Prim Bonuses and LODs

[Chic Aeon]

You might want to look at the Kettle Quest Video (I think maybe the third in the series)?  It is VERY OLD but still does a very good job explaining LODs. It was made for folks moving from sculpts to mesh, but still lots of good ideas in their. I still send folks off to watch it.

 

Looking forward to seeing what you come up with.

[arton Rotaru]

---

Chic Aeon wrote:

---

---

LOL. That gave me a laugh. Actually the things that I see with horrible LODs are mostly smallish  home and garden furniture type items -- which really can almost always use the same LOD as the highest setting -- at least if the original model was low poly.

 

---

Yeah, it depends on at what do you look at. As a creator of a couple of mesh car vehicles, I took a look at some other vehicles from time to time. When mine weight in with 45 LI with about 18k triangles in it's high LOD, and still recognizable as a vehicle down to it's lowest LOD, there are others which have a land impact of ~22 but 450k triangles (highly detailed, low land impact). The worst I came across was a sports car with a tri count of 770k. I forget how much land impact it had, but it wasn't much. The tri count was so ridiculous already, that I didn't paid that much attention to its LI. :matte-motes-little-laugh:

Of course, these vehicles turn into triangle salad with default viewer settings from just a couple of meters away.

[Medhue Simoni]

Well, I figured I'd do 2 videos. 1 for the average consumers, and 1 for the creators. I doubt any consumers want to sit thru a half hour video on how to make LODs. So, the first is short and sweet, at least to me.

 

[Codex Alpha]

All I know is I want to make the highest quality, most efficient models for Second Life as possible, from LODs to textures, so LI bonuses or not will not change my approach.

I learned mesh just from the thrill of packing more details into an object than I could do with prims. I haven't rebuilt any of my building projects in mesh yet (still working up to that in mesh).

The one thing I hate is LOD4 (Lowest?). It doesn't matter what I do to make the model, if it's anythingmore than a 1-plane billboard, I am punished astronomically for it. Nothing make crumples down to that, without looking like anything anyway. Perthaps I am missing something using too many materials (for functionality, buttons, effects,etc), or I'm missing some trick on keeping the texture on straight on lower LODs.. Other than the posts on the forums here, there isn't much specifics on how to optimize or even organize your decimation down, so I do my best.

So now I just do LOD1, optimize my LOD2, destroy (but try to maintain silouhette) of LOD3, and let LOD4 be damned. Try crumpling down a multipart piece into 3 triangles so you dont get charged 20Li for having 64 verts....

[Chic Aeon]

Well again some of this has to do with complexity as well as SIZE but I think most folks forget that lowest LOD setting (as unfortunately many disreguard physics all together). I can't actually remember using more than the minimum on that lowest LOD setting.  The third LOD (next to lowest) can certainly be important and I try to keep that holding reasonably well if at all possible.

 

It does very much depend on your model, how you build, and if you possibly have a heavy finger on the subsurf modifier etc (not a personal comment, just in general). If you try and make things as simple as you can and still get the look you want then the LODs are a bit easier to deal with. Sometimes though on items that need to look smooth in order to be believable, you can take a land impact hit in order to keep viewability (sort of what the original post was about).

 

I made some spectacular wine glasses with wine recently and to get them to hold at a resonable distance (thin stems as wine glasses will have) they came in at 3. Since they were going to be worn and 3 oddly enough doesn't (or didn't in this case) add that much to avatar complexity it was really OK. I sure would have liked to have seen them at 1, but I wasn't willing to mess with the design and make them "clunkier" .

 

Every now and then I design something that IS problematic, that maybe we just shouldn't be making in our virtual world. Not too many vertices, but more about THINNESS or CURVES and such. Not everything you want to make is viable.

 

I have a tendency to use a lot of materials as I build in Cycles. I "know" I am doing that but the way I make things it is pretty much built into the equation. I do now have my textures down to what I feel is a minimum (others might not - we all have our opinions). I remember when mesh was very young there was on full perm creator that made like 8 - 1024 textures for a chair. He had a big store and gave away group gifts (hence my knowledge of how he worked) so I thought that was "correct". Now I cram as much information as I can into one texture plane and have learned some tricks that can get things looking still very detailed even with smallish texture areas. I think that is a big help.

