schlosser at st.cs.uni-saarland.de
Wed Jan 7 19:42:18 CET 2015
in the 'Distribution Raytracing' slides it says we should scale the
light sources intensity by incident cosine and the area of the
source-solid. As far as I understood it we achieve the behaviour of an
area light by simulating N pointlights each having an intensity divided
by N. The integral in the area light equation then gives us N=source area.
I don't think the cosine scaling is done at all in the reference picture
because the top of the walls should be way darker (Also we cannot easily
reach the sources normal vector in the given structure).
So what I ended up doing is dividing the intensity by (distance^2 /
area), mind the paranthesis because the scene is very 'small' and both
distance and area are below 1.
This achieves a very similar image to the reference, I do not know if
this is correct though.
On 01/07/2015 06:05 PM, Gereon Fox wrote:
> Discussing the issue with a fellow student, I realised that I had
> forgotten to mention the following: Before we apply said division, we
> clamp the radiance we got from the texture. This is what makes our
> approach especially suspicious, because the texture's radiance of 25.f
> seems to have been chosen carefully and we just discard every
> information beyond 1.f through the clamp...
> Am 06.01.2015 um 01:47 schrieb Gereon Fox:
>> we are having some trouble with the AreaLight: The constant
>> texture/material attached to it in the sample scene emits a radiance of
>> 25.0. Obviously, simply returning this radiance in
>> AreaLight::getIntensity makes the scene much too bright. That is why we
>> tried to follow all the clues we could find and divided that radicane by
>> the (sparsely convincing) divisor pi * irr.distance^2. This made our
>> images look quite similar to the samples, but the bright half-circle on
>> the white back wall is now a little too small (see attachment).
>> What are we doing wrong here?
> Cg1 mailing list
> Cg1 at cg.uni-saarland.de
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Cg1