"Kenneth" <kdw### [at] gmailcom> wrote:
> Looking through the "rad_def" include file, I see that all of the radiosity
> entries there are missing a  maximum_reuse  value-- which means that it defaults
> to 0.2 in all of the cases. That's a rather large value, and can produce some
> large-sized splotches. (From the docs' explanations of the various rad settings,
> and possibly from code examples I've seen there(?), the 'typical' spread between
> minimun_reuse and maximum_reuse should(?) be approximately 1:13; I guess that's
> a trade-off between quality and render speed. And the spread is also designed so
> that large flat surfaces get the maximum value-- large smooth rad patches--
> whereas curved surfaces or other 'non-flat' areas get smaller patches.
>
> Personally, I now like to use a *small* spread between min and max_reuse; it
> kind of defeats the purpose of the render-speed trade-off, and produces patches
> that are all of the same (smaller) size; but I like the higher-quality look,
> especially if rad's count value is greatly increased to produce more patches.
>
> Anyway...
>
> As an experiment when using, say, the "rad_def" Radiosity_Fast entry (which has
> a minimum_reuse value of 0.025), add a maximum_reuse there of 0.026. And greatly
> increase the 'count' value. It will be a slow render, so maybe pick out a
> smaller section of the image to test (and turn off AA). I can't predict what the
> results will be, but this *might* improve the results.
>
> BTW: In your image post, it looks like some of the objects' colors might be
> exceeding <1.0,1.0,1.0>, perhaps due to the combination of lights used. This
> could be a reason for some of the rad artifacts. Try decreasing the objects'
> diffuse values a bit, so that nothing in the scene produces a max color
> brightness exceeding 1.0. It will probably be a guessing game of trial and error
> ;-)
I have started doing some of the trial and error and have noted a few things as
I start to learn how all of the available levers are related and affect the
image.

Smaller error bounds and larger counts don't necessarily reduce the black
artifacts that show up when using radiosity.  In fact, in most of my
experiments, the artifacts disappear with an error_bound of 1 or higher and a
very small count.  That, of course, can lead to blotchy surfaces, missing
shadows, etc., so most of my work now has been trying to figure out what else I
can adjust to get a decent error_bound/count combination while also eliminating
the artifacts.

Since most of the artifacts show up on my isosurfaces, and on very thin
cylinders, I am wondering if the extra radiosity rays are finding isosurface
"holes", similar to what happens when max_gradient is too small.  The problem is
that I am not seeing any max_gradient warnings.  When I render without
radiosity, the max_gradient seems to be fine and I get no warnings.  When I bump
them up significantly for radiosity renderings, the warnings I get are that my
max_gradients are too high, yet I see the black splotches.

I am also seeing cases where the isosurface appears to be somewhat transparent
when rendered with radiosity, while it appears solid without.  I had thought
this might be an error_bound issue, but even reducing error_bound to 0.06 or
smaller doesn't help.

I'll keep fumbling until I really understand how it all works.

-- Chris R.

Post a reply to this message