 

Sure other folks will have their own thoughts on this.

[Drongle McMahon]

Good start, except that I think an elephant would have a voice at least two octaves lower! :matte-motes-evil:

[Devriv]

During a

following the interview, the designer CubeRepublic made the suggestion that those who make custom LODs should perhaps be rewarded with “better LI” than those who just use the uploader’s LOD options.  Paraphrasing Patch…

“All I’m going to say is that we’re working on that next…I feel we should be rewarding content creators for creating highly optimized content…Right now the system actually penalizes….”

This is another welcome news and will be interesting to see.  Not being very techy however I’m wondering what “highly optimized” in general means in terms of LODs.  We know for instance that where

are concerned, the smaller and fewer the textures, the more optimized the texturing.  But in terms of triangles and actual geometry, isn’t the very problem under discussion here in a sense one of “over optimization”?  If here too less meant better and more optimized, then objects with zeroed lower LODs would be considered the most optimized!    

[arton Rotaru]

Yeah, that's a non-starter without approving each and every piece uploaded by a human being.

[Chic Aeon]

Custome LODs are definitely not my best skill in Blender (I tried again the other day and the uploader did a better job than I did, still   -- or I guess I should say that the uploader and "I" did a better job than me by myself .   Most of my items are pretty optimized to start out with so that might just be a "me" thing.

 

It seems like you can have less than optimal LODs EITHER by making custom LODs or by using the uploader. There is both skill (or lack of it in my case)  and intent (including that nasty single triangle trick that Drongle pointed out and happily that I haven't seen; I must just hang out in cool places *wink*).  And too, each creator will have an idea of what they want to have stick in a shape. 

 

Complex really.

 

Good we are talking about it though .

 

I can't see that it will help much but who knows.

And along that line I have repeatedly (again this morning) found that the uploader on the beta grid and the uploader on the main grid can have quite a difference when computing the DOWNLOAD part of the equation. This has been going on for awhile and maybe they are testing the new system over there.  No clue really but on more complex items (models that are over 1 land impact) the numbers aren't the same when using the exact same settings -- for me anyway. 

 

 

[Chic Aeon]

Great elephant by the way.

So you are saying that the RenderVolumeLODfactor in the Debug settings is not the same as the LOD Factor setting on the Quick Preferences in Firestorm? (presumably the same slider that you should on the Linden view)

 

I didn't know there was a difference!   I will say that even though I have a very hefty computer things seem to work better at 2 than 4 on the Quick Preferences Panel. 

 

Be sure and post the creator video here when you are done. Sure you will .

[Pamela Galli]

I am just glad there is official recognition that too often better optimization results in being punished by having too high LI slapped on something. Especially when talking about the lowest LOD -- with rare exceptions, no matter how careful I am to make and efficient lowest LOD, anything other than a plane explodes the LI. After wasting so much time doing it, I just stopped. Why bother? 

 

[Devriv]

**Hugs Chic**  Chic, There’s nothing wrong with letting the uploader create the lower LODs.  Please don’t think for a moment that I’d permit myself to criticize those who do or question their skills.  That’s why it’s there, to be used!  It was, after all, all I had when I first began building.  It’s wonderful that we’re talking about the subject, instead of ignoring it, and that is the very objective of this thread.  So again I thank you for it :smileyhappy:

The worst one may say about the uploader is that sometimes it tends to do a less than ideal job of making the lower LOD models.  In fact, I’d much rather that builders use the uploader’s LOD options than to completely eliminate the lower LODs.  Objects coming apart or disappearing when we cam out is my issue, regardless of the methods used.  After all, as you very correctly point out, custom LODs do not automatically translate into visually optimal LODs…by any means.

Pamela, I understand, and I think everyone here is sympathetic.  I’m glad the thread was made a “sticky”.  "Hot topic" anyway, hehe

[Chic Aeon]

LOL.  I had no belief that you were trying to pick on the uploader LOD guys.   I do know however that SOME folks think it is the only possible way to make "good" mesh. 

