Why is there so much high-poly mesh in SL?

[Penny Patton]

On 11/26/2018 at 10:34 AM, Arduenn Schwartzman said:

but it it was more detailed and wonderful than most 2018-triple A games will ever be.

But SL users will only ever be able to appreciate that detail in heavily photoshopped still screenshots on Flickr. The rest of the time they'll be running SL at low quality graphics settings just to maintain double digit framerates. Meanwhile, those triple A games always look better in motion.

On 12/1/2018 at 7:53 AM, Kyrah Abattoir said:

The beauty of shared spaces, ultimately we are collectively responsible for it, but we don't really see it because individual impact is not perceptible.

Content creators aren't going to police themselves, and the average user doesn't have either the tools or the knowledge they would need to tell good content from bad. If there's ever going to be any real improvement in terms of content optimization, it has to be lead by Linden Lab. Specifically, LL needs to provide:

• Better tools, to nudge people towards better content creation habits.
• Better education. LL has everything from their blog and wiki, to the login screens every user sees every day, they can use all of these to promote better content creation habits, and make people aware of what sort of content issues drag down performance.
• Some method of reigning in the use of resource intensive content by pushing content creators to better optimize their work moving forward, without taking away existing content from the userbase. SL has been around for 15 years, and unless LL really manages to screw something up, it will probably be around another 15 years. If LL can get content creators to start making better content now, the issue will resolve itself with time. And LL has plenty of ways to go about doing this.

 And it goes beyond simply telling people "too many textures eats up all your system's memory" or "SL isn't built to handle that many polygons in a scene". LL needs to be pro-active. Whenever LL does create content, it should be done by people with not just the talent, but also the skill to make content that looks great and doesn't kill your framerates.

• When LL releases a new monthly Premium Gift it should be a showcase of good content creation habits. They should release a blog post showcasing the item, showing how it was made and why it was made that way.
• They should do the same whenever they showcase a new Premium Sim, or any other public content, like the SL games they've been producing over the years.
• LL should regularly be holding contests challenging people to create better content. From avatars, to vehicles, to full sim builds. Rewarding content creators that are employing good habits to build content that both looks amazing and is light on the performance cost.

These are all positive ways to approach the issue.

[animats]

I'm  trying. I've been writing Blender addons to help generate impostors and help with mesh reduction. They're on Github, but fully ready for use yet. With the existing tools, making good lower level of detail models is just too much work.

The really crappy decimator in the viewer's uploader is a big part of the problem. That's what generates terrible default low-LOD models. If it was as good as Blender's "Decimate" (which is open source) things would be much better.

High-detail models wouldn't have so much load impact if you only saw them in high-detail in close up. "High" models ought to drop to "Medium" sooner, especially for large objects. There's no reason to be at "High" when you're 30m from a stairway. (54m in Firestorm.) On the other hand, more triangles at "Low" and "Lowest" for larger objects wouldn't hurt.

Blame the tools, not the creators.

(Right now, I have a nice model of a real-world escalator in Blender from a purchased 3D model. And I have a rideable escalator prototype in SL using some of its parts. Getting all that to a low-LI working model in SL is going to take more effort than making it rideable was.)

[Theresa Tennyson]

7 hours ago, Penny Patton said:

But SL users will only ever be able to appreciate that detail in heavily photoshopped still screenshots on Flickr. The rest of the time they'll be running SL at low quality graphics settings just to maintain double digit framerates. Meanwhile, those triple A games always look better in motion.

 

Which is what significant numbers of SL users enjoy doing, so they buy the items and therefore provide financial support for both those merchants and Linden Lab.

[Kyrah Abattoir]

7 hours ago, Theresa Tennyson said:

Which is what significant numbers of SL users enjoy doing, so they buy the items and therefore provide financial support for both those merchants and Linden Lab.

The big question is whether they do that because they will never be able to run SL at normal settings or if they really only care about SL for making snapshots.

[ChinRey]

4 hours ago, Kyrah Abattoir said:

The big question is whether they do that because they will never be able to run SL at normal settings or if they really only care about SL for making snapshots.

