Am I applying too many vertices for LOD?

[Tommy Rampal]

So I'm still a bit confused about setting LODs. It's not fully dependant on distance, right? It also depends on the client's preferences.

So I've been battling with myself on how to set my LODs for my building exteriors (walls, frames, etc). By default on the lowest setting, Secondlife provides auto-triangled LODs that are almost unrecognisable as a building (shape change due to loss of vertices; even a simple triangle for lower vertice meshes.

I tend to find myself setting the triangles to these settings:

• High: 100% triangles
• Medium: 50-100%
• Low: 50%
• Lowest: 25-50%

Typically, my personal rule is: always raise the lowest triangles until the mesh is reasonably identifiable. Is this good practice for not-so complicated meshes?

 

[Mikki Miles]

For my builds I wouldn't use a lod with 25-50% of the high level tri count. It's way to much. You may need that if you set the rendervolumefactor to .5, but who does that?

You can save a lot of prims, no, wait: PE, err, I mean: LI, when you keep the lowest lod as low as possible. Ask yourself, if anyone will ever see the lowest lod.

[Drongle McMahon]

You have a good example of when the auto-LOD is not good. You should be able to make much better lower LOD meshes by hand. Generally, the auto LOD works best for highly detailed and/or rounded models. For others it is better to make them by hand. This may not be trivial as you have to make sure the UV maps are still right.

The effect of the LOD meshes on PE/LI depends on the size of the object. For example, for a flatish squarish object like yours, if it's greater than about 7.6m on its side, the lowest LOD will have no effect at all. For low LOD, the limit is double, and for medium it's 8 times that size. However. that doesn;t mean you should ignore the lower LODs that don't affect PE, because people with long draw distances (>181m) and/or low object detail (renderVolumeLODFactor  < 1) will still see them.

[Chosen Few]

How about working the other way around, from the bottom up, instead of the top down?  In your modeling program, create the model first with the fewst amount of triangles you can possibly get away with while still representing the basic shape and allowing for the basic UV layout you're going to need.  There's your low LOD model.

From there, step it up, to manually create the higher LOD versions.  Add more detailing (while remaning mindful of the UV framework).

As a very simple example, say you're making something cylindrical.  Cylinder prims in SL have 24 quads around the circumference.  But obviously, you don't need 24 to suggest that something is a round shaft, especially if it's not going to be very large on screen.  At the very lowest LOD, it could have as few as four sides, and assuming the endge normals are soft, it will look round enough to get by.  For the next setting up, give it six sides (most small to medium sized cylinders in game art have only six sides anyway).  The next higher one gets eight sides.  Only the very highest level should get anything approaching 24.  Really, 12 is plenty for most cases.

Of course, it takes a bit longer to work this way, creating multiple versions of each model.  But it's the only way to ensure your models look the way you want them, at all LOD levels.

[Kwakkelde Kwak]

You can compare your LOD with sculpted prims, the low LODs are pretty messed up most of the time, because the wrong vertices/faces are surpressed (as has been said this is especially noticable on boxy builds). The number of vertices for the sculptLODs looks reasonable though....

sculpt:

LOD high 32x32x2 = 2048 faces

LOD medium 16x16x2 = 512 faces

LOD low 8x8x2 = 128 faces

LOD lowest 4x4x2 = 32 faces (?)

So this is 100%, 25%, 6,3%, 1,6%

Often I use 100%, 10%, 0%, 0%..depending on what I'm building. (the 10% is still pretty detailed in these cases, the 100% more or less insane))

In most cases it really pays off to make all your LODs yourself. Smooth and "blobby" meshes can often get away with the triangle limit. If you can spare one or two materials it can also pay off to make "old style" walls with partial alpha maps for windows, ofcourse more work, but one wall can then be 2 faces, where a wall with a hole is at least 8 (both per side). I wouldn't bother with the lower LODs for the interior..unless you have really big windows and can see through the building. From the inside you will never be far enough to see the lower ones. I also split most of my LOD building in two, where LOD high is either an upgrade of LOD medium, or LOD medium a downgrade from LOD high and the two lower LODs are far more simple, just an impression of the model with the same overall shape and color/texture. i just read 7.something meters is the limit for LOD lowest, my experience is anything bigger than 5 or so meters doesn't need the lowest LOD, so that sounds about right.

