This is something on my to do list, but in my case performance is not
a big issue.
I've tried alpha to coverage but I didn't get acceptable
results with less than ~15 levels of transparency and even that wasn't
great.
I've been hoping to accomplish what John suggests but haven't
made the attempt yet. If that doesn't work out I will probably use
depth peeling which I hope will give good results with few layers. This
is probably too demanding for your application but if you're interested
there is an nvidia demo with source.
----Original Message----
From:
BO...@vv...
Date: 29-Jan-2008 08:26 PM
To: "Game Development
Algorithms"<gda...@li...>
Subj: [Algorithms]
Maximizing overdraw == Optimizing hair rendering?
Hi,
I'm currently
looking into some simple solution to the ordering problem
for complex
transparent meshes, particularly hair. I've seen the older
work, but
our ultimate use case is less demanding (and I don't think we
can
justify multiple passes at run-time).
Obviously, in the simplest
case, it makes sense to render something like
hair "inside-out". It
seems (to me) like that is the exact opposite
criteria used when trying
to minimize overdraw. Thus, it follows that
using any algorithm to
minimize overdraw, but inverting the cost
function, should produce a
"mostly correct" ordering for hair?
In particular, I'd like to use
the tipsy algorithm from "Fast Triangle
Reordering for Vertex Locality
and Reduced Overdraw" [Sander, Nehab,
Barczak], to instead increase
overdraw. I've tried lots of searches to
unify transparency sorting (eg
hair) and overdraw, but there don't seem
to be many papers that make
the connection. Is there some flaw in my
logic, or is it just so
obvious that everyone takes it for granted?
-Brian Osman
-
Vicarious Visions
|