I suspect it's a little bit of both. But Theresa's argument is very valid in any case because she describes how SL actually is today. When we discuss the could-have-beens and should-have-beens and can-becomes, we have to take the status quo into account.

[Penny Patton]

8 hours ago, ChinRey said:

But Theresa's argument is very valid in any case because she describes how SL actually is today. When we discuss the could-have-beens and should-have-beens and can-becomes, we have to take the status quo into account.

This is only true if you believe that those who currently enjoy SL would be put off if SL looked better and ran better for them, and at the same time you must also believe an SL that looks and runs great would fare no better in drawing in users. 

If LL had addressed the content optimization issue at any point in the past, the people Theresa brings up would at worst enjoy SL just as much as they do now. At worst. And on top of them, and the revenue they provide to both SL merchants and Linden Lab, there would be many more users adding to that.

[Bitsy Buccaneer]

I wonder if there could be any traction for a marketing tag like "Ideal for High Rez Photos" for the hyper-detailed stuff. Just thinking out loud here, so please don't trash the details but maybe make friendly improvements instead, like better wording.

Something which would feel like a plus for the uber-close-up crowd but would also introduce the idea that it was extra and had a specific use. That might be a way to slip in the idea that it's going to create problems for other uses.

A downside I can see is that it would might encourage even more hyper-detailed stuff and its popularity in photos would encourage more sales and most users would just wear it anyway. So I think I've talked myself into disagreeing with my original hope. I'll post this anyway in case it sparks a better idea for someone else. I can't see "Optimised for a better SL experience" catching on, even though it would appeal to folks like us.

[ChinRey]

2 hours ago, Penny Patton said:

If LL had addressed the content optimization issue at any point in the past...

Yes but that is a big if.

The difference between you and Theresa here, is that you talk about how it could and should have been, Theresa about how it actually is.

Me, I fully agree with both of you in this.

 

1 hour ago, Bitsy Buccaneer said:

I wonder if there could be any traction for a marketing tag like "Ideal for High Rez Photos" for the hyper-detailed stuff. Just thinking out loud here, so please don't trash the details but maybe make friendly improvements instead, like better wording.

I've had similar thoughts myself but as you say, there are a number of pitfalls along that road. A distinction between "still photo" and "dynamic" content might or might not have worked if it had been drilled into people right from the start but in either case, it's too late now. If I understood @Grumpity Linden right, the original purpose of the RenderVolumeLODFactor was to allow people to temporarily override the LoD system for still photos and we know what that led to.

I also think Linden Lab has made it absolutely clear that optimising SL for dynamic performance is not on their priority list. They have to do a little bit of damage control but it's not something they're going to put any substantial effort into. The Town Hall meeting last month spoke volumes about this, not by what was said btu by what was not said. There were several performance related questions in the forum thread but but apart from a brief mention of sim crossings none of them were brought up during the meeting.

As much as I hate it, I do think LL is right there. Effective solutions to all of SL's performance issues would require a huge amount of work and it would disrupt the current SL communities and SL economy in so many ways. Yes, it could have been different and yes, it probably should have been different. But since we don't have an llGoBackInTime(); function, this is what it can be.

Edited December 5, 2018 by ChinRey

[Theresa Tennyson]

7 minutes ago, ChinRey said:

Yes but that is a big if.

The difference between you and Theresa here, is that you talk about how it could and should have been, Theresa about how it actually is.

Me, I fully agree with both of you in this.

 

I've had similar thoughts myself but as you say, there are a number of pitfalls along that road. A distinction between "still photo" and "dynamic" content might or might not have worked if it had been drilled into people right from the start but in either case, it's too late now. If I understood @Grumpity Linden right, the original purpose of the RenderVolumeLODFactor was to allow people to temporarily override the LoD system for still photos and we know what that led to.

I also think Linden Lab has made it absolutely clear that optimising SL for dynamic performance is not on their priority list. They have to do a little bit of damage control but it's not something they're going to put any substantial effort into. The Town Hall meeting last month spoke volumes about this, not by what was said btu by what was not said. There were several performance related questions in the forum thread but not a single of them was brought up during the meeting.