 

I am a bit more lenient with my judgements (wasn't always that way *wink* but I am mellowing in old age).  To me, the bottom line is TESTING.  Not just LODs, pretty much everything in life. I test and test and test and my alt tests and sometimes I even test in Opensim and then the beta grid and THEN the main grid. It is amazing what you can spot if you test enough. Now and then I actually get something on the main grid and have to upload a new texture or a model. I try VERY HARD not to do this. It isn't really the 10 lindens, it is the waste.

 

Now I am not trying to pick on anyone here, just explaining how "I" look at LODs.   If we go back to Medhue's video (which I hope everyone watches) 

  he shows his sneaky way to make the LODs on the guitar and still hold the shape -- by getting rid of all depth (and likely other things too but let's talk about depth). That works fine as long as you are looking from the front. It is certainly better than that silly triangle.  But I have no idea where people are going to put my stuff. 

 

In a dark corner surrounded by other things, outside in the back yard where people can walk all the way around it -- you get the idea. So for "ME" that isn't going to be a viable method. I probably have some things out there that break down into a triangle, but my GOAL is to have them just disappear completely at a fairly long distance. 

 

One of the best things a creator can do -- in my mind  -- is to turn down their LOD setting to 2 or less (depending on their temperment, viewer and who they feel their customer base is) and live with that for a week. It can be a real eye opener. 

 

Something else that I have noticed lately (I don't think it is a change, just me noticing) is that the more versions of SL I have running and other programs as well, the more the LODs break down. I guess this makes since, I just hadn't really noticed that before. After (AFTER) I made high land impact filler buildings to add to my filler building pack, I went back up to my city demo area with ONLY one copy of SL going and try as I might I couldn't make the original 3 prim wonder break down within a 128 long surround. That was GOOD news, but it certainly broke down when I had more applications running on my computer.  So that's something to think about also.

 

Drongle made a glorious visual explaintion of turning a pillow (I think it was a pillow) into the lowest possible LOD with all the steps from high to low. Took me awhile but I found the screenshot:

 



 

The complete thread is here.

 

https://community.secondlife.com/t5/Mesh/Simple-mesh-pillow-13-prims/td-p/2482869/page/2

 

That was a LOT of typing. Tired .  

[Chic Aeon]

Now I know that the uploader is often spoken of badly. But I have an example of why I think it is pretty good. I also think rather than The Lab giving bonus points to folks making their own LODs (which we know can be bad, cheaty or just plan faulty as well as elegant like Drongle's) they should work more on improving the uploader.

 

I don't know if it can be improved. Perhaps not making the long distance so heavily penilized like in Inworldz.   

 

Here is a table that I just made. More complex than many of my things I was wondering how the uploader would do with the LODs.   I was amazed how well it did.

 

Land impact is 1.2 at one meter height. Photos are taken at LOD 2. Difficult of course to see at at distance but you can tell there is something cylindrical there even from far away. Not much futher out it disappears all at once.

 

Honestly I think it did a damn good job!

 

 OOPS forgot the screenshots.

 



[Pamela Galli]

Looks good to me, and is a very pretty table :-)

[Chic Aeon]

Thanks. The final upload of course had the two cylinder shapes. I took them out for baking and then later remembered that LOL.  Anyway I was impressed with the uploader. Frequently It does a very good job (for me) .  

[arton Rotaru]

Well, it's not so much a matter of the importer LODs are good or bad. The point is that you can do much better manually than the importer can in most cases. Meaning, you can go lower in triangle count while keeping more of the shape, than the importer can. You can always get the result the auto LOD will have manually, but not vice versa.

Improving the decimation function of the importer would be nice indeed. I doubt that will happen any time soon though. However pretty much any modeling program has optimisation/decimation functions which might be much more sophisticated than the importers decimation algorithm. mootools Polygon Cruncher is probably the best available yet.

A human can still do much better than this, but certainly not as fast.

