 2.

---

The closed shed wouldn't have 400 polygons, because you can leave out all the geometry for the outside. That's what can cause lower landimpact. But I think you mean something else.

Maybe the LoD (a model that shows when you are further away) is what you were thinking about. If you only need to see the inside, you are never further away from the object than half its length. So you can leave out all the lower LoD models. That can save a lot of data, causing the lower LI.

---

Usually I find SL has broken up each quad into 4 triangles or some such, quadrupling my polycount. As you may guess, this isn't good.

---

If your quads are broken into 4 triangles something is seriously wrong with your model. Every quad is two triangles by definition. Graphic cards can't render quads, they render triangles. Even in you modelling software you don't have actual quads, you can see for yourself by making a single quad and then pulling one of the vertices away. You'll see two opposing vertices are connected, you just don't see the edge.

You can triangulate before upload to have more control over how this "invisible edge"  is positioned, since it can be two ways for every quad.

[Aquila Kytori]

Hi  

 

edited to delete my post , When i read through first time i thought it was a wind up . My bad, apologies

 

---

[Chosen Few]

---

Mizana Imari wrote:

 

The shed is 400 poly's either way, but one is less LI than the other. Why? I know it has something to do with the mesh being more complex, but it's still the same polycount, so that's what is confusing.

---

Since you're just making up the numbers, it's hard to really speak intelligently.  Before we go any further, let me make sure I'm understanding how you envision the two models.  In the closed version, you're putting all 400 polys into the exterior surface.  In the open version, you're putting less polys into the exterior, and using the remainder to construct the interior.  Is that right?

Until you say otherwise, I'm going to assume that's what you meant.

 

You seem to be missing some key principles in your understanding of how land impact is calculated.  It's not just about poly count alone. The physics shape, the various LOD's, the UV layout, the number of materials, the normals, all play active roles.  The size of the model also matters, but I'll set that aside, since I'm assuming the two buildings in your example are of equal size.

I can think of several ways the closed version could end up having a much lower land impact than the open one.  In no particular order, these are:

• The the two models would have different UV layouts.  If the closed version has more UV points than the open version has, then it's going to have a higher download weight.
• The two models would have different physics shapes.  If the open version has a more complex physical shape than the close version has, then it's going to have a higher physics weight.
• The two models could have different numbers of materials.  If the open version has more materials than the closed version has, then it's going to have a higher download weight, and server weight.
• The two models could have differing amounts of hard and soft normals.  If the open version has more hard normals than the closed version has, its going to have a higher download weight.
• The LOD's are the biggest wild card.  With the exception of physics, all of the above applies individually to each LOD model.  If the factors are higher in each of the LOD's for the open version than for the closed version, then the open version's land impact would likewise be higher.

For more info on how land impact is calculated, you can read the Knowledge Base entry. You can also dig through this forum, as there have been tons and tons of discussions on it.

 

---

Mizana Imari wrote:

What is the secret to low Land Impact...?

---

Two things:

1. Understand all the factors that go into how land impact is calculated.
2. Use as few resouces as possible, for each factor.

 

 

---

Mizana Imari wrote:

 

Onto my second subject; Triangulation. When uploading to sl, I find SL automatically triangulates quads, which is a problem since I model in quads.

---

That's perfectly normal, for any real-time engine.  Quads are just a way of keeping things organized for yourself, while working in your modeling program.  The game engine doesn't know what quads are.  Everything in-game is triangles.

 

 

As Kwak mentioned, it's also ultimately just triangles in your modeling program, as well.  It's just that the modeling program gives you the option not to display certain edges, if you don't want it to.  This is entirely superficial, just to benefit you while you're working.  As far as the computer is concerned, it's all triangles, no matter what. 

For any model made entirely from quads, the actual poly count of the model will always be double the amount of quads.  The triangles are what count. 

Anything with more edges than three is just a group of tris, under the hood.  If it's a quad, it's two tris, and the poly count is two.  If it's a pentagon, it's three tris, and the poly count is three.  If it's a hexagon, it's four tris, and the poly count is four.  And so on, and so on...

 

---

Mizana Imari wrote:

 

Usually I find SL has broken up each quad into 4 triangles or some such, quadrupling my polycount. As you may guess, this isn't good.

---

No, that's really not good.  As I said above, each quad should be two triangles, not four.  If it's four, then you've got stay vertices in your model, which have no place being there.  I don't know enough about C4D to tell you why that might be happening.

 

 

---

Mizana Imari wrote:

 

Is it better to triangulate your build yourself in your 3D modeling program? Or am I missing something and it doesn't really make any difference?

---

Generally speaking, it's best pracice to triangulate before export.  That way, you can control in which direction the quads are divided. Any modeling program worth its salt will let you flip the triangle edges, if you don't like the direction your quads are divided in.  Again, I don't know enough about C4D to tell you how to do it in that program.  In Maya, which is what I use, it's as simple as selecting the quad(s), and clicking Edit Mesh -> Flip Triangle Edge.  I would assume C4D has a similar command.  (If it doesn't, then in a more game-friendly modeling program.)

Most destination platforms, SL included, will triangluate automatically, if you neglect to doit yourself.  This doesn't allow you to control the direction of the spit, though.  Sometimes, things won't look quite as they should.

So, it's just like anything else.  If you want it done right, do it yourself.

 

 

[Maeve Balfour]

Land Impact (LI) is based around the 4 levels of detail (LODs) of a mesh object. These can either be set via automation in the uploader (with generally crappy, random results), or made by the creator themself, which is far better as it gives full control over how the different LODs look, and how the object will appear at various distances.

