Proposal - FAST MESH

[Coffee Pancake]

Land Impact (Li) is not a measure of how detailed (or well constructed) a model may be, it's an method of accounting for the purposes of billing that's more related to scale than complexity. Tweaking Li in such a way that half the grid doesn't get randomly autoreturned (or encourages objects to be left for fear of increased Li when re-rezzed) doesn't solve the specific problems caused by the defaco-standard viewer mishandling LOD, and wont on its own address the point in the OP.

Having a consistent LOD system across all viewers matters when people are creating content for all viewers. If that means the debug setting to hack LODs up has to go away too, so be it. (and I'm not suggesting FS should go it alone)  This standard is a cornerstone of the shared experience.

Yes, static objects are not even half the problem when an animated rigged mesh avatar, festooned with multipart rigged attachments enters the scene, but for the sake of examining the subject of mesh objects, ones that aren't prone to wandering off and are easily inspected is easier to wrap everyone's head around. Obviously any blanket change to the uploader would also affect rigged mesh destined for an avatar, wouldn't it be nice if that meant creators of rigged mesh bodies had to ship a full set of manual LOD's in their dev kits, so clothing makers could make their own matched LODs.

The point that bad-content-is-bad needs to be publicly made (and enforced). Capping complexity alone moves the problem to consumers in such a way that encourages people (and viewers) to find a ways around it, it will not have the desired effect on content creators. Changing the uploader and enforcing LODS everywhere will (can you imagine a single mesh body that chose not to release an updated FAST MESH version surviving commercially?)

New hardware doesn't solve this problem, and even if it did that's of no use to SL users unable to even contemplate a high end gaming rig.

We're not making content for systems the users actually have (or aspire to). This hurts us all.

[ChinRey]

6 hours ago, FinnfinnLost said:

The thing is, most best practices talked about here are not difficult for an amateur to learn.

Building with polylist meshes will never be as easy as building with prims of course but if you know how to make mesh at all, it's not too difficult to make LoD models that are solid enough to stand up at LoD factor 1 and still low enough in land impact you'll run out of space long before you run out of prims on your Homestead.

---

That's another reason why I don't like this LoD factor thingy btw. With the generous prim quota the Gods of Second Life have bestowed upon us all I've yet to see a single static object that actually needed the LoD factor trick to work. Sometimes strenghtening the LoD means adding an LI or two but as often as not it doesn't and it shouldn't be a big deal.

There has been, and still is, quite a bit of hysteria about getting everything down to 1 LI and I'm afraid I've been guilty of fueling that hysteria in the past. I can make a good solid 1 LI mesh house. Nobody else can except possibly NEKKA if she's still around. But that's just showing off; there's no real need for that kind of extreme optimization in SL. (Sorry about posting links to my own MP store here but I'd need to illustrate what I mean.) If you can make a house like this at 10 LI with good 1 LoD factor Lod, you're doing fine. This one at 50 LI would be excellent (the mesh that is, I really ought to upgrade the textures for some of my old builds) and considering it has a mesh timberframe, 70 or 80 wouldn't be bad either. For this family house - 150 perhaps.

---

The real problem is the lack of documentation, tutorials and examples to follow.

The way Second Life handles LoD is so radically different from other game/virtual reality engines so any generic info about LoD is likely to be more misleading than informative.

LL's official documentation, written by developers for developers, is useless at best downright harmfull at worst. Except for the bits that Hyper Mole added that is. But that was back when nobody had figured out how to handle mesh in SL and she had barely started before she passed away.

The current Moles go for the brute force aproach to LoD, strengthening the models to a ridiculous and seriously LI heavy amount. That 21 LI house I linked to is about the same size and complexity as the Bellisseria suburb homes. Check their LI and you'll see what I mean.

For those who prefer to build with mesh and convert to prims, MeshStudio used to have an active support group helping new users through the basics. I don't think MeshStudio exists anymore though and the surviving competitor, Mesh Generator, comes with no support whatsoever and a bunch of horrendously poorly made examples you should never ever dream of following. Contrary to what many seem to think, it is actually possible to make excellent LoD models with Mesh Studio or Mesh Generator but who's gonna tell you how?