All in all I'd make sure LOD medium looks good, LOD high looks great....and most important of all, LOD low and lowest are as small as possible, these two really make the difference in prim weight.

 

[Sabanco]

Sincerly, this looks really bad to me.

Everytime you build up a mesh in Blender, Maya or e.g.  the first you do is build it in lowpoly, the lowest, the better. Start with a box primitive and then extrude or cut it until you get a cheap look of what you want to achieve. When you do so, ALWAYS check whats the looking of it in Sub-D-View. If it looks exactly like you want it to be, then hit the Sub-D to give your mesh more Polys. Sometimes hitting Sub-D once will do it(esp for SL cuz u dont want to crash or let ur pc smoke in flares). Best is to come to approx. 6 to 20k. 20k is really highpoly for the SL Viewer and also for many PCs of the users. After subdividing your mesh model, you can tweak it to refine the structure or giving more detail bugles or so. Alternatively you can save your lowpoly(thats before you Sub-Ded your model) to the disk for further exporting it as lowest LOD into SL.

Another tip: If you are messing with UVW Unwrapping and are getting bad results, just try Unwrapping your lowpoly mesh before you Sub-D it.

[Kwakkelde Kwak]

If Ctrl-D is a modifier that subdivides every polygon into 2x2 polys..I would advice against it for "box builds". You might get far too many faces where they really aren't needed. A simple wall can be virtually the same in all LODs. For things like balloons or cars or whatever, with very curvy surfaces...that is the way to go I think. For a car I build LOD medium and enhance it with that modifier. My low LOD is far more boxy. Lowest is minimum.

But the advice of building low and enhancing I can completely agree with (In most cases:) ).

EDIT: Also, the number of faces you should end up with is HIGHLY depending on what you want, are you aiming at the lowest prim count? or the best looks? A combination?..And what's the size of the model? and what exactly IS the model? There really is no optimum here, too many variables including very subjective ones.

[Chosen Few]

---

Kwakkelde Kwak wrote:

You can compare your LOD with sculpted prims, the low LODs are pretty messed up most of the time, because the wrong vertices/faces are surpressed (as has been said this is especially noticable on boxy builds). The number of vertices for the sculptLODs looks reasonable though....

sculpt:

LOD high 32x32x2 = 2048 faces

LOD medium 16x16x2 = 512 faces

LOD low 8x8x2 = 128 faces

LOD lowest 4x4x2 = 32 faces (?)

So this is 100%, 25%, 6,3%, 1,6%

Often I use 100%, 10%, 0%, 0%..depending on what I'm building. (the 10% is still pretty detailed in these cases, the 100% more or less insane))

In most cases it really pays off to make all your LODs yourself. Smooth and "blobby" meshes can often get away with the triangle limit. If you can spare one or two materials it can also pay off to make "old style" walls with partial alpha maps for windows, ofcourse more work, but one wall can then be 2 faces, where a wall with a hole is at least 8 (both per side). I wouldn't bother with the lower LODs for the interior..unless you have really big windows and can see through the building. From the inside you will never be far enough to see the lower ones. I also split most of my LOD building in two, where LOD high is either an upgrade of LOD medium, or LOD medium a downgrade from LOD high and the two lower LODs are far more simple, just an impression of the model with the same overall shape and color/texture. i just read 7.something meters is the limit for LOD lowest, my experience is anything bigger than 5 or so meters doesn't need the lowest LOD, so that sounds about right.

All in all I'd make sure LOD medium looks good, LOD high looks great....and most important of all, LOD low and lowest are as small as possible, these two really make the difference in prim weight.

 

---

It's rather unfortunate that the sculpty was what so many SL users were exposed to as their first experience with the concept of LOD. The restrictive nature of sculpties leads to bass ackwards thinking, distorting the perception of what LOD really is.  It becomes about enforced level of poly count instead of actual level of pertinent DETAIL.  Details themselves become almost irrelevant. 

It also makes doing things the smart way appear overly complicated, which in reality, couldn't be further from the truth.  Had SL users been trained to think more critically about LOD right from the start, nobody would ever dream of letting the system just forcibly chop their models in half like that.  People would just naturally be making intelligent decisions about LOD throughout the entire creative process.