As much as I hate it, I do think LL is right there. Effective solutions to all of SL's performance issues would require a huge amount of work and it would disrupt the current SL communities and SL economy in so many ways. Yes, it could have been different and yes, it probably should have been different. But since we don't have an llGoBackInTime(); function, this is what it can be.

There's a problem with the nature of Second Life that I think a lot of the "Well, they otta..." school overlooks. Unlike most other graphically intensive applications, there isn't one single "right" way to use Second Life. It certainly is possible to use it for highly detailed, largely static scenes; it's certainly possible to use it for having a lot of avatars moving around in a single area. People show the desire to do both of these things, and there's nothing "wrong" with either of them.

The problem is something that's well-suited for one of these uses will be ill-suited for the other. Most of the "Well, they otta..." proposals involve making things better for that person's preferred use.

[Bitsy Buccaneer]

6 hours ago, Theresa Tennyson said:

There's a problem with the nature of Second Life that I think a lot of the "Well, they otta..." school overlooks. Unlike most other graphically intensive applications, there isn't one single "right" way to use Second Life. It certainly is possible to use it for highly detailed, largely static scenes; it's certainly possible to use it for having a lot of avatars moving around in a single area.

This is at the heart of what I was trying to get at above. Thanks for putting it better than me Theresa.

I think I'll leave off trying to say more though because I just don't have it in me to hear another lecture about things I already know.

[ChinRey]

7 hours ago, Theresa Tennyson said:

The problem is something that's well-suited for one of these uses will be ill-suited for the other. Most of the "Well, they otta..." proposals involve making things better for that person's preferred use.

... while others resist any proposed change because they don't see them as making things better for their preferred use.

I don't think anybody can claim they speak on behalf of the majority or even a substantial minority of all SL users.

Edited December 5, 2018 by ChinRey

[animats]

Quote

Chin Rey: "I also think Linden Lab has made it absolutely clear that optimising SL for dynamic performance is not on their priority list. They have to do a little bit of damage control but it's not something they're going to put any substantial effort into."

The big problem with Linden Labs is that they think they have the right to suck. Known bugs stay unfixed for years. In the game industry, that gets you laughed out of business. As Linden Labs will discover when they put Sansar on Steam.

This is because nobody else has a big virtual world that's any good.

That will change in 2019. SineSpace is looking into hooking up to Spatial OS, so they can have a big world instead of just little isolated sims. Sominium Space is in beta. There's a deal beween Improbable and NetEase in China to do something big. NetEase brought Minecraft to China. It's getting much easier to put together something like Second Life - the standard game dev tools can now almost do it. One guy put together a big virtual world called Community Garden, as a demo. It's a big multi-user virtual city where many people can visit, chat, and tend the plants.

The successor to Second Life isn't here yet, but it's getting close.

[Kyrah Abattoir]

However, a lot of the "high res content" I tend to shoot down is not "smart" high res. A lot of objects in SL can be simplified and made better with NO loss of visual fidelity because the creator shot up so high to begin with that even 50% of what they've used is still more than enough for any possible use.

[Penny Patton]

9 hours ago, ChinRey said:

The difference between you and Theresa here, is that you talk about how it could and should have been, Theresa about how it actually is.

No, if that were the case I'd have no problem with it. She's using "tail wagging the dog" arguments to say things like this:

9 hours ago, Theresa Tennyson said:

The problem is something that's well-suited for one of these uses will be ill-suited for the other. Most of the "Well, they otta..." proposals involve making things better for that person's preferred use.

There is no truth to Theresa's statment. None, whatsoever. She's fallen for the misguided idea that this is somehow a zero sum game when it absolutely is not. Improving SL's content optimization issues would not ruin SL for anyone who enjoys SL now. (Except possibly the content thieves. Anyone want to stick their neck out for them?)

34 minutes ago, Kyrah Abattoir said:

However, a lot of the "high res content" I tend to shoot down is not "smart" high res. A lot of objects in SL can be simplified and made better with NO loss of visual fidelity because the creator shot up so high to begin with that even 50% of what they've used is still more than enough for any possible use.