[Drongle McMahon]

Interested in your eaxmple. So I made a thing like the base of your table and uploaded with the default LODs generated by the uploader, and with hand-made LODs. Then looked at it with RenderVolumeLODFactor=2. The triangle counts for the auto-LODs were 3576, 894, 220, 110. For my hand made LODs they were 3576, 452, 168, 24. Here they are at the transition points where you can get both flanking LODs at the same size by zooming in and out a bit. The autoLOD is on the left, mine in the middle, and the one on the right is using the high LOD at all LOD steps. The download weights are 3.5, 0.6 and 126 at 1.2m high.



[Pamela Galli]

I doubt anyone would find the auto loaded LODs acceptable at the defaults you show --- what someone would likely do is fiddle with the numbers to see how good a quality they could get without huge cost. If it's not possible, then they might make their own. As chic says, it's not one size fits all. Sometimes one thing works, sometimes another. In any case, the uploader seems to have done an acceptable job in Chic's example.

[Chic Aeon]

Well the difference is easy to explain.

 

As I have stated SO MANY times , the DEFAULTS of the auto uploader are seldom the best. The only time I can remember using the defaults was on a fairly simple house. Most everything else works better if you change the default settings and test. After doing this awhile you get a feel for what the optimum setting will be, but still .....

 

TEST TEST TEST TEST *wink*

 

Here is my screenshot of what I used to upload that mesh and the mesh at a long distance view on my building platform. This is on the main grid. 

I can't actually tell from your tiny photo if you added the table top and bottom (two 32 segment cylinders that are two sided) so if not, that would add more geometry to your example.

 

I definitely agree that using the default uploader settings is a VERY BAD plan. That has always been my position.



[ChinRey]

---

Medhue Simoni wrote:

That's SL tho. It's a place of learning and advancing, with hobbyists and professionals.

---

I think that's a crucial point here. Even before the prim increase, a skilled 3D modeller could easily make mesh with LoD good enough to work with LoD factor 1 and land impact low enough you ran out of space long before you ran out of prims. But most SL content creators are not skilled 3D modellers, they're hobbyists. And that's how it shold be because a streamlined, optimized Second Life designed by porfessionals ... well, it wouldn't be Second Life.

[ChinRey]

---

Devriv wrote:

Not being very techy however I’m wondering what “highly optimized” in general means in terms of LODs.  We know for instance that where

are concerned, the smaller and fewer the textures, the more optimized the texturing.  But in terms of triangles and actual geometry, isn’t the very problem under discussion here in a sense one of “over optimization”?  If here too less meant better and more optimized, then objects with zeroed lower LODs would be considered the most optimized!    

---

There's no definite answer to what is highly optimized neither when it comes to mesh nor textures. In both cases, the more details you add, the more lag you get. The trick is to include exactly the details that are necessary in the textures, in each LoD model and in the physics model. There are lots of good rules and guidelines for that but in the end it's art, not science.

[ChinRey]

---

Chic Aeon wrote:

As I have stated SO MANY times

, the DEFAULTS of the auto uploader are seldom the best.

---

There's more to it than that. As Arton and others already have said, with a little bit of patience even the most inexperienced human can make better LoD models than the best computer algorithm but GLOD, the imp that lives inside the uploader and tries to help us with the LoD models also has one unique and fatal flaw.

Proper simplification algorithms work on vertice level, removing corners, merging triangles to maintain as much of the shape as possible with a simpler geometry. GLOD (the name means Geometric Level of Detail btw) doesn't care about these tiny little vertices, it removes triangles, leaving huge gaps in the model. It's a very different, much cruder and usually far less effective aproach than any other 3D software I know about.

You can get good LoD with it of course, just set the thresholds high enough it keeps all the essential triangles. But then you also get lots and lots of superfluous geometry in the LoD models, adding both to the LI and the lag. I'm not too surprised it did a fair job with your table those. All those small triangles that don't contribute to the overall outline are exactly the ones GLOD would eat first and in this case they also happened to be the ones you could best afford to loose. And of course, the lattice structure helps a lot disguising the holes GLOD leaves. 1.2 download weight is still a bit high for a model like that - it should be 0.6 or 0.7 - but it's low enough to be rounded down to 1 so it doesn't affect the nominal land impact.

