[opensource-dev] Standard Human Mesh Avatar

Daniel danielravennest at gmail.com
Mon Dec 12 09:24:04 PST 2011


There is no single number you can define for all cases.  During the mesh 
beta one Linden
staffer noted the mesh cost formulae were made assuming 300K triangles 
for the entire scene being
rendered would allow low end PCs to run at adequate frame rates.  Note I 
said "triangles", which
is what graphics cards process, not vertexes.  High end machines can 
handle a lot more.  Crytek
(designer of the Crysis games) recommends level designers keep to under 
3 million triangles per
frame, versus whatever PC hardware they recommend for those games.  How 
much work the
SL viewer does with each triangle depends on graphics settings, but 
generally a given hardware
will process X triangles per second, and so frame rate will tend to go 
inversely with scene complexity.

The existing default SL avatar is about 8000 triangles, each sculpt map 
can add 2000, and each prim
can add 100 to 2000 or so depending on type.  So an avatar wearing a lot 
of detailed attachments
can add up to a lot of theoretical triangles.  I say theoretical because 
the SL viewer applies "Levels of
Detail" (LOD) on the theory that objects at longer distance would end up 
with geometry which is
less than one pixel on the monitor.  This is unnecessary processing, so 
the viewer applies simplification
of what it renders at various distances.  For prims and the standard 
avatar the LOD levels are built in.
For sculpts it discards alternate rows and columns of the map texture, 
and for mesh you explicitly
define the LOD geometry at the time of upload.  LOD levels kick in at 
specified multiples of the object
size vs camera distance.  So if you design a mesh avatar as a single 
object, the levels will switch at
multiples of the full body size.  If you design it as separate body 
parts, the switches will happen per
part, so generally at closer distance.

The final consideration is what environment will the avatar be used in.  
One made for fashion photography,
where only one will be in view, and frame rate does not matter, can be a 
very detailed.  One made
for SL role playing games, where speed matters, or for visiting popular 
clubs, where lots of other attachment-
heavy avatars will be in view, should stick to lower detail.  For the 
latter cases I would suggest staying under
double the default avatar (ie to < 16K triangles), on the assumption 
that the default avatar when hidden
by an alpha texture subtracts 8K from the rendering task, and an extra 
8K total is not a large impact given
whatever else the avatar has in attachments.  Levels of detail by 
default go down by a factor of 4 per level
in the upload window.  You are free to upload whatever geometry you want 
for each level to maintain
decent looks for your avatar, but that is a reasonable guideline for how 
aggressively to simplify each LOD.

Message: 3 Date: Mon, 12 Dec 2011 09:40:26 -0500 From: Robert Martin 
<robertltux at gmail.com>
> the trick with MakeHuman is it can be used to generate multiple meshes
> it would be understood that it would need to be downsampled (to below
> XK vertexes) before it could be used on the grid. The big question
> would be What is the recommended Maximum Vertex count?? (define X
> please)
>



More information about the opensource-dev mailing list