The uploader steers the inncovent towards that monstrosity of an automatic LoD model generator.

There is this forum of course. I think everything you need to know about mesh LoD is here. But it's all buried deep and you have to keep in mind that we too had to figure it out the hard way so older posts from even the best experts may still be seriously flawed.

I would recommend Chic's videos. They're more than a little bit outdated by now but they still present the basics in a nice friendly way. (I can't remember the URLs, @Chic Aeon, can you help us?)

Since I'm not in a modest mood today, I'd also recommend the two case studies I posted in my blog:

• https://chinrey.blogspot.com/2019/09/creating-lod-models-case-study-1.html
• https://chinrey.blogspot.com/2021/01/creating-lod-models-case-study-2-lesson.html

I wish I could do more of those but unfortunately there are only 24 hours in a day.

Edited March 27, 2021 by ChinRey

[Mollymews]

7 hours ago, FinnfinnLost said:

The thing is, most best practices talked about here are not difficult for an amateur to learn. If you're going to create stuff for any domain (be it your own game or a mod/content for an existing), you just have to read up on it a bit so you can create decent stuff

this is fair comment

i just say that when I say amateur I am trying to be kind

when a person takes a professional approach to their work then they are a professional. Whether is fulltime or a hobby, whether they are just starting out or been in it a long time. Professionalism Is about the approach

what do those who take an amateur approach do?

consider: They make some badly formed thing. Sometimes because they don't know any better because just starting out, and other times they do know better but do it anyway because they can reduce costs. And then they put it on the market and sell it. And then when we (everybody else) goes: This is awful! Is killing the view. They go: What did you expect for 1LI and L$20

with a metrics standard that awful newly-made thing is never going to be 1LI, is going to be 4LI. Person shopping goes: 4LI for a teacup ?! I am not buying that. I will get me some 1LI ones. 1LI tea cups are not going to crumble when I move my chair back from the tea table

Edited March 27, 2021 by Mollymews
typs

[Coffee Pancake]

Amateurs are not the problem.

[Mollymews]

11 hours ago, Kyrah Abattoir said:

So essentially do "nothing"? At some point we have to stop assuming that creators are just misguided, if they've been doing business for any length of time they know exactly that what they are doing is wrong, they simply won't care unless it kicks them in the wallet.

i have been trying to avoid this conversation. But yes, you are right. It does have to be had

with pick-a-number then the thing that is most important is that there is a number. In the Linden viewer which drives all of this then the number per user is pretty much anything people like. And this has to change

people on less powerful computers do tho need to be able to dial it down

4 may be too high. It can be less. Linden have a lot of data on resident computers and I am sure that they will be able to pick a number that will be ok for most people

is the same with a metrics scale. 8+4+2+1 may be too aggressive for people starting out making stuff to achieve. It might be that 4+3+2+1 is a better balance.  So 4+3+2+1 = 10 is optimal. 4+4+4+4 = 16 is least optimal

the main thing is that there are numbers that people can work too

Beq made a point earlier that a really good designer can do 4+2+1+1 so why should this be penalised. From a professional builder's pov then this is a valid point. The issue with this tho is how to write an algorithm that can distinguish between a good 4+2+1+1 model and a bad 4+2+1+1 model, for every possible model that can be made within these parameters

and my view is that because there isn't such an algorithm it shouldn't result in paralysis

[Mollymews]

1 hour ago, Coffee Pancake said:

Amateurs are not the problem.

yes you are right

amateur is maybe not the right word for this conversation. As strictly speaking anyone who sells their stuff is not an amateur. So my bad on using this word to describe the difference between people who take (and/or aspire to) a proficient approach to their work and those who don't

maybe proficients and non-proficients might have been better for me to use

[Coffee Pancake]

5 hours ago, Mollymews said:

maybe proficients and non-proficients might have been better for me to use

I was being very literal, newbies, the inexperienced, learners are not the problem.