Smart modeling is never simply about percentages or "dumb division".  If I remove half the subdivisions from a simple a plane, so that the poly count drops, but the model's real appearance remains exactly the same, have I lowered the level of detail?  By the sculpty definition, the answer would be yes.  But by any other standard, all I've done is clean up the model by removing wasted polygons that never should have been there in the first place, while keeping the detail the same. 

If you're going to add polygons to a model, there should be a legitimate geometric reason for it.  It shouldn't just be to make up for the fact that some algorithm is going to forcibly cut the poly count in half when the object is viewed from a distance.

 Low LOD means just the base shape (usually rather angular), the way the object would likely look from far away in RL.  High LOD means including those surface features that are too fine to see from far away, so their presence is only justified when the object is viewed from up close.  This means things like beveling of hard edges, increasing of divisions on rounded surfaces for better curvature, etc.  Sometimes the differences are dramatic, sometimes they're subtle.  It depends on the particular model.

In most cases, I'll create the low LOD version of a model first, with the fewest polys I can get away with, to suggest the general shape of the object.  Then I'll add refinements to create the higher levels.  Sometimes, though, it's more practical the other way around.  Again, it depends on the model.

In every case, the important thing is it's the real details that drive the poly count, rather than the poly count driving the details.

[Chosen Few]

---

Sabanco wrote:

Sincerly, this looks really bad to me.

Everytime you build up a mesh in Blender, Maya or e.g.  the first you do is build it in lowpoly, the lowest, the better. Start with a box primitive and then extrude or cut it until you get a cheap look of what you want to achieve. When you do so, ALWAYS check whats the looking of it in Sub-D-View. If it looks exactly like you want it to be, then hit the Sub-D to give your mesh more Polys.

---

I agree with you that it's generally best to start with the lowest poly count possible, and then add to it to create the higher LOD's.  However, I can't that simply subdividing arbitrarily is ever a good idea (except for purposes of texture baking, normal mapping, etc., which is a whole other topic, no relevant to this discussion).

To rephrase the example I gave in my last post, if I subdivide a simple plane from one quad to four, but it still remains just a simle flat plane, I haven't increased the level of detail.  I've just wasted polygons.

Only add geometry where it actualy matters, where it will make a real difference to the model's appearance.  Don't ever just subdivide for the sake of subdividing.

 

---

Sabanco wrote:

Sometimes hitting Sub-D once will do it(esp for SL cuz u dont want to crash or let ur pc smoke in flares). Best is to come to approx. 6 to 20k. 20k is really highpoly for the SL Viewer and also for many PCs of the users.

---

I don't understand why you'd be suggesting max polycounts without qualifying the statement with any reference to the kind of model in question.  You said 20K is high.  That's true if you're talking about a character, or a piece of furniture, or row boat, or anything else that conceivably can have a MUCH lower count.  But what if the model in question is an entire sim-sized castle?  In that case, 20K could be very low.  

It's worth keeping in mind that 20,000 polys is the equivalent of just 185 cubes.  It's obviously not at all uncommon for objects in SL, even small items, let alone castles, to be made from far more than that.

Obviously, I'm not trying to say that 20,000 is actually a good number for the kinds of things people tend to make from a couple hundred prims.  In most cases, you can mesh-model the same thing for just a tiny fraction of that count.  My point is simply that we need to know WHAT we're modeling before we can determine what is or isn't a reasonable poly count for it.

[Kwakkelde Kwak]

@ Chosen

I'm not quite sure what you're getting at....whether it's a common post where quoting me is useless or you are completely missing my point. I never said the percentages should be a guide in any way... and sculpts my first experience of LOD? erm...no not quite...

But if the OPer uses the percentages as seen in the pics...the primcount will probably be staggering compared to look of the higher LODs. So I merely gave an example, several actually. And in all of them the number of faces/verts falls off much quicker than in the OP. I also tried to explain every build requires a different approach. using one model as a base for all 4 LODs is not always a wise choice, sometimes it is... Sometimes it's wise to let SL do the calculating, sometimes it's not. Sometimes you don't need all 4 LODs, sometimes you do...etc....etc....

I think my main point was to watch out with the lower LODs or the prim cost will be higher than most customers/users will accept.