Drongle McMahon is far better at explaining this than I am (he will probably post in this thread at some point). In essence though, LOD1 is the main mesh object, and LODs 2, 3, 4 swap into view at different distances as the cam pulls away. SIZE of the mesh plays a big part also, the larger the mesh object is, the further each LOD is visible for. Smaller objects have their LODs swap at shorter distances than larger ones.

So the size inworld of a mesh, combined with how efficiently made the LODs are, have a large factor on the final LI. An efficient physics shape for your mesh (how AVs interact with your mesh etc) also have a major bearing. (I will let Drongle explain these in more depth).

Relatively low LI is definitely achievable IF one is prepared to put in the work to attain these efficiencies (LODs and physics hulls).

.....

In regards to triangulation: I prefer to divide each quad into triangles by hand (for more predictable results and inworld rendering results). However with larger, more complex meshes, this is often quite tedious.

Not sure how triangulation works in Cinema4D, but I know from experience that Hexagon's triangulation method converts quads into FOUR triangles - NOT GOOD. As an option, if Cinema4D has the same issue, you could consider importing your mesh OBJ file into Blender and let it triangulate automatically from there (Blender converts quads into TWO triangles).

:matte-motes-smile:

EDITED TO ADD: Chosen Few beat me to the post - we must have both been typing at the same time :matte-motes-wink:

[Chosen Few]

---

Maeve Balfour wrote:

I know from experience that Hexagon's triangulation method converts quads into FOUR triangles - NOT GOOD.

---

Really?  Wow, that's awful.  What the heck was DAZ thinking on that one?

I just Googled it, and apparently, there is a solution.  Someone posted this on the DAZ forums: 

"You can also use “Utilities-> Decimate” and set (in the right hand properties) the level to 1. That will triangulate quads into 2 triangles, whereas using “Triangular tessellation” will split each quad into 4 triangles."

It's totally ridiculous that you should have to do that, of course.  But at least the option does exist, so it's not as bad as it could have been.

Remind me never to try to do any work in Hexagon.

 

---

Maeve Balfour wrote:

if Cinema4D has the same issue, you could consider importing your mesh OBJ file into Blender and let it triangulate automatically from there (Blender converts quads into TWO triangles).

---

That's worth a try, but it's not guaranteed to work.  If the OP's description is accurate, it sounds like C4D is defining "quads" like the figure on the left, instead of like the one on the right, in the following image:

If that's indeed what it's doing, then that bizarre structure would be preserved upon export, which would explain the described behavior in SL.  In that case, Blender (or any other modeling program) would end up doing the same thing with it.

If this theory is correct, then the reason decimation appears to solve the problem in Hexagon is because that command retoplogizes the model, rather than simply making the existing edges visible.  It's a work-around in post, not really a production fix.

To do the same thing in other modeling programs, you'd need to similarly retoplogize.  What a pain.

I cannot think of any good reason, technical or artistic, for defining "quads" in this manner.  If that's really how C4D does it, then I'd have to recommend C4D not be used for modeling for any real-time environment.  When modeling for film, which is what C4D is primarily used for, this stuff doesn't matter quite so much (although it still does matter).  For real-time, it's absolutely crucial.

I have a hard time believing there's not a simple solution, somewhere in the program.  It would just be too ridiculous if there weren't.

[Kwakkelde Kwak]

From what I recollect, the strange triangulation happens on dae export, not on obj. I could be wrong though.

Anyway, if it's the case, the Blender workaround should work.

[Chosen Few]

---

Kwakkelde Kwak wrote:

From what I recollect, the strange triangulation happens on dae export, not on obj. I could be wrong though.

Anyway, if it's the case, the Blender workaround should work.

---

Interesting.  Yeah, if that's true, then using Blender (or any other properly functioning modeling program) as an in-between should work. 

It's just awfully silly that the program would let the exporters have any say in the tesselation.  All an exporter should be able to do is spit out the existing data, in a new format.  It shouldn't have any power to change the data. 

And of course, it begs the question, if you need to bring in another program anyway, why not just do the work in that program, right from the start?  But that's a whole other subject, of course.

 

The more I hear about stuff like this, the more I'm reminded why Maya, Max, Blender, and the like continue to be the standard tools of choice across the industry.  While I do think it's great that there are so many smaller fish out there to help make the digital arts more accessible to the masses, some of the behaviors of these programs are just too crazy for comfort.

[Drongle McMahon]

I'm afraid I have to disappoint Maeve. She has indicated the variety of possible cause of higher land impact, and since it isn'r a real object, we can't even ask the questions that could narrow it down - is the LI download weight or physics weight? - is it all one mesh or several in a linkset? It should certainly be possible to make the open version for an LI of 2, unless you need to have unreasonable detail in the woodwork. To make the interior accessible, you certainly need a simple but accurate physics shape. If that needs to include a colliding pillar, it will have to be made using "Analyze" because the pillar is narrow. It does sound to me as if you are giving an example of cases where it's the physics shape and its weight that causes the problem.

[Maeve Balfour]

Chosen Few: Yah, Hexagon has some pretty odd ways of working at times. I mentioned the program mostly because of my memory of its painful triangulation method - your solution from the DAZ forums is a workaround, although of course not ideal. I think Hexagon was probably designed more for quad-based modeling - specifically for rendering etc.
Your diagram pretty much demonstrates how Hexagon converts quads into four triangles. Pretty useless as a default.

I can't complain, since all those years ago Hexagon was my introduction to 3D modeling, so I guess I have a (biased) soft spot for it despite all of its quirks.

However, I have outgrown its limitations - especially for SL modeling (no native Collada exporter, no rigging capabilities etc etc). I have since switched to Blender, with it having pretty much every capability I will ever need. Like a massive breath of fresh air - makes me wish I had made the change long ago!

:matte-motes-smile: