Has there been any update on project ARCTan?

[Coffee Pancake]

6 minutes ago, Verbena Melodious said:

I agree with ArabellaJones on this one. Creators should optimize their meshes to help reduce lag. I understand wanting something beautiful and complex, but not all of us can afford 1,000,000 mbps internet or the top of the line hardware needed to render SL content remotely gloriously. I, personally, feel creators don't care about "the complexity thing" nor care about the lag it causes, for the sole reason that making these lagtastic contraptions pays their bills.

Mesh complexity is entirely based around manipulation and gaming of Li calculations.

SL meshes are, as a general rule, not really over complicated. There will always be exceptions, most of them found attached to an avatar, but in general, anything rezzed in world is comparable to modern game content.

As a counter point to perceived over complexity, how can a mesh object be better optimized than one that ceases to exist the moment you're 3 feet away from it. It's technically impossible to beat "4 random tris".

The problem comes when users crank up the LOD settings in their viewer to counter this hyper aggressive Li driven optimization, now you're rendering everything, all the time.

 

I would be in favor of LL junking the existing system entirely, throwing all lower LOD meshes in the trash, and having the viewer generate lower LOD models on demand based on users graphics settings (there is also the side effect that LOD mesh generation can be improved over time just by shipping new code in the viewer).

There is a special case to be made for artist created LOD models as they are almost always visually superior to generated ones, sadly these would also be junked in any such move by LL (I asked at the last TPV, and no, they can't tell which meshes were uploaded and which were generated).

To correct for this, LL should continue to allow artist created LOD meshes to be uploaded after "operation clean-slate" removed all LOD meshes and LOD generation in the uploader, creators who cared about making such things (myself included) could then issue updates to their products to restore the lost work.

This could also incentivize other creators who previously relied on generated LOD meshes to get into doing this. There should be no Li difference between identical objects using artist created LOD mesh and realtime generated ones to prevent gaming, it would just be a something creators could use as a badge of quality.

[ChinRey]

Necroposting is usually frowned upon here of course but this thread needs to be bumped. It's been almost two year since Kyrah posted it and three years since ArcTan was first announced and still nothing has happened.

 

26 minutes ago, Verbena Melodious said:

I, personally, feel creators don't care about "the complexity thing"

Some don't care, some are simply ignorant, some take the official LI and ARC (or whatever it's called these days) calculations at face value.

But for those of us who do care and understand, it's very frustrating that those two numbers are so unreliable, ARC unreliable enough you might as well pick a random number. It makes it all but impossible for us to estimate how much load our works cause and for the users to tell the good from the bad.

Those of us who know enough to be able to make our own more reliable estimates face and additional dilemma: should we optimize for the lowest possible LI/ARC to make our works more sellable or should we optimize for the lowest possible actual load to make them more SL friendly?

ArcTan was supposed to sort this out and I'm really sorry to see that it appears to be dead. I'm not surprised though.

[arabellajones]

I would just like to note that the LOD switching distances are calculated differently for rigged Mesh, and that ripples through in display-weight/complexity calculation and what people can see, but too many Creators seem to neither know nor care about this.

I don't claim to have got this right on my clothing products, the High-LOD can look a little poor if you zoom in pervy-close, but I try to have things looking OK.

[Lucia Nightfire]

20 minutes ago, arabellajones said:

I would just like to note that the LOD switching distances are calculated differently for rigged Mesh, and that ripples through in display-weight/complexity calculation and what people can see, but too many Creators seem to neither know nor care about this.

Many do know as they LOD starve their content to save on upload costs. Costs that would have been recouped on the first sale.

I have engaged many creators on this LOD starvation tactic and the majority have actually told me they do it to save money on the upload cost and because it won't look like crap when worn, only rezzed. With unrigged mesh, many typically use the same LOD triangles for Medium as are in High, while some further do the same for the Low LOD level. I have tried telling them the uploader in most cases offers sufficient triangles by default.

There are still many that have told me they haven't a clue what triangle values to use and just type in anything or leave it alone.

Edited December 12, 2021 by Lucia Nightfire

[animats]

It's come up at Creator User Group a few times in the last year.

A solution has to be backwards compatible, not increasing the LI of existing objects. Otherwise, when deployed, some people will have their parcels go over limit and objects will be returned. There's already been one change like that, with "old" and "new" LOD accounting. Give a prim torus a material and watch the LOD increase as it changes to new accounting. Adding a "newer" LOD accounting is not a popular idea.

