LL: Mesh is useless?

[Drongle McMahon]

Just a note - If the slm mechanism is giving you unexpected results, you can disable it by setting MeshImportUseSLM to FALSE (In the Advanced->Show Debug Settings dialog). It stays off until you reset it. I have it off because I want to be precisely in control of all uploads.

I would expect a bedpost like that to have a LI of 1. I'll have to have a go at one.

 

[Pamela Galli]

---

Drongle McMahon wrote:

Just a note - If the slm mechanism is giving you unexpected results, you can disable it by setting MeshImportUseSLM to FALSE (In the Advanced->Show Debug Settings dialog). It stays off until you reset it. I have it off because I want to be precisely in control of all uploads.

I would expect a bedpost like that to have a LI of 1. I'll have to have a go at one.

 

---

Thanks -- I don't think the slm is creating the problem I am seeing because I have been using my models (on the bedpost and everything else I have made) as the top 3 LODs (and getting variable but generally low LI's except on the bedpost), and setting the bottom to 25 (if it was set above 25).  However it is good to know that the slm functions that way.

[Drongle McMahon]

Here we are. Based on Chosen'e. This uses the same mesh for the top three LODs and the same simple box for lowest LOD and physics. I know it's a bit crude, but there's 0.2 left for polishing. It's 504 tris and 430 verts in the uploader (it has quite a lot of sharp edge loops).

[Pamela Galli]

---

Drongle McMahon wrote:

Here we are. Based on Chosen'e. This uses the same mesh for the top three LODs and the same simple box for lowest LOD and physics. I know it's a bit crude, but there's 0.2 left for polishing. It's 504 tris and 430 verts in the uploader (it has quite a lot of sharp edge loops).

---

Thanks -- that's the kind of number I was expecting for my post.  I will do as you have and add a box for the lowest LOD and Physics and see if I get something more like this.  If I do, I will still be in the dark why the results for similar meshes with the same upload parameters should be so different. I look at the chair, I look at the bedpost, and the only variables I see is the size of the post and the version of the viewer.

[Drongle McMahon]

"...the only variables I see is the size of the post and the version of the viewer."

The size is crucial, both directly in determining the LI, but also determining which LODs are important for reducing the LI. . That's why I put that bit of the edit box in the picture. It wa 2m, which seemed good for a bedpost. I did post a thread about the effects of size on download cost part of PE, shortly after release. Possibly too mathematically oriented for many users, but... As far as I know, none of the relevant parameters has changed since then, except that it became LI instead of PE. The failure of the SLM mechanism to keep the scale parameter (I think it keeps the scaled mesh instead) was one of the reasons I turned it off. I don't know if it still does that.

[Siddean Munro]

Pamela, I have seen that upload screen before, once on aditi, where it generated a hairstyle I was playing with to have 12000 tris in each LOD level instead of less tris for the lower LODs as it should have, as a result the LI and cost were astronomical. It could be a bug that you're looking at, as I have not seen the same behaviour on Agni..

 

In any case, I would also recommend making your own LOD and physics shapes. Also, don't build the bed like sculpties and assemble it inworld (I don't think you're doing this, but just in case!) build it entirely first in your modeling package, complete with LODs and physics shape (which should have approx 9-11 quads depending on if it has a headboard), material groups and UVs, and you should find the LI much improved.

[Medhue Simoni]

As Drongle said, size is crucial here. If what you are making is a 4 poster bed, meaning large, then this could be your whole problem. When uploading, you have to treat larger objects a bit differently. Larger objects change LODs at different distances than smaller meshes. I have not done many large items, but the few that I did make, i never, ever saw the lowest LOD. This means that you can drop the lowest to a very small number which will save you tons of prims.

Many "modeller" are going to tell you that making your own LODs is the only proper way to create mesh, or the only way to get a low land impact. This is completely false, tho you will always get different numbers with LL's reducer. Depending on the model, an automated reducer, such as the 1 in the upload window, can do basically the same thing you would do anyways most of the time. Granted, the latest features in the uploader are not half as good as they were a few months ago. "Modellers" have major problems with reducers because they overly reduce their final finished models, which leads to the reduced having nothing left to reduce on sections of the model, hence completely removing that part of the model. If you make all the parts of your model in a uniform way, then an automatic reducer has an easier time reducing the model.

