What I thought would be a straightforward night-time scene has turned
out not to be. The first two images at the link below shows a
conventional scene with the lights replmarks-ed in. All the lights are
based the same fitting, a Thorn Contrast C (33000lm).
http://uk.f1.pg.photos.yahoo.com/ph/bsounes/lst?.dir=/ambient_mess&.view
=t
After doing a few test renders with just one light, a wall and a ground
plane I realized the splotchy areas beyond the site where the result of
the ambient calculation picking up the bright wall areas just below the
fittings. No extra ab's, ad's ,as's, extra ar or setting aa to 0 solved
the problem. The lamp is modeled as a 10mm sphere therefore source
sub-structuring didn't seem to be the issue either.
This suggested to me using mkillum to eliminate the gamble of the first
ambient bounce finding the bright surfaces: feed the brightest surfaces
to mkillum while retaining the original surface modifiers for the final
render. This involved switching to AutoCAD to 'radout' the polygons
(previously I was using .obj files), a degree of re-modelling , checking
the surface normals and splitting the model in two. The result (third
image: 'AlternativeA_pcond') is still less than satisfactory.
I experimented putting a void illum polygon a few millimeters in front
of the wall. A single bright splotch still appeared. There also seemed
to be a slightly different light distribution, which made me question
using opaque surfaces as illum sources.
An illum 'fence' could be spaced a distance from each building to try
and catch the splotches before they appear, but I haven't tried that
yet.
I am interested primarily in the final appearance of the image. The
illuminance beyond the site boundary is quite low - below 3 lux.
Does anyone have any suggestions?
Bruce Sounes
Hi Bruce,
Interesting problem. I presume that you don't see this unless you run the image through pcond, which makes the darker areas visible. Did you try applying the -h option of pcond? This might reduce the appearance of these splotches considerably.
Getting at the heart of your errors, they are really due to the low-angle light coming in from the horizon on the outer perimiter of your model. Your idea of placing a "mkillum fence" around the perimiter is a good one, and might well eliminate these splotches. Make your polygons out of "void" and completely surround your building with a fence as high as the top of the roof. You might get another artifact from this, though, which will be the unexpected appearance of a bright line right near the mkillum fence, even though the fence itself will be otherwise invisible. I'm not sure how to avoid this.
Do you really want to see the spill light on the surroundings? Typically, moonlight would drown this out. If you added a moon, you might not even notice your splotches anymore. You could also exclude the surrounding material using the -ae option. You won't get the spill light anymore, but the splotches would be gone with it.
-Greg
Gidday Bruce,
Ok this is bit of a punt - and hopefully it won't end up looking like a dog's breakfast.
if you don't mind flat shading of the surrounding surfaces and loosing the light spill you could treat them as glow surfaces (with low RGBs to produce something equivalent to the slightly darker areas of the land or sea) and set the 4th parameter maxrad -ve so they don't contribute to scene illumination.
Effectively it should be a way of killing off the ambient calculation in those areas while still producing at least the presence of a coloured surface. Vertical surfaces might need their own lower glow definition to make a contrast.
The abstractness might help indicate the 'extent' of the lighting rendering, but if the flatness of the shading ends up being an aesthetic problem, then a Brightfunc could be written to modify the glows and slowly fade them out with distance from the project.
hope this helps
cheers
alex
···
*******************************************************
A. J. Summerfield [email protected]
Faculty of Architecture, University of Sydney
Outside of a dog, a book is man's best friend.
Inside a dog, it's too dark to read. Groucho Marx
*******************************************************
On 12 Nov 2003, at 2:53 pm, Bruce Sounes wrote:
What I thought would be a straightforward night-time scene has turned out not to be. The first two images at the link below shows a conventional scene with the lights replmarks-ed in. All the lights are based the same fitting, a Thorn Contrast C (33000lm).
http://uk.f1.pg.photos.yahoo.com/ph/bsounes/lst?.dir=/ambient_mess&.view=t
After doing a few test renders with just one light, a wall and a ground plane I realized the splotchy areas beyond the site where the result of the ambient calculation picking up the bright wall areas just below the fittings. No extra ab�s, ad�s ,as�s, extra ar or setting aa to 0 solved the problem. The lamp is modeled as a 10mm sphere therefore source sub-structuring didn�t seem tobe the issue either.
This suggested to me using mkillum to eliminate the gamble of the first ambient bounce finding the bright surfaces: feed the brightest surfaces to mkillum while retaining the original surface modifiers for the final render. This involved switching to AutoCAD to �radout� the polygons (previously I was using .obj files), a degree of re-modelling , checking the surface normals and splitting the model in two. The result (third image: �AlternativeA_pcond�) is still less than satisfactory.
I experimented putting a void illum polygon a few millimeters in front of the wall. A single bright splotch still appeared. There also seemed to be a slightly different light distribution, which made me question using opaque surfaces as illum sources.
An illum �fence� could be spaced a distance from each building to try and catch the splotches before they appear, but I haven�t tried that yet.
I am interested primarily in the final appearance of the image. The illuminance beyond the site boundary is quite low - below 3 lux.
Does anyone have any suggestions?
Bruce Sounes