They are probably making all kinds of ugly mesh with bad textures and no baked normals and no LODS, but they aren't turning out 5000 tris shoes on purpose, or finding creative ways to get around what few limits exist .. or even aware of the limits till they mess up and post about it on these forums.

Professionals have to know what they're doing to get the kind of top tier exposure events bring, which means they either don't care or don't believe it matters. Even if they aren't making the mesh themselves (highly likely they aren't), if this was important, the people who are could EASILY hit reasonable tris targets and make LOD's by hand.

This is why asking or hoping creators do the right thing always fails. The newbies at least try to get it right.

Edited March 28, 2021 by Coffee Pancake
typo

[ChinRey]

4 hours ago, Mollymews said:

people on less powerful computers do tho need to be able to dial it down

No, I don't think they do because LoD is not a variable. There are many factors that can be tweaked to reduce the load on less powerful computers (and improve the visuals on stronger ones) but LoD is not one of them. As Beq pointed out, reducing the LoD isn't actually that efficient a way to reduce the actual render load and it's quite destructive to the overall visuals. It needs to be what it needs to be, nothing more and nothing less. Use simpler shaders, limit the texture resolution, even (I hate to say this) lower the draw distance. Those are only some of the far less destructive and far more effective ways to reduce the render load.

 

[Mollymews]

33 minutes ago, ChinRey said:

No, I don't think they do because LoD is not a variable. There are many factors that can be tweaked to reduce the load on less powerful computers (and improve the visuals on stronger ones) but LoD is not one of them. As Beq pointed out, reducing the LoD isn't actually that efficient a way to reduce the actual render load and it's quite destructive to the overall visuals. It needs to be what it needs to be, nothing more and nothing less. Use simpler shaders, limit the texture resolution, even (I hate to say this) lower the draw distance. Those are only some of the far less destructive and far more effective ways to reduce the render load.

 

you are right about this as well

i think we agree that people on slower computer do need ways to downtune. Draw distance, no shadows, number of avatars, render objects low to high, etc

how this is implemented I leave to those who know how to code it up

I tend to look at these things from a product pov. What is it that we want from the product?  The product in this case being the resident inworld visual experience. At the moment the experience is not all that great. So the OP question. How might this be improved?

i think that any movement along this path is more about the design/direction of the product model. Technical efficiencies can inform the implementation of a product model for sure. But what they shouldn't become is the sum of the product's direction

as I mention in another post if it was just about raw FPS (a technical efficiency) then I would use a different viewer to the one I normally use. But it isn't just about technical efficiencies that inform this kind of choice. There are other factors (functionalities) that come into this. And these informing factors are different for different people.  However, that I might choose other factors for myself shouldn't overly interfere with the product itself, nor should my choices impact adversely on other users' overall visual experience obtainable from the product

and in the case of user-made objects which do impact on other users, then if my choices in this aspect can't totally be prevented from affecting other users then I should expect that there will be a cost/penalty for me in doing this

[NaomiLocket]

The only thing you do, do (if you do), is implement a better decimator that ensures a minimum hull integrity for auto generation. No social engineering, no degrading loss of ethics and moral standing with abusing responsibility through marketing twists, deceits, sarcasm, or trying to be 'cleaver'. You just ensure a minimum hull integrity for auto decimation, and leave manual hulls with full control. However in doing so, you would also expect to ensure the lower limit of cost has not risen.

Yes, that does mean some that will do it anyway will do it externally.

That solves the importer.
Leaves the newbies, intermediates, and well practiced alone.
And leaves the effectiveness of good content to their own marketing responsibility and strength of their merit. It is not the responsibility of the import dialog to push any one content over another. It has one job. To validate and accept a valid mesh. That is its only job.

I would fundamentally agree with Molly and Beq on this topic, and also ChinRey's last post.