Me, generally, for smaller objects, I make the medium model the same as the high. I have the uploader create the low, at about 1/3rd to 1/4 of the high. The lowest, I try to have it look like the model, but usually prim cost and other factors play into my decision here more. Weirdly shaped things with different combinations of meshes can require me to make LODs, but these are few and far between. For Larger meshes, I reduce the medium slightly, most of the time with the uploader. The low gets something that is recognizable, and the lowest gets the lowest number it will take. Personally tho, I've made a bunch of different types of things, now I'm working on an avatar, and It seems to me that just about every object presents unique decisions.

Good luck Pam, I know mesh is not quite as straight forward as it really should be.

 

[Pamela Galli]

---

Drongle McMahon wrote:

"...the only variables I see is the size of the post and the version of the viewer."

The size is crucial, both directly in determining the LI, but also determining which LODs are important for reducing the LI. . That's why I put that bit of the edit box in the picture. It wa 2m, which seemed good for a bedpost. I did post

a thread

about the effects of size on download cost part of PE, shortly after release. Possibly too mathematically oriented for many users, but... As far as I know, none of the relevant parameters has changed since then, except that it became LI instead of PE. The failure of the SLM mechanism to keep the scale parameter (I think it keeps the scaled mesh instead) was one of the reasons I turned it off. I don't know if it still does that.

---

Thanks -- okay I am going with the theory that the size of the bedpost is making that much difference in PE. I had not noticed that much of an effect before.  And I will try making models for the lower LODs to compensate, instead of reusing the highest model -- ie treating a larger mesh like a different animal.

[Pamela Galli]

---

Siddean Munro wrote:

Pamela, I have seen that upload screen before, once on aditi, where it generated a hairstyle I was playing with to have 12000 tris in each LOD level instead of less tris for the lower LODs as it should have, as a result the LI and cost were astronomical. It could be a bug that you're looking at, as I have not seen the same behaviour on Agni..

 

In any case, I would also recommend making your own LOD and physics shapes. Also, don't build the bed like sculpties and assemble it inworld (I don't think you're doing this, but just in case!) build it entirely first in your modeling package, complete with LODs and physics shape (which should have approx 9-11 quads depending on if it has a headboard), material groups and UVs, and you should find the LI much improved.

---

Thanks -- My plan was to make components and assemble them inworld, which would have been nice, but I discovered that my 5 prim chair went to 9 when I dropped a script in, BUT when I joined just the legs and shadow, it went back to 5. So yes, now I have to join the meshes in Blender, but they have to be very specific dimensions and I have to learn how to do that, and I suspect that will involve: math. (For some reason my history and English background never come in handy in making 3D objects!)

[Pamela Galli]

---

Medhue Simoni wrote:

As Drongle said, size is crucial here. If what you are making is a 4 poster bed, meaning large, then this could be your whole problem. When uploading, you have to treat larger objects a bit differently. Larger objects change LODs at different distances than smaller meshes. I have not done many large items, but the few that I did make, i never, ever saw the lowest LOD. This means that you can drop the lowest to a very small number which will save you tons of prims.

Many "modeller" are going to tell you that making your own LODs is the only proper way to create mesh, or the only way to get a low land impact. This is completely false, tho you will always get different numbers with LL's reducer. Depending on the model, an automated reducer, such as the 1 in the upload window, can do basically the same thing you would do anyways most of the time. Granted, the latest features in the uploader are not half as good as they were a few months ago. "Modellers" have major problems with reducers because they overly reduce their final finished models, which leads to the reduced having nothing left to reduce on sections of the model, hence completely removing that part of the model. If you make all the parts of your model in a uniform way, then an automatic reducer has an easier time reducing the model.

Me, generally, for smaller objects, I make the medium model the same as the high. I have the uploader create the low, at about 1/3rd to 1/4 of the high. The lowest, I try to have it look like the model, but usually prim cost and other factors play into my decision here more. Weirdly shaped things with different combinations of meshes can require me to make LODs, but these are few and far between. For Larger meshes, I reduce the medium slightly, most of the time with the uploader. The low gets something that is recognizable, and the lowest gets the lowest number it will take. Personally tho, I've made a bunch of different types of things, now I'm working on an avatar, and It seems to me that just about every object presents unique decisions.

Good luck Pam, I know mesh is not quite as straight forward as it really should be.

 

---

Thanks Medue :-)  Until this bedpost came along I was doing fine not making LOD models at all but using the high LOD model for all but the last LOD; I guess with the size of the bedpost I will have to make at least one low LOD model, and hope that solves the problem/mystery. 