Kyrah is hitting the nail on the head here. Theresa would have you believe that optimized content must look worse, therefore it would push those who use SL for Photography away. That is absolutely not the case. The bulk of poor optimization in SL is waste. 1024x1024 textures where over half the texture is empty unused space. Huge textures on tiny objects that you won't even see in those heavily edited Flickr photo albums. High poly models that look no better than a low poly version properly utilizing spec and normal maps. Things that could be fixed easily and the only reason any current content creators might not know how to do it is because they've never seen any need to learn how, but they could pick it up very easily if they needed to.

[ChinRey]

6 minutes ago, Penny Patton said:

No, if that were the case I'd have no problem with it.

Keep in mind that I was only responding to your previous post and Theresa's immediate reply to it. What she added later is a completely different matter.

Edited December 5, 2018 by ChinRey

[Theresa Tennyson]

 

58 minutes ago, Penny Patton said:

There is no truth to Theresa's statment. None, whatsoever. She's fallen for the misguided idea that this is somehow a zero sum game when it absolutely is not. Improving SL's content optimization issues would not ruin SL for anyone who enjoys SL now. (Except possibly the content thieves. Anyone want to stick their neck out for them?)

 

On 12/3/2018 at 11:42 PM, Penny Patton said:

But SL users will only ever be able to appreciate that detail in heavily photoshopped still screenshots on Flickr. The rest of the time they'll be running SL at low quality graphics settings just to maintain double digit framerates. Meanwhile, those triple A games always look better in motion.

So, why do those games look better in motion?

Edited December 5, 2018 by Theresa Tennyson

[Theresa Tennyson]

3 hours ago, animats said:

This is because nobody else has a big virtual world that's any good.

That will change in 2019. SineSpace is looking into hooking up to Spatial OS, so they can have a big world instead of just little isolated sims. Sominium Space is in beta. There's a deal beween Improbable and NetEase in China to do something big. NetEase brought Minecraft to China. It's getting much easier to put together something like Second Life - the standard game dev tools can now almost do it. One guy put together a big virtual world called Community Garden, as a demo. It's a big multi-user virtual city where many people can visit, chat, and tend the plants.

 

And yet, many in Second Life who are in a "big virtual world" prefer to live in a skybox with a security orb. What makes you think they'll be any more interested in other "big" worlds?

[Penny Patton]

1 minute ago, Theresa Tennyson said:

 

 

Ummm, ok? Not sure what you're getting at here. You still seem to be saying that if SL can be made to look better in motion, then the SL Photography community must somehow suffer, without explaining why this must be. It is not a zero sum game.Most SL content can look just as good but without the performance hit. In the extremely rare cases where excessively high-res textures and high poly are needed for photography props, environments, whatever, you could still do that, but I'm saying there should be some sort of discouragement from using that photo-specific content for everyday/non-photo environments/avatars so people aren't using such content where the extra resource use has no benefit and only serves to drag down the experience for everyone.

[Theresa Tennyson]

5 minutes ago, Penny Patton said:

Ummm, ok? Not sure what you're getting at here. You still seem to be saying that if SL can be made to look better in motion, then the SL Photography community must somehow suffer, without explaining why this must be.

No, you were the one who said that those games looked better in motion than standing still. Because they were optimized for fast action.

You keep saying, "Second Life should be optimized!" "Optimized" used that way is a weasel word. Of course it's good when something's "optimized."

But what do you optimize it for?

[Penny Patton]

2 minutes ago, Theresa Tennyson said:

And yet, many in Second Life who are in a "big virtual world" prefer to live in a skybox with a security orb. What makes you think they'll be any more interested in other "big" worlds?

Lot of assumptions here.

• First, the assumption that since many SL users wish to have a private space for themselves means that these people NEVER, EVER leave those security orb laden skyboxes.
• Then there is the assumption that SL users represent the entirety of the market for virtual worlds. If someone is not in SL, they must have zero interest in any sort of virtual world. The SL userbase is it.
• This in turn relies on the assumption that SL is perfect. It has no flaws that push users away, none that another virtual world could avoid. SL has successfully attracted and held on to everyone who holds any interest in virtual worlds. It has never lost a single user beyond those who were never interested in virtual worlds in the first place.

