Any ideas on the LI on this?

[Pamela Galli]

Following up on my thread featuring my deceptively simple looking 8,000 LI arched wall is my "Faberge" egg.  My friend (who likes to remain nameless or I would give her credit) modeled this egg and ornate stand for me to texture.  

After texturingm I decimated the legs to about .3 or .4, so they kept some of their definition.

 

Joined the parts and imported so it would have been .25 m (tho I intended to shrink it more).  If I had imported it at that size it would have been 134 LI.

<sigh>  Any ideas?  I suppose this must have to do with tiny triangles?

[Drongle McMahon]

You didn't say what you did for LODs and physics, whether the LI is download or physics weight. Triangle and vertex counts are also useful. Without these, it is guesswork.

My simplest guess would have to be assuming auto-LOD and no specified physics mesh. In that case, the physics would be from the low LOD mesh. With something like this, with so much curved detail, the default convex hull can still have very many vertices and consequently very high weight. If a physics mesh was specified based on any of the LODs, and Analyzed, it would be even higher unless some very severe simplification was done. In either case, these hull-based physics weight should be independent of size*.

This is certainly the sort of model that should use a simple cube as the physics shape. If you did that already, then we would have to start thinking about triangle counts in the LODs. The geometry of the stand is very irregular, and it's possible that the LOD generator would struggle with that, although the cases where I remember we have seen that effect before were all with toroidal geometry. Were the LOD triangle counts as low as expected? Can you reduce them further?

*A triangle-based weight would be inappropriate, and would revert to the default convex hull at this size anyway.

[Pamela Galli]

Sorry -- the physics weight was like .5, the download 135.  Auto Med LOD, zero Low and Lowest LOD, no physics. 12,000 faces.  I dont see the point of this thing having physics. 

 

 

[Drongle McMahon]

Not physics then. It's certainly hard to understand that sort of LI if the lowest LOD was very simple. What were the triangle and vertex counts for each LOD in the uploader?

[ChinRey]

You don't mention what size you intend to shrink it too, Pamela, and that's crucial for an item like this.

But at a size of 0.25 m along all axises, the theoretical switch points would be 0.9, 3.6 and 7.2 m with LOD setting 1. Even at LOD setting 4, switch points would be as low as 3.6, 14.4 and 28.9 m. Actual switch points would be lower, depending on the client hardware and the surroundings. In other words, somebody standing right next to it with default graphics setting and default camera position would see the low resolution model, not the high or even mid. Shrink it down and the switch points get even closer to it of course. That means you can't zero out the low res model, and possibly not the lowest res either. Sorry.

But that also means you shouldn't get much download weight at all if you zero out those two models. The explanation to the relationship between size and downlaod weight is fairly simple really. The formula includes an estimate of how often each model will be downloaded. In this case, that estimate says that the high and mid res models will hardly eve be downlaoded at all so their triangle and vertice counts shouldn't make much difference, it's all about the low and lowest res models.

What are the triangle counts for those two models when you zero them out? Sometimes the uploader keeps far more triangles than necessary. It's mainly a problem for items you've uploaded before so the uploader starts with data from the pevious uplaod but it can happen even at "fresh" uploads. I never figured out why. I just make a quick custom zero model - on tri for each face, aligned for max. compressibility when I run into that problem.

As for what LI you should expect, it's certainly possible to get it down to 2 or 3, probably even 1, and still keep a decent LOD. That might require a lot of trial and error and tweaking of each LOD model though. If you want to go for it, pack a lunch and dinner, head over to the beta grid and be prepared to spend the day there. More realistically - my best guesstimate is 5-10. Provided the LOD is good enough I can actually see it, I'd certainly be more than happy to have something as beautiful as this in my house even with those LI figures and I'm sure many otehrs would too.

 

Edit:

---

Pamela Galli wrote:

Auto Med LOD, zero Low and Lowest LOD, no physics.

---

Hardly relevant in this case but generally it's a good idea to include a simple phyics model even if you don't actually need one. Using a cube as Drongle suggested gives a physics weight of 0.36, lower than anything the uploader can possibly manage. A single tri or quead as physics model reduces it even further, down to 0.2. The difference isn't very significant perhaps but make a few standard "zero physics" models and make it a habit to use them rather than no physics at all. Over time you'll save a bit of LI with hardly any extra work or effort needed.

Btw, just noticed Drongle had already asked aout the triangle counts. Sorry for repeating his question. :matte-motes-wink:

[Pamela Galli]

Thank you Drongle and Chin.