[ChinRey]

1 hour ago, Mollymews said:

i think we agree that people on slower computer do need ways to downtune. Draw distance, no shadows, number of avatars, render objects low to high, etc

I said earleir to Beq that we couldn't tackle all problems at the same time but although that's usually good advice, it was probably out of place here.

What we really ought to do, is start with the question: How do we make a virtual reality that offers a good visual appearance and performance on low spec hardware but still allows people with high spec computers to take full advantage of the power they have? This big picture is sadly missing and it's not about content or software, it's all about content and software.

[ChinRey]

1 hour ago, NaomiLocket said:

The only thing you do, do (if you do), is implement a better decimator that ensures a minimum hull integrity for auto generation.

A more suitable decimator would certainly help but at the end of the day there is no algorithm for good design just as there is none for truth.

I won't say a "better" decimator because that's not fair to GLOD but I'm glad you did because it illustrates so well the core problem with mesh in SL. I once had a talk with Nyx Linden about the decimator and he insisted that GLOD was the best one available. I thought he was insane until Beq explained it to me. GLOD is indeed the best there is for the task it was meant to do. It was specially made to generate LoD models on the fly in environments where there are no pre-made ones. It was never meant to do a good and thorough job, it's all about getting it done as fast as possible. When you upload a mesh, you can always wait the few seconds it takes to do a more thorough analyzis of course but the LL developers totally missed that point because they did not understand well enough what they were working on. They had just read somewhere that GLOD is good and it never occured to them to ask what it was good at. To them a decimation algorithm is a decimation algorithm, no further distinction. To use a music metaphor, they tried to build a violin using the construction drawings of a trumpet and ended up with this:

(Disclaimer: the Stroh vioin is actually a very cool instrument. I love it and wish I had one but I wouldn't use it in a symphony orchestra.)

This lack of basic task analyzis is a recurring theme throughout SL's mesh implementation and it's the root of nearly all the problems we have with mesh today, including the LoD issue.

Edited March 28, 2021 by ChinRey

[Kyrah Abattoir]

3 hours ago, NaomiLocket said:

And leaves the effectiveness of good content to their own marketing responsibility and strength of their merit.

It doesn't work. Even skyrim modders (who should know what they are doing given how much of a pain it is) are usually completely clueless as to why all the textures aren't in 4K and why details aren't fully modeled all the time.

[NaomiLocket]

@ChinReyPretty much, and one up to the actual task wouldn't be a walk in the park for sure.

[NaomiLocket]

11 minutes ago, Kyrah Abattoir said:

It doesn't work. Even skyrim modders (who should know what they are doing given how much of a pain it is) are usually completely clueless as to why all the textures aren't in 4K and why details aren't fully modeled all the time.

It doesn't need to work. There is a point where it is not the problem of the system and the system doesn't need to make it the concern. Skyrim modders not understanding that Skyrim was made before 4k was a thing has nothing to do with when the next remaster hits. Second life based on freeform content doesn't have the same argument any other platform does for sitting on its engine too long either, but that is another topic.

The fact of the matter is you do not need to, and should not, coerce an ideal of creation at this level. It is simply not the way. Using a more suitable decimation solves the original described point about collapsing objects that don't need to be collapsed fully. Or basically shifting the bottom end of the threshold. Seeing as when you strip all the nonsense out of the OP that was tacked onto LOD and the need to ignore balanced settings, that is the matter you are left with and going no further is required.

[Kyrah Abattoir]

8 minutes ago, NaomiLocket said:

Skyrim modders not understanding that Skyrim was made before 4k was a thing has nothing to do with when the next remaster hits.

You are completely missing my point, the reason why it's not all in 4K has more to do with engine stability and the target computer the game was made for.

I keep saying that LL need to step in an decide "this is the performances target for this year and deviating from that target will have you penalized." but they never will judging from the way they bend over backward when high profile creators break the platform.

