RE: [GD-Consoles] Textures vs. Polygons

[Jacob T. (C. D. Ltd) <Ja...@Co...>]

I looked into luminance splitting compression method for textures and it is
a massive saving for 24-bit textures, but not that big a gain if you using
256 colour paletted textures.

Just for the record we texture each face individually on a character (well
actually the artists do the texturing ;-) ) and let the artists use as big a
texture as they want and then we split it up into 256x256 chunks when we
build the game data.  This allows the freedom to put more texture detail
into those parts that need it e.g. faces, hands, feet.

Jake

> From: Tom Forsyth <to...@mu...>
> To: Jamie Fowlston <ja...@qu...>, 
> 	gam...@li...
> Subject: RE: [GD-Consoles] Textures vs. Polygons
> Date: Tue, 20 Nov 2001 15:07:53 -0000
> 
> DX/OGL vertex shaders have limits as well, and in some cases they're
> possibly lower than the VU ones. So yeah, splitting meshes up 
> like this is
> something people have to do.
> 
> I don't know of any magic stuff that wil help. There's detail 
> textures - but
> that just adds some "noise" to things - it's good for gravel 
> and grass and
> other landscape stuff, but probably not very useful for 
> people until you get
> horrendously close (cloth weave, fingerprints, pores, etc). 
> If you're aiming
> for ultimate realism, they will be useful, but my impression 
> is that you're
> having trouble well before then.
> 
> Having different bits of the body at different sizes on 
> thetexture can work
> well. We always have faces take up about four times the 
> texture area (i.e.
> twice as big in each direction) as equivalently-sized body 
> parts, because
> people look at faces much more closely, so you notice 
> pixellisation much
> quicker there than elsewhere. They're still on the same 
> texture as the rest
> of the body though.
> 
> For really high detail there's not much that cn beat just using a big
> texture. Obviously you can compress them with the S3TC/DXtn 
> stuff for space
> advantages. Another compression trick I've been thinking 
> about is to take a
> huge texture made by the artist, then extract the luminance 
> as a separate
> texture. Then quantise it to 4 bits. Then remove that 
> influence from the
> original texture, downsample the texture quite a bit down to 
> more sensible
> sizes, and compress using S3TC/DXTn.
> 
> Then we can render the combo using multitexture hardware. The 
> colour and
> subtle gradient info is encoded in the (relatively) low-res 
> RGB texture, and
> the high-frequency luminance info is in the high-res 
> monochrome texture
> (some hardware may not understand A4 texture, so it might 
> need expanding to
> A8 unfortunatelky).
> 
> It's the same trick as used by television of course, and I 
> believe there's a
> Sony sample using this method too.
> 
> Tom Forsyth - Muckyfoot bloke.
> 
> What's he up to now (and can I have a go)?
> http://www.eidosinteractive.com/downloads/search.html?gmid=86
>