[Drongle McMahon]

Indeed, I was not trying to disagree with you. My intention was to show the poor quality of the default LODs, to add to the context of your model. It was too tempting because this is the type of model GLOD does its worst with. However, since you raise the issue of GLOD with selected triangle counts vs. hand-made LODs, I will make the following comparisons.

First, your medium LOD uses the high LOD. This is not unusual in a model of this size, as the effect on download weight will be very small. However, it does mean that, compared with the hand-made LODs, the viewer has to download and display about eight times the number of triangles when the object is displayed at medium LOD. This is a load for the download and rendering system, compared with using the hand-made medium LOD, that might be considered unecessary if you find the hand-made medium LOD's appearance acceptable.

Second, you have used the zero-triangle fechnique at the lowest LOD. The result is unpredictable because it depends on which few triangles GLOD decides to preserve, and that's not even always the same for the same input. If they happen to be  visible, this is unlikely to be satisfactory (for those using low RVLF). This model is a good example of where the minimal triangle technique can be used effectively with a custom lowest LOD mesh. A single polygon with one triangle for each material can be placed so that it's on the ground facing downwards. Then it will never be visible, and the table is guaranteed to disappear completely at the lowest LOD. You can't arrange that with the GLOD generated zero-triangle mesh.

I wasn't trying to make an exact comparison, but anyway, I added top and bottom to the model (mine is 30-sided cylinders). I also added UV mapping, which increases the data size (although not needed if everything is blank textured). These didn't really make much difference. The triangle counts were 3803, 472, 188, 28, and the download weight went up to 1.00.

I think these considerations do convince me that there are advantages to using hand-made LODs instead of using GLOD. However, I would guess it took me two to three times the work than just using GLOD. With many models, the extra work is much more than that. So I would not disagree that it is perfectly reasonable to decide that it's not worth the extra investment in cases where you can achieve acceptable results without. I guess I just prefer messing around in Blender than fiddling with the uploader settings. I also enjoy the challenge of optimisation. So, of course, I am biased towards the hand-made LODs. In othe words, I am still not trying to say you are wrong.

Oh - and a note about the size of the picture: There isn't really any choice if you want to show the model at the transition points as they happen in the viewer. These were original size screen captures from a viewer session at full screen on a 1600 x 900 pixel monitor. So they show exactly what you see in the viewer. There is a point in the zooming system where you get one LOD if you are zooming in, and the next if you are zooming out, while the overall size is the same. That's what I call the transition point. So this kind of picture shows exactly what you see in the viewer at the extreme range of each LOD with whatever RendeVolumeLODFactor setting you are using. It's not really universally exact, however, because it will still be different with different resolution displays. If my monitor had twice the resolution, the pictures would be twice the size. Using RVLF=2 is acompromise. Obviously, the pictures would be larger, and the lower LODs would look worse, if the equivalent points were captured with RVLF=1.125 (default on LL viewer).

[Chic Aeon]

And I seem to love playing with the UV maps and creating a pleasing look. I like the puzzle aspect of that . 

So far as the small photo, I do get that, but my monitor has a very high resolution so even smaller for me hence I couldn't see if you had the "glass" in there.

 

Along with the discussion -- and I don't argue that you and some others can do better than the uploader making LODs -- I will say that I could have chosen to go with a lower land impact than 1.2 BUT I was aiming to "embrace" the bonus prim aspect with longer distance viewing -- the OP theme here.

 

In the best of all possible worlds folks would make lovely UV maps wiht correct ratios that get a large amount of detail in a relatively small space with stellar clarity AND have perfect LODs (or as good as is practical) by practicing a long while and making their own.  But we definitely don't live in a perfect world. A lot of folks creating mesh will not go to the effort to make their own LODs unless it is really needed (and I have on occassion found the need, but hardly ever in my models). A ton of folks let Blender unwrap their model which now and then works fine, but (again for me and what I am making) most often doesn't.

 

If modelers -- especially those just starting out -- will recognize that all parts of a model are important (UVs, LODs, Physics etc.) and remember that we are living in a LOW POLY world, I think we will be just fine. 

 

We don't have to actually make it to perfection. This exercise IS supposed to be FUN .