[Radiance-general] windows: illum useage?

[Greg Ward]

Hi Lars,

I recommend your "choice a)", and suggest you first try creating a 
rectangle that just covers the opening from the inside, then use 
mkillum to average it's output.  Keep your gensky command in the scene 
(let "rad" automate things for you if you like), whether you want to 
look out the window or not.  The largest source of error from applying 
skyfunc directly to fenestration is the lack of the building shadow on 
the ground plane, leading to overestimation of the upward component 
from the window in cases where the sun is on the other side of the 
building.  Applying mkillum avoids this error, besides accounting for 
the framing of the window panes and the circular shape of the opening.

-Greg

> From: "Lars O. Grobe" <grobe at gmx.net>
> Date: July 1, 2004 3:50:01 AM PDT
>
> Hi list!
>
> Once again I have to ask a rather basic question, still I could not 
> find an answer so far.
>
> I have a model of a building which is illuminated solely trough 
> windows consisting of a grid of small rectangular glass panes in stone 
> framing, the whole window has an arc. So far I didn't model the panes, 
> and the light just came in through the frames, which gave not the best 
> results in the rendering, but I didn't care so far. Now that I will 
> start modeling the panes, I have to think about how to use the illums 
> later. As I don't want to keep the artefacts caused by the small 
> openings in my final rendering, and as I want to speed up rpict, I 
> need to use illums.
>
> - The first question is how to define the illum geometry.
>
> a) I could define one polygonal surface covering the whole window from 
> the inside, and using mkillum to "see" the effect of the frame which 
> is "outside" than. The pro is, that I have to model just two surfaces 
> per window (one for the illum, and one glass pane which could 
> intersect with the frame), and I have few light sources in the scene 
> (one per window). The negative point is that the mkillum is necessary, 
> although I am not really interested in the outside, so I get some 
> overhead. It will also be not to easy to define a polygon for the 
> surface, as the round outline is already converted into a polygon from 
> the cad, and both have to fit exactly. However, this is what I would 
> conclude from the advices in Rendering with Radiance (complex 
> fenestration).
>
> b) I could model each pane, place it in its correct location, and than 
> simply put the gensky pattern on it. No mkillum needed any more (I 
> actually don't want to look through the windows). The drawback is that 
> I get a lot of small lightsources (around 40 per window), and I have 
> to model lots of panes... don't laugh, modeling a rectangular surface 
> is not that difficult, but I have to make sure that the surface 
> normals point to the inside, and that is not that easy - I found some 
> problems in formZ regarding normals and dxf export.
>
> So which way to go? As I noted, I have also have to think about 
> rendering speed and memory - the model is large, and I am supposed to 
> produce animations later. And I found the following on the list, which 
> makes me tend to the b) way:
>
> > Illum's, like all light sources in Radiance, are preferred as 
> rectangles or (at least)
> > 4 or more sided convex shapes.  Since Radiance samples a rectangle 
> with
> > equivalent area to the source shape, triangles are a particularly 
> poor choice.
> > I recommend replacing any windows you plan to use as illum's with 
> rectangles
> > for that reason. (Greg Ward)
>
> - A second question, would it make sense to remove the gensky output 
> except the sun after the illums have been generated if I use procedure 
> a)? If all openings are covered with illums, I don't need the diffuse 
> sky and ground any more, as they will be blocked by the illums, right? 
> Could that give some additional performance? This could also reduce 
> errors caused by "leakage" in the model.
>
> TIA+CU Lars.
> --
> Lars O. Grobe
> grobe at gmx.net