RE: [Algorithms] State of Real-Time Depth-Of-Field?

[Rowan W. <ro...@ir...>]

I am yet to hear of anyone coming up with an acceptible solution for DOF
and translucent objects (which dont write into the depth buffer).
=20
rowan


  _____ =20

	From: Wolfgang Engel [mailto:wengel@RockstarSanDiego.com]=20
	Sent: Thursday, 17 November 2005 8:50 AM
	To: gda...@li...
	Subject: RE: [Algorithms] State of Real-Time Depth-Of-Field?
=09
=09
	Hi,
	I am quite happy with the approach from Thorsten Scheuermann in
ShaderX3. It is easy to understand and to use and you can change all the
things that you do not like pretty intuitively. For example you can go
back to a simpler Gauss blur or try out a few other things to get rid of
color leaking.
	In some games you want to remove the near blur plane.
	=20
	One of the general problems of each DOF algorithm is, that they
only work well in scenes, where you can estimate the viewing distance.
If the viewing distance varies a lot, you need to think about an
"auto-focus" feature. I think this will be hard to make work well. So
from my current point of view DOF is pretty good for scripted scenes and
scenes where you know that the viewing distance won't pass a certain
amount.
	=20
	My 2 cent,
	- Wolf

  _____ =20

	From: gda...@li...
[mailto:gda...@li...] On Behalf Of
Aaron Lefohn
	Sent: Wednesday, November 16, 2005 12:29 PM
	To: gda...@li...
	Subject: [Algorithms] State of Real-Time Depth-Of-Field?
=09
=09
	Hi All,
	   I'm currently working on a new interactive depth-of-field
algorithm (will be submitted to Siggraph 2006) and am looking for
comment on the usefulness/problems with current interactive DOF
techniques. My main question is:=20

		Is the quality of the current DOF techniques "good
enough?"=20
		  - Examples include ATI's variable-width separable
gaussian blur (ShaderX2) and the mipmap techniques summarized in GPU
Gems, Ch.23.=20
		  - Is color bleeding across depth discontinuities still
a problem or do the various fixes remove all objectionable artifacts?

	Please note that I am not digging for solutions, but rather your
thoughts about existing DOF algorithms.
	=20
	Thanks,
	Aaron
	http://graphics.cs.ucdavis.edu/~lefohn/