[NaomiLocket]

Seeing as you can upload something upwards of 60li and skyrocketing without following their rules for a given reason, I'd say they did exactly as you said. My point is, content creators that create same or similar goods in a kinder fashion need to also be merchants doing a merchants job. Can't complain if someone that owns venue x, buys item y that looks great to them at a price they don't mind, and puts it in venue x & z of their choice. Ignoring that is missing the point of SL's design and function along the entire chain.

Edited March 28, 2021 by NaomiLocket

[Mollymews]

11 hours ago, NaomiLocket said:

Using a more suitable decimation solves the original described point about collapsing objects that don't need to be collapsed fully. Or basically shifting the bottom end of the threshold.

there isn't a more suitable decimation method than what there is now for this particular use case.  To realise a more suitable decimation method then somebody yet unknown is going to have invent a yet unknown method

shifting the bottom end of the decimation threshold in terms of the view is already available in the viewer on a individual user basis. Render Objects = High

while this can be helpful  it doesn't address the fundamental issue that the OP posits

 

some more thoughts. There is a social aspect to this

when we leave it to the object creators then we get what we have now

which raises the question: With user-created objects, is what we have now visually good enough?

if the answer is Yes then all we have left to do is make what we have go faster (perform more efficiently). And the conversation ends here

if the answer is No then what is the cause? The follow-up question being: If we know the cause then how might this be addressed so that the answer to the first question can be Yes

Linden have done quite well in how they have approached these questions. Quite well meaning there is some method to their thinking

1) Make a decimator. And in doing so, use the best one available and make it go as fast as possible. Texture decoding is a result of the same method of thinking. Use Kakadu as is the best available for that particular task   
2) When creators are not satisfied with the output of the decimator then make it so that creators can do their own decimations

the OP raises the 3rd next question. What to do with creators who decimate in ways that result in a No to the original question: With user-created objects, is what we have now visually good enough?

this is not wholly a technical question. Is a lot of social in this. The standard social response to this is: Let the market decide. Customers will buy stuff that looks good on their screens, and creators who do not make good-looking stuff will find few to no buyers

so the 4th question. How is that working out in terms of answering question 1?

so the 5th question. Why is let-the-market-decide not working out for all of the residents all at the same time?

the answer to that is: Well! it works on my screen

works-on-my-screen is the fundamental social thought/aspect which holds the world back. A social aspect which is enabled by the viewer

Linden have undertaken with LDPW to show what their understanding of works-on-my-screen means. Linden My means my residents, my customers. All of them, all at the same time, with all of their differing screens

and internal to the LDPW object creation processes there is no let-the-market-decide, works-on-my-individual-mole-screen. There is a stringent LDPW design model with tightly defined parameters

the big decision for Linden is: Do we impose a stringent design model on all resident creators? And if the answer is Yes then how would this be technically implemented. And when creators do not follow the stringencies then what are the penalties for not doing so

is a big decision this and I would like to think that Linden can find a way to make it

as should Linden make this decision, then the next big decision is less big than it appears now. That decision is: What is the cap on avatar appearance?  Hit the cap and have to take stuff off to stick other stuff on

i don't suggest a avatar appearance cap lightly either. A avatar appearance cap will impact on me personally. I am a bling bunny of the works-on-my-screen kind. 38 attachments is often not enough for me. I have to detach to be able to stick other stuff on myself. And when I do then I go: Gah! blinking Linden. oh! well  

 

[Gabriele Graves]

2 hours ago, Mollymews said:

which raises the question: With user-created objects, is what we have now visually good enough?

I cannot imagine a time when the visual quality of what we have would no longer satisfy me.  That goes for both items I wear and that I rez down.  I am definitely very picky about visual quality but I don't equate high visual quality with a high degree of realism.  It doesn't have to be "real" looking for it to look great.

