[Bf-committers] Re: Re: [Bf-blender-cvs] Re: CVS commit:
blender/source/blender/blenkernel BKE_node.h
blender/source/blender/blenkernel/intern ipo.c node.c node_composite.c
blender/source/blender/makesdna DNA_camera_types.h
DNA_node_types.h blender/source/blender/render
Michael Crawford
psyborgue at mac.com
Sat Dec 23 23:34:25 CET 2006
On Dec 23, 2006, at 5:08 PM, Alfredo de Greef wrote:
> --- GSR <gsr.b3d at infernal-iceberg.com> wrote:
>> Those are the cases in which having something beyond
>> plain blur really
>> matters.
>> The linked image is from a demo page of a
>> real DSLR.
>>
> http://www.pentax.co.uk/images/products/photo/uk/K10D/Image%204.JPG
>>
In these cases I've found using a seperate layer for the background
works great. Isolating elements in their own layers can solve a lot
of the mentioned "bleeding" problems (which i don't think can really
be avoided otherwise).
>> Or you can use mouse to get the focus depth. ;]
>> Click in any render
>> and look for the printed Z value.
Easy way here:
http://homepage.mac.com/psyborgue/focalplane.blend
Now i have a suggestion.
Foreground objects isolated in their own layers end up with a black
halo after a blur (from the adjacent pixels). In order to get a
correct result those black pixels must be changed to the *blurred*
background prior to the foreground element's blur (keeping the same
alpha and Z of course). This can be done manually but i propose a
"layered DoF" node which automagically blurs elements from background
to foreground based in order of acending ZDepth (or perhaps by layer
order). Such a node would have a variable number of inputs and
function as a Z combine node as well. This node would likely be the
last step prior to output. Assuming there is a way, perhaps the node
could accept "multilayer" input as well. If this was done, however,
it would likely require a seperate "multilayer combiner" node. Such
a setup would avoid the "bleeding" edges problem where a foreground
element was is away from the background.
More information about the Bf-committers
mailing list