Sorry Bob, maybe I wasn't clear about what I perceive as the bug.

With recursion_limit 1 the scene gives exactly what I would expect to
see - bright areas on the surfaces by the high ambient ball, intensity
falling off in a roughly circular fashion, smoother with higher count
and more patchy with lower error_bound. I would expect to see something
similar with a higher recursion_limit.

With recursion_limit 2 or more the bright patches are no longer
circular, they are clear (axis aligned?) bands on the surfaces. You even
get a brighter area where the bands cross. These bands become more
distinct as you lower the error_bound and get less patchy as you
increase the count.

You can see these bands on the radiosity passes with your settings too -
the very high error_bound on the final pass is simply smearing them
across the surfaces, the underlying problem is still there.

This seems to indicate to me that there is a sampling problem within the
radiosity code.

After looking through the MegaPov radiosity code, the only odd thing
that I can see is that exactly the same sequence of radiosity samples is
taken each time. With the way the local (axis aligned?) basis vectors
are set up, could this be producing up a sampling bias? I'm sorry, but
this is about where my brain starts to gibber and finds a dark corner to
hide in ...

I hope this clarifies my point a little.

Bye for now,
	Mike Andrews.

"Bob H." wrote:
> Not a bug, just bad settings; and possibly scene setting too considering
> there's a large black emptiness behind the camera.
> 
> Try the following:
> 
>   radiosity {
>     pretrace_start 8/image_width
>     pretrace_end   4/image_width
>     count 123  // does not need to be as high now
>     error_bound 1.6  // higher smoother
>     low_error_factor .04 // lower smoother
>     nearest_count 3 // lower faster, less quality
>     max_sample 4 // ?? higher better ??
>     recursion_limit 2 // 3 is really slow, 1 is dimmer
>   }
> 
> It only seems there's a problem when you muck around with the settings and
> get bad ones.  Not easy to get right, which is why leaving all out and
> letting defaults be used generally looks okay.  But certainly the recursion
> limit isn't faulty.
> 
> Bob H.

Post a reply to this message