[Bf-funboard] Proper soft shadows in composer

Matt Ebb matt at mke3.net
Sat Nov 3 04:01:10 CET 2007


On 03/11/2007, at 2:34 AM, toontje at telbonet.an wrote:

> I would like to see the shadow pass Z-valued. Every pixel in the  
> shadow
> area should have the information what the relative distance is from  
> the
> lightsource(s) to the object(s) that is causing the shadow. That  
> way, I
> think, you can feed this information into the defocus node. The result
> should be the same as raytraced soft shadows (sharp near object and  
> blurry
> far from the object). The other aspect is that the blurrier the shadow
> gets, the less intense it is. But that is not a problem because you  
> can
> stack alpha on the very same z-values.
>
> I wonder why this was never coded into Blender, since I think this  
> is the
> cheapest way (render time and coding) to achieve this.

It was never coded because it won't really work - it's a lot more  
complicated than that.

It's not really worth going into detail about here, but you can find  
out why, by reading up on how raytraced shadows work, but a few hints  
to consider why this isn't that feasible is:
* it's not the distance from the light source to the shaded point  
that's important, it's the distance between the shaded point and the  
shadow caster (and the relationship to coverage of the light area)  
that's important, and there's no way to get enough accurate  
information on the whereabouts of geometry in the compositor (for  
that you need, surprise, the scene geometry information and an  
acceleration structure like an octree)
* there can be multiple light sources, multiple light directions,  
light areas/shapes. This is much more info than can be stored in a  
single float channel.
* shadows do not get uniformly 'blurred' in screen space, the  
softness happens along the geometry itself relative to the light  
location and angle.
* shadows behind transparent/refracting/motion blurred objects
* and so on...

cheers

Matt


------------------------------------------
Matt Ebb . matt at mke3.net . http://mke3.net



-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.blender.org/pipermail/bf-funboard/attachments/20071103/3cdfdc32/attachment.htm 


More information about the Bf-funboard mailing list