Re: [Algorithms] Decals and deferred rendering

[Megan F. <sha...@gm...>]

I'd also suggest G-Buffer, including blending.  As far as the diffuse
channel of the G-Buffer goes, the results will always be right (since
decals are one of the few cases where you can guarantee sorting / not
require depth peeling / etc), and as far as any lighting parameters,
it "should" read right so long as the lighting parameters of your
decal aren't grossly different from the surface being decalled.

... but for that reason, you would need to reign in your decals and
keep them reasonable.  A high-specular shiny decal thrown onto a
low-specular muddy wall would look silly indeed unless it was
hard-edged, given that the lighting properties will be entirely
hard-edged.

I'd wonder, though, if you couldn't mask the edge with a bit of
alpha-blending even on those properties?  So long as you're not
changing any of the values that are rigid (ie. specular mul is
probably safe, but any ID-buffer-esque data is not), you'd be able to
mask the edges pretty nicely with the blend.  So, again, you're
seemingly safe so long as the lighting properties of your decal aren't
grossly removed from that of the surface.

Just depends what kind of game you're doing, really.  Personally, I'd
just make sure the decals matched, since that'd probably look better
period, and bake them into the G-Buffer to allow the much-improved
lighting, AO et al they'd get that way.

-- 
Megan Fox
http://www.shalinor.com/