[Aqsis-development] =?iso-8859-2?Q?Re=3A=20=5BAqsis=2Ddevelopment=5D=20DOF=20is=20hard?=

[<Ky...@se...>]

> > Well, I think this is semi-wrong, since what the projection transfo=
rm is
> > NOT supposed to do a projection of the z-values on the screen plane=
, but
> > rather it is supposed to scale/translate the z-values to have them =
in the
> > [0-1] range (see Ri-spec 3.2 page 26). So the screen space is NOT a=
 2D coordinate
> > system.
> 
> Maybe that's the space in which DOF is supposed to be computed, then?=

> 
> Cheers,
> Andrew Bromage

I doubt. It should be computed in camera space, since you simulate
the physical properties of a real lens. In screen space, the Z coord
doesn't contain the 'real' (we're talking about virual worlds :)
depth, but you can either perform an inverse transform (one
multiplication and one addition only - cheap) or modify the CoC
computing equations so that they work with the screen space (in fact
the same as the first, but inherently a part of a single equation).

Kyosuke

______________________________________________________________________
Reklama:
P=F8edpla=BBte si Tarify nov=E9 generace na =B9est m=ECs=EDc=F9 a z=EDs=
k=E1te vol=E1n=ED o v=EDkendu
v s=EDti Oskar zdarma. 
http://ad2.seznam.cz/redir.cgi?instance=3D41039%26url=3Dhttp://www.oska=
rmobil.cz/services/whatsnew.php