[Mollymews]

1 hour ago, Gabriele Graves said:

I cannot imagine a time when the visual quality of what we have would no longer satisfy me.  That goes for both items I wear and that I rez down.  I am definitely very picky about visual quality but I don't equate high visual quality with a high degree of realism.  It doesn't have to be "real" looking for it to look great.

yes is definitely a difference between visually suitable and visually satisfying/pleasing

in the context of this conversation then I think visually suitable means that an object looks like it is holding to a representation of its intended form when viewed from a distance. The operative being looks like

 

[Kyrah Abattoir]

@MollymewsYou raise a good point here and I think it is worth repeating one of the issues that the current "model" has that, in my opinion, is causing the most problems:

1. Avatars are not metered in any meaningful way: This is a difficult problem to solve because pretty much everything you can buy today, and have been able to buy for the past decade for your avatar has been built with the baked assumption that "it is a wearable, therefore its complexity doesn't matter". It is also not a "new" problem either, back in 2004, being able to wear half a region worth of prims was already a concern. But back then Linden Lab didn't have much concern for aspects of Second Life that did not affect their servers.
2. Complexity is unenforceable, by design: It is as far as Linden Lab has gone to address the issue of the load of avatars on the client, and it essentially boils down to "we aren't taking any real decision here because we don't want to be the bad guys." What little good it did to empower resident to be the arbiters here falls because it was intentionally designed to prevent landowners to use it in any capacity (don't unionize plz). There is no support for it baked into the parcel/region tools, and the scripting support is entirely based on trusting that clients will report it in good faith.
3. Land impact calculation promote bad behaviors: This is at least aknowledges, there is no incentive in place to produce reasonable LOD models (whether they are generated or user supplied) as the lowest setting will produce the most "marketable" value for the seller while being "mostly" hidden from the customer at the moment of sale (you won't notice bad LODs immediately, and that is assuming that you know this is intentional and not "SL bad".
4. Metering is not consistent: Regardless of the usage all metrics should matter, LODs should matter for rigged wearables as well as unrigged wearables, texture count should matter for wearables and for rezzed objects, script count should matter for wearables and for rezzed objects. From a rendering standpoint, it doesn't matter whether you are making a chair or a jacket. You can make a case that avatars should be more detailed than the environment, and you would absolutely be right, but at the end of the day, it all has to go through the same clogged rendering pipeline.

Edited March 29, 2021 by Kyrah Abattoir

[FinnfinnLost]

Avatars might actually be a bigger problem overall. Rezzed furniture and such is predictable, performance for it won't change unless something is removed, replaced or added. If an area is bogged down with horrible sculpts, at least you know and can avoid it.

However, a 10000 tri hair avatar or ten may just appear within render distance at a moment's notice and ruin performance for everyone involved, no matter how clean and optimized the rezzed objects in the vicinity are.

In addition, I'm getting the feeling that the "low LI" measure of quality is applied to every object offered on the MP, with "low" not being a variable based on complexity, but some absolute value expected by customers. I'm currently creating a neon sign outlining a human. I've been optimizing the crap out of it, retopologizing multiple times, and created custom LODs. Since the shapes are complex and curved, I can only do so much and it clocks in at a minimum of 17 LI while still somewhat resembling the shapes. Can probably shave off one or two more, but loss of detail would increase drastically.

My point is, I see how people are tempted to just zero out every LOD to get to as low LI as humanly possible. It just looks good in the description and that generates sales. How would we even go about tackling that psychological aspect? FAST MESH is a decent strategy to begin with, but would it be enough?

[Coffee Pancake]

9 hours ago, FinnfinnLost said:

Avatars might actually be a bigger problem overall.

Oh they are. Animated avatars covered in stacks of rigged objects is why we can't have nice things (with more than 2 friends at a time). The performance hit explicitly due to avatars in groups is staggering and is explicitly due to rigged mesh.

Would enforced LOD targets help this?

yes

[Gabriele Graves]

Am I right in thinking that it comes down to how many triangles and textures there are overall and it largely doesn't matter how that is broken up into different attachments, link-sets and objects with the exception that if everything an avatar was wearing were all one mesh object then it could be better optimised due to it all being designed as one piece?  Or does it make a big difference how it is broken down into each attachment?

[Coffee Pancake]

Attachments are just collections of objects, the more objects you have the more things need to animate.