There's also the issue that the LI accounting is tuned to the way the viewers draw, using OpenGL. Someday the viewers will switch to Vulkan, which has totally different relative costs. (No "draw call" cost.). Speeding up the viewers (which is actually being worked on by LL, after a long hiatus) is more productive than messing with the LI accounting.

Avatars have their own problems. Avatar LODs don't work, because every attachment has the LOD distances of the main avatar. If you don't do it that way, the avatar starts coming apart at distance due to much existing clothing having terrible or non-existent lower LODs. 20,000 triangle shoes are common. You can try this by putting purchased mesh clothing on an an animesh character. The LODs will usually be terrible, clothing parts will disappear at distance, and the LIs will be huge. Which is why avatar complexity is such a problem.

(There's clothing with proper LODs made for animesh - Meli Imako and Duck Girl make some. It's rare.)

[Paulsian]

something's funky is a 12 sided object less complex than an 11 sided object prolly is and prolly answered my own concern.. both are max lod and perfect. 

Edited December 15, 2021 by Paulsian

[ChinRey]

11 hours ago, Paulsian said:

something's funky is a 12 sided object less complex than an 11 sided object

That's genuine and not a flaw. The mesh asset file is compressed before it is transferred and it's always easier for a compression algorithm to handle a symmetrical object than an asymetrical one. So the file size and thus the download weight for the 12 sided shape will be slightly lower than for the 11 sided one. The difference is marginal but in some cases, such as this, it's just enough to flip the LI up or down. If a shape has a download weight of 1.499, it's rounded down to 1, if it's 1.501 it's rounded up to 2.

Edited December 15, 2021 by ChinRey

[Paulsian]

2 hours ago, ChinRey said:

That's genuine and not a flaw.

ty  

[Quarrel Kukulcan]

10 hours ago, ChinRey said:

The mesh asset file is compressed before it is transferred and it's always easier for a compression algorithm to handle a symmetrical object than an asymetrical one. So the file size and thus the download weight for the 12 sided shape will be slightly lower than for the 11 sided one.

I'm confused. 11-sided polygons are just as symmetrical as 12-sided ones if they are also aligned with world coordinates and regular -- which this one probably is, given how most software generates arbitrary N-gons by default.

[Paulsian]

 

Here's another situation with 2 identical objects. I build a two piece arch with max LOD and when the are uploaded together 5LI when I unlink 1 side is 2LI and one side is 3LI.

When I uploaded them seperate each is 2LI when grouped 5LI.

It looks tacky to have a 3 LI side and a 2LI side. Something is not calculating each side of arch evenly when multiple piece models are uploaded at same time. I could upload each seperate but that will cost twice as much to have the 2 each even though they will be 5 when linked. 

Altogether it's 4.027. Each should be 2.0135 / 2.014 That's odd  

I think it's a test grid calcuation issue because when I uploaded to the main grid it uploaded as 4 linked and when unlinked 2 each. yah and ugg.. 

Edited December 16, 2021 by Paulsian
after thought

[ChinRey]

9 hours ago, Quarrel Kukulcan said:

I'm confused. 11-sided polygons are just as symmetrical as 12-sided ones if they are also aligned with world coordinates and regular

Not really. Look at this rather crude illustration:

This 12 sides polygon only has four different y coordinates for its vertices and one of them is zero. Well, it has seven but three of them are just negative versions of some of the others so most of the actual data is the same.

This 11 sided polygon on the other hand, has six completely different values for its y coordinates:

It's a similar story with the x, u and v coordinates and the normals and sometimes it adds up to enough to affect the LI.

Edited December 16, 2021 by ChinRey

[ChinRey]

2 hours ago, Paulsian said:

Here's another situation with 2 identical objects. I build a two piece arch with max LOD and when the are uploaded together 5LI when I unlink 1 side is 2LI and one side is 3LI.

You shouldn't do it that way anyway of course; upload one arch and copy it in-world. No need to spend more L$ in upload fees and add more entries to the assets database than necessary.

But there's another factor that affects the compressability of the mesh file, the order the elements are sorted. This is why it's usually recommended to export meshes as quads from Blender and let the uploader sort out the triangulation. Blender's way to sort the vertice list tends to be rather messy while the uploader tends to do a much cleaner job. (Yes I know, I've just said something nice about the uploader! Wait for the sky to fall.)

If you click on the "More info" link on the edit palette you'll see that the download weight for the two arches are pretty much the same, they may even both show up as 2.5. But again, since the weights are rounded to the nearest integer when LI is calculated, there is a difference between 2.499 and 2.501.

Edited December 16, 2021 by ChinRey