[Medhue Simoni]

If you want to play with another automatic reducer, look for MeshLab. I think it was made by Stanford, or some college and does a damn good job. There are lots of tools you can use to only reduce some parts of the model and leave other parts alone to hold it's detail.

[Pamela Galli]

---

Medhue Simoni wrote:

If you want to play with another automatic reducer, look for MeshLab. I think it was made by Stanford, or some college and does a damn good job. There are lots of tools you can use to only reduce some parts of the model and leave other parts alone to hold it's detail.

---

Oh thanks -- I actually have that but had forgotten about it.

[Gaia Clary]

---

Medhue Simoni wrote:

Many "modeller" are going to tell you that making your own LODs is the only proper way to create mesh, or the only way to get a low land impact. This is completely false, tho you will always get different numbers with LL's reducer. Depending on the model, an automated reducer, such as the 1 in the upload window, can do basically the same thing you would do anyways most of the time.

---

I think an automatic reducer always has the problem that it does not know which parts of the model are critical to the definition of the object shape. I believe that an automatic reducer can do a good job when

1. The model is high poly
2. The model is mostly uniform in density
3. The model has a smooth/organic surface

And it will fail when

 

1. The model has a high variation of face density
2. The model has sharp edges which define the shape
3. The model is already optimized (to some extend)

According to LL themself their reducer is not one of the best. It is a convenience tool which allows you to quickly create something reasonable. But it was not meant to create high quality mesh (thats how i understood it). I looked at a few reducers and meshlab was one of the best i found on the free market. I do not know if there are commercial reducers available which perform better.

What really works well (for me) indeed is:

When my original is a high poly model: I use a reducer to turn the poly count of a mesh down from extreme high to reasonable. then i tweek that result and make it the highest LOD model. After i have tweeked the highest LOD i proceed by reducing further by hand.

But i must say that i rarely work like this. I mostly start with a low poly modelling (box modelling) and use subsurf to get more details. That makes later reduction steps much easier. The only issue i have found so far is that manual reduction often means that i have to do separate unwraps which is tedious and unpleasant. But well... it gives the best (optimized) results for me at the end.

[Duke Parkin]

I think the point here to me is this...

that bedpost could be made with a 1 prim sculpty.

In fact all 4 posts could be made from a 1 prim chopped cylinder scuplty?

 

Or am i missing something?

[Braydon Randt]

Gaia

 

I have found that you can sort of reuse your UV map on lower LOD items by pinning the areas that define its size and location,  like around the edges in the uv editor,  and then asking it to unwrap again  and most of the time it will move the internal vertices of the UV map to compensate for the holes generated by removing ring loops.

 

Bray

[Gaia Clary]

---

Duke Parkin wrote:

I think the point here to me is this...

that bedpost could be made with a 1 prim sculpty.

In fact all 4 posts could be made from a 1 prim chopped cylinder scuplty?

 

Or am i missing something?

---

Well, you can do that. Remind that for sculpties the UV Map is fixed and the fractions need a whole lot of "collapsed" vertices". Hence you will waste a lot of texture space to get the fractions defined. I believe fractional sculpting is a bad option when you have meshes as an alternative.

 

[Duke Parkin]

---

Gaia Clary wrote:

---

Duke Parkin wrote:

I think the point here to me is this...

that bedpost could be made with a 1 prim sculpty.

In fact all 4 posts could be made from a 1 prim chopped cylinder scuplty?

 

Or am i missing something?

---

Well, you can do that. Remind that for sculpties the UV Map is fixed and the fractions need a whole lot of "collapsed" vertices". Hence you will waste a lot of texture space to get the fractions defined. I believe fractional sculpting is a bad option when you have meshes as an alternative.

 

---

Thats true also

however in the case of this bedpost that is likely to be either a metal or wood texture, its easily done as a sculp.

 

I think this forum topic is a good reminder that Mesh doesnt replace sculpts.

A good build can be a combination of both mesh and sculpts, to get the best detail and prim count ratio, dependant of course on what your building.

sorry to keep refering to them as "prim counts" ahah old habits

The suprise of the Land impact/Prim count on newly built Mesh objects is a topic that will rage on for a while yet

Not only have you got to learn how to build, now its important to learn how to actually import your object too!

[Kwakkelde Kwak]

I'm pretty sure a mesh post well built will be less than one prim/PE/LI... not to mention You win a tremendous amount of rendering time....