I'm not saying SineSpace will be a hit, I have no experience with it. Like Blue Mars and other failed attempts to enter the arena  it could have it's own crippling flaws that prevent it from taking off with users. I just don't think it's an aversion to "big worlds" that is the problem.

[Theresa Tennyson]

1 minute ago, Penny Patton said:

Lot of assumptions here.

 

Thanks for letting us know.

Isn't asking a question the opposite of making an assumption?

[Penny Patton]

Just now, Theresa Tennyson said:

No, you were the one who said that those games looked better in motion than standing still. Because they were optimized for fast action.

You keep saying, "Second Life should be optimized!" "Optimized" used that way is a weasel word. Of course it's good when something's "optimized."

But what do you optimize it for?

Whoa, whoa, whoa. I never said AAA games look bad when standing still. Again, you're applying that zero sum logic, "Well, Penny said optimized games look better in motion, that MUST mean they look worse in still shots."

And "optimization" is a technical term used universally across the CG modelling industry and communities. SL being the exception, a bubble community of people whose experience with CG modelling begins and ends with SL.

And here again you're falling into that zero sum logic, if content is optimized to render better in a realtime environment, it MUST look worse in still renders/screenshots. 

[Penny Patton]

2 minutes ago, Theresa Tennyson said:

Thanks for letting us know.

Isn't asking a question the opposite of making an assumption?

Context is important. Your question is loaded with these assumptions.

[Theresa Tennyson]

1 hour ago, Penny Patton said:

Whoa, whoa, whoa. I never said AAA games look bad when standing still. Again, you're applying that zero sum logic, "Well, Penny said optimized games look better in motion, that MUST mean they look worse in still shots."

 

How can situation A be better than situation B but situation B not be worse than situation A?

Is there a dairy in Lake Wobegon?

ETA - Upon reflection I think you were trying to say AAA games looked better than Second Life in motion, but you left out the clause that would have made that clear. Regardless, Arduenn said that Second Life could look better in a static situation.

Edited December 6, 2018 by Theresa Tennyson

[animats]

Here's my shopping list on basic tool fixes for creators. This is from the perspective of someone who's not doing clothing. Clothing creators have different problems. These are all relatively simple.

Mesh uploader

• Better mesh reduction in the mesh uploader. Take the open source code from Blender's Decimate or MeshLab's quartic mesh reduction with texture preservation. Both are open source. I've been trying various mesh reduction algorithms, and Blender's Decimate is pretty good. I thought MeshLab's would be better, but it doesn't seem to be. The current built-in one is terrible.
• The uploader's decimator should not force models below 20 or so triangles, unless that can be done without creating holes or significant errors at the outer edges. You should never get those awful see-through models.
• The upload cost algorithm should always allow you 20-30 triangles at lowest LOD for 1 LI. There's no real win to reducing below 20-30 polys per draw call. A thousand objects with 30 triangles are more useful than one object with 30,000 triangles. (An SL scene can have about a million triangles before the frame rate chokes. More if you have a high end GPU and a lot of RAM. Visual quality is about using that million-triangle budget effectively.)
• Get the uploader to accept multi-layer Collada files, so all the LODs can be in one file. If you're uploading multiple models that way, assume they all have the same origin and scale; don't align and stretch bounding boxes to make them fit. This will get more precise alignment of LODs.
• Fix the bug which prevents textures being uploaded via a Collada file, so you can see a useful preview of what you're uploading. Sometimes you may want to upload the textures separately, for "fat packs" and such, but single-file upload should work. Faster development cycle.
• Bigger preview window! (Beq Janus is putting this in Firestorm.)
• The uploader should handle the issues of having to have exactly the same material list for each LOD, whether needed or not. Having to put dummy triangles on a mesh to force a reference to a material is silly.

Blender tools

• Get Blender to put layer info in Collada files, so you can upload one big file with all the levels of detail. (This is an accepted request and may be in Blender 2.8, which also has named layers.) This helps with getting the levels of detail and the physics model precisely aligned.

Edited December 6, 2018 by animats