I re-decimated and removed the shadow plane and tried again at the same size, 41 LI.  The triangle count does not change on the lowest LODs even if I zero them out, even if I use a cube.

Added a cube as Phys shape, still 41.

 

 

 

 

 

 

 

 

 

[Aquila Kytori]

Hi Pam

As you have already noticed its the incorrect Auto generated Low and Lowest LOD tri and vertice count that is causing the high LI cost. They should have been calculated much lower.

I have had similar problems in the past     https://community.secondlife.com/t5/Mesh/LI-count-something-not-right/m-p/2254573#M23880     when the mesh was Smooth shaded but not UV unwrapped. Drongle mentioned it earlier so can you confirm that the eggs legs have been UV unwrapped ?

I'm guessing the problems is in the legs somewhere so  would suggest to try uploading just one leg and checking if the auto generated Vertice and Tri counts for the Med , Low  and Lowest LODS are correct and that you can manually set them to zero or not.

If they are wrong try removing a loop of tris ( to open the leg ) and test uploading again ................Try test upload of a leg before it was decimated.............

Cay Trudeau in message number 12 of the thread link above mentions she had the same problem and solved it by simply removing doubles.

Just noticed that the Physics weight is also wrong. If you have used a cube for physics and Analyzed it in Step 2 it should be only 0.36 and not the  4.560 shown in your image.

 

[ChinRey]

---

Pamela Galli wrote:

The triangle count does not change on the lowest LODs even if I zero them out, even if I use a cube.

---

The cube Drongle suggested was to reduce physics weight, not download weight. He didn't know at that time which weight was causing the problem. It's stilla good idea but it won't actually help in this case.

Have you ever uploaded this mesh here or to the beta grid? Triangle counts for autogenerated models will never go lower than those of the previous model no matter what settings you try the second time around. The solution there is to eitehr delete the slm file or rename the dae file so the uplaoder won't "remember" what values it used before.

Even so, Pamela, I would strongly advice against using auto generated mid and low models for this. I would always advice against autogenerated models of course but in this case it's even more critical than usual. Those are the main models for an item like this, not just "fill ins" for distant view. The high resolution model will only be applied when you cam in on the egg to study it in detail.

[Pamela Galli]

Thanks for letting me know about that bug Aquila. It only went away when I completely separated the parts.

 

Got it down to 3!  I separated the legs, ring, and egg, zeroed out the Low and Lowest LODs and they look okay to me at Lod level 2.5, which is what I normally build for. 

 

[Drongle McMahon]

I guess we just have to put it down as another case of geometry that the Analyze function can't handle. I made similar a model to see if there was somethging reproducibly in that sort of shape that would stop it, but it had no problem. Redundant now that you have it done, but now I've made it, I might as well show it. I got the LI down to 2, but only with a 27 triangle lowest LD which is not to my liking. The three lower LODs of that version on the right of the pairs.The auto-LOD version on the left of the pairs had LI=18. RenderVolLODFactor 2.0, each at max size for each LOD. Triangles <10288, 2570, 642, 296>, <10288, 3512, 1684, 27>. The LIs are for 0.5m high.

[Pamela Galli]

Thanks for going to such trouble to unravel this Drongle -- I hate to think where ppl like me would be trying to understand mesh, without ppl like you.

[Pamela Galli]

---

ChinRey wrote:

---

Pamela Galli wrote:

The triangle count does not change on the lowest LODs even if I zero them out, even if I use a cube.

---

The cube Drongle suggested was to reduce physics weight, not download weight. He didn't know at that time which weight was causing the problem. It's stilla good idea but it won't actually help in this case.

Have you ever uploaded this mesh here or to the beta grid? Triangle counts for autogenerated models will never go lower than those of the previous model no matter what settings you try the second time around. The solution there is to eitehr delete the slm file or rename the dae file so the uplaoder won't "remember" what values it used before.

Even so, Pamela, I would strongly advice against using auto generated mid and low models for this. I would always advice against autogenerated models of course but in this case it's even more critical than usual. Those are the main models for an item like this, not just "fill ins" for distant view. The high resolution model will only be applied when you cam in on the egg to study it in detail.

---

This was the first time to upload this model, btw -- thanks, that is new info there about tri count caps for auto generated models.

WRT auto generated models:  I check and see how they look. If they are good, I go with them. For most things, they are. If not, then I make the LODs. Really the only things I find I need to pretty consistently make LODs for are large architectural features like windows, rows of balustrades,  and columns.