[Chosen Few]

---

Kwakkelde Kwak wrote:

@ Chosen

I'm not quite sure what you're getting at....whether it's a common post where quoting me is useless or you are completely missing my point.

---

 

Your post inspired mine.  What's wrong with quoting you to show the flow of ideas, and to give credit where credit is due?  Not every quote has to be argumentative or indicate disagreement.

 

---

Kwakkelde Kwak wrote:

and sculpts my first experience of LOD? erm...no not quite...

---

 

Did I say YOU?  No, I said "many SL users".  I have no way of knowing whether you in particular fit into that category or not.  I was simply remarking on the state of things.  Again, your post was what made me think of it, so displaying the quote seemed only appropriate.

[Kwakkelde Kwak]

okokokokokokok:)

It was very late last night Chosen....and as I said I wasn't sure what you were getting at...reading it this morning with a cup-o- coffee works a lot better... I think we're on the same page on this whole subject anyway...

[Sabanco]

To Chosen few:

You wrote: 'I can't that simply subdividing arbitrarily is ever a good idea (except for purposes of texture baking, normal mapping, etc., which is a whole other topic, no relevant to this discussion).

To rephrase the example I gave in my last post, if I subdivide a simple plane from one quad to four, but it still remains just a simle flat plane, I haven't increased the level of detail.  I've just wasted polygons.'

 

You can simply do Sub-D. But you are right with your example. Sometimes it does not make sense to subdivide a simple plane. OK. In film you need to subdivide that plane, bute were in SL and the Viewer may render it like Unitiy or other game engines, so it does not matter. The only thing that counts is the Polycount. But the original question wasnt about 'hey guys Im doing a castle, ship, clothing, a model, toothbrush - am I applying too many vertices for LOD?'. His Q was more general. Youre right to quote me for writing some tipps down that were more for doing clothing and models and more expensive things in SL. Everythings ok with that.

But If someones does not know what hes doing and he wants to do polygonal modelling for his first time, I would say go to google hit the button and watch tutorials till you do so. With these Sculpies people could do a lot and there were not so many things they can do wrong with that esp to polycount. But now they have the power to do everything in poly. Everyone can do a wrong mesh with thousands of polys, upload it and bomb SL with that. This is what I care for. I dont want people to get over... over.. oh man I did not get the word for it. I mean I dont want the people to be too much enthusiastic about that new feature. It took me a few years to get to know maya, wire loops etc. etc. pp. and in these years some things went wrong to worse and of course you need to have the experience to know what went wrong and why. Polygonal modelling is not the same as nurbs modelling. All the beginners should know this. When you, like I already wrote, do polygonal modelling, extrude faces, do more edges, your UV map is going to be some crappy crazy lines. And if you forget to stitch some verts together your mesh gets worse. Same with freezing transformations and working in bad units.

You wrote: 'My point is simply that we need to know WHAT we're modeling before we can determine what is or isn't a reasonable poly count for it.'

Thats also right. In my humble opinion that counts a lot to your working techniques. If we are talking about a small castle 20k verts may be too much. It will always be a matter of definition and detail how many vertices you need to distribute what you want to. But guys, not everyone has a powerful pc and 6k+ bandwidth, so try not do increase the polys where its not needed, if you do so you may bomb other habitants or self crash.

You wrote to another guy: 'In most cases, I'll create the low LOD version of a model first, with the fewest polys I can get away with, to suggest the general shape of the object.  Then I'll add refinements to create the higher levels.  Sometimes, though, it's more practical the other way around.  Again, it depends on the model.

In every case, the important thing is it's the real details that drive the poly count, rather than the poly count driving the details.'

This is absolutely right. And for everyone whos not an artisan in polygonal modelling: this is law.

*hands Chosen Few a cup of coffee out*

(Sorry again for my bad english)

[Tommy Rampal]

I also figured out, that if viewer settings have object detail set to the lowest; the mesh constantly fluctatates between low and lowest. Which made me a bit concerned...

[Kwakkelde Kwak]

Yes that's a decision for the creator...I always use my own settings, object detail high and RenderVolume at 2.000

i think most people will have it at this setting or higher, because most sculpts and even normal prims will look awful with lower settings.