[Siddean Munro]

A properly made mesh bed need not be as many prim or Land Impact as the same bed in separate sculpties.  You can separate out components into their own material groups so you get better texture resolution and use of texture space, as Gaia mentioned, sculpties waste a lot of texture space because of they way they must be built.  They also take longer to load, as you need to load a sculpt texture, and a image texture for each sculpty.  Mesh models with custom lods degrade much better at distance than sculpties do as well.  Sculpties can be up to 2048 tris per sculpty, whereas a well made mesh bed can use 2048 tris for the entire model.  Not to mention that building in mesh is the 3d industry standard, and sculpties are a clever, SL specific hack.  If you can make sculpties in an external 3d program like Blender, you can make meshes.  

[Pamela Galli]

Siddean, part of my frustration stems from the availablity of a continually updated central repository of important Best Practices information, specifically the information about joining meshes that you mention.  Below is what I found in a UG transcript, and have been using for a guide.  Yet I found with my chair I had to join at least part of the mesh to eliminate the script penalty. I am going to have to go back and join all the meshes in everything I have made so far that needs a script.

This is all confusing and frustrating -- and I think if LL wants to promote widespread adoption of mesh, it needs to be aware of this.

 

	Charlar Linden	#2 Joining submeshes prior to upload results in a higher PE than uploading them as a linkset. This seems to be due to the more differentiated LOD calculation; still it is confusing since you would assume that one joined mesh would result in less PE than several meshes, linked together (= more server load). Is there some documentation (or any additional information) available on this topic? [Domsson Lean]
[12:11]	davep (runitai.linden)	You're probably getting a higher streaming cost when combined because combining them effectively makes every part as large as the whole linked set for LoD calculations, which means the number of visible triangles will be larger over a larger area. See the mesh streaming cost wiki page for more details:
[12:12]	Nyx Linden	if they're a linkset we can LOD each piece individually.
[12:13]	Charlar Linden	Unfortunately I don't see Domsson... so we're moving on...

[arton Rotaru]

If it isn't confusing, frustrating, complicated to some degree, it wouldn't be Second Life. Actually that's what makes it interesting. If it were all straight forward, easy and simple, well, it would be boring after a week. :matte-motes-nerdy: :matte-motes-big-grin-squint:

 

[Drongle McMahon]

Here is a couple of simple rules-of-thumb....

1. Always join part meshes up menough that the results have download weights greater than 0.5. Otherwise the server weights will dominate.

2. If you join up your meshes enough that the parts have download weights greater than or equal to 1.0, then you will never see and increase in LI when you add a script.

In fact it's a bit more complicated. Since the unrounded download weights are added up for the whole linkset, a part with a higher download weight can compensate for some with weights less than 1.

There is a price for joining, in the higher LI from the increased size and more distant LOD switch. There may be a gain from joining in that you can use a simpler combined physics shape, which matters if the physics weight is high.

Finally, the un-coordinated LOD transitions of separate mesh parts can be disconcerting, so joining can provide a benefit there. Conversely, when the separate parts are specifically designed to separate small high detail parts and large low detail parts, the closer LOD switch of the high detail parts can achieve good savings in download weight while LOD behaviour is still acceptable.

[Ashasekayi Ra]

Duke: This isn't the case where a sculpty is needed. A mesh bed post could be made with 1 LI.

As for the Medhui's comment about LODs, most people have been talking about the "quality" of handmade LODs not the lower weights. Of course, you could use lower percentages for the LODs made by the automatic SL uploader and get a size as low as you want. But, they won't look good. Unless the person makes a high LOD with too many triangles in the first place or makes a highly organic form, the decimator algorithm used by the SL uploader does a pretty horrid job visually, especially for the low and lowest LOD.

[Nacy Nightfire]

Bray, you can also bake  the texture of the highest lod and use that as a guide for setting up the lowest LOD versions.  You don't have to just accept what the unwrap does.  If it doesn't line up, just move things around.  And an alternative to using the baked texture as a guide  is to export the UV template of the highest lod uvs, reimport it and use that as a guide to arrange your lower lod versions onto the texture space.

[Pamela Galli]

---

arton Rotaru wrote:

If it isn't confusing, frustrating, complicated to some degree, it wouldn't be Second Life. Actually that's what makes it interesting. If it were all straight forward, easy and simple, well, it would be boring after a week. :matte-motes-nerdy: :matte-motes-big-grin-squint:

 

---

My guess is that there will be a rather small number of people actually making mesh, and a large number buying it. Like with sculpts, except with those, there were not all these upload rules to try to keep straight.