[Radiance-general] Shadow above highlight?

[Greg Ward]

Hi John,

I spent a little time looking at your model, and there are some  
irregularities that I'm not sure are intentional.  Specifically, I  
rendered a view from the darker region of the ceiling looking down  
like so:

-------------- next part --------------
A non-text attachment was scrubbed...
Name: from_ceil_ab2.jpg
Type: image/jpeg
Size: 14446 bytes
Desc: not available
Url : http://radiance-online.org/pipermail/radiance-general/attachments/20090813/83904086/from_ceil_ab2-0001.jpg
-------------- next part --------------



(apologies for breaking my own rule about not attaching images, but  
I'm in a rush and I shrunk it down quite a bit)

The light patch is the one seen at the bottom of your original view,  
but the dark patch below it (actually towards the camera in your first  
view) is peculiar.  It appears to be an opening, where light goes but  
does not return.  Was this what you meant to do?  Are you radiosity  
calculations OK with this?  Infinite rays may cause some oddities in  
the interpolation, which I think is at least part of what we are  
seeing in the 10-20% level drop in the dark patch on the ceiling.

Setting -aa 0 reduces this problem considerably, since it turns the  
indirect interpolation off altogether, but something is noticeable  
even so.  (I suggested setting -aa 0 in my first reply by the way -- I  
thought you had tried that.)  The only way to get rid of the dip  
entirely is to set -as 0 as well.  You may have to increase -ad at the  
same time.  Let's explain why this is so...

The -as samples are sent only to those places where Radiance  
determines there is significant variance.  If you don't send out  
enough initial samples, then the additional samples may not go where  
they need to.  The polar region is doubly-cursed with respect to the - 
as setting, because the sampling density is low at the zenith AND  
there is no check for "over the top" differences.  Thus, the zenith  
region tends to be underestimated if all the variance is there.

-Greg

> From: schwaj3 at rpi.edu
> Date: August 13, 2009 4:05:22 PM PDT
>
> Wow, reducing the -as to 0 seems to have relieved the problem. I had  
> no
> idea that they don't shoot directly up (I had guessed it might have  
> been
> something along those lines, but couldn't track it down).
>
> The RwR book seems to suggest that you use a -as equal to 1/4 your - 
> ad,
> so I can't imagine that reducing the -as to 0 should be the correct
> solution... Perhaps the model I am using simply exacerbates this  
> issue?
>
> Could you explain when it's reasonable to use -as 0 and a large -ad,  
> as
> opposed to the values recommended in the book? Does it depend on the  
> size
> of the highlight on the floor (smaller = more/less -as  )?
>
> Naturally like you suggested Andrew, this model is much faster to do  
> with
> radiosity. It's really just a test model, and naturally more complex
> models would have to be used. However after I saw this, I naturally
> couldn't proceed with image comparisons until I figured out what I was
> doing wrong! As you've probably seen, I've sent out the scene, so if  
> you
> have any further insights, especially into what rpict options I  
> should be
> using, do tell.
>
> --John Schwartz
>
> ==============Original message text===============
> On Thu, 13 Aug 2009 18:01:09 EDT Andrew McNeil wrote:
>
> John,
>
> I'd also like to see your scene, but while I wait I'll go out on a  
> limb:
> AFAIK in radiance, an ambient sample ray is never emitted normal to a
> surface.  I suspect your shadow results from the (slightly) decreased
> probability of sampling the highlight from the area directly above the
> highlight.  This effect can be compounded by the -as setting (which  
> never
> samples the upper edge of a division, but can sample the lower edge  
> of a
> division).
>
> My suggestion: try a high ad value coupled with no ambient super  
> samples.
> -ad 20000 -as 0
>
>
> As far as your comparison between raytracing and radiosity, You pick  
> a scene
> that plays to all the strengths of radiosity.  Radiosity with a  
> dense mesh
> produces a highly accurate result in relatively little time for  
> scenes that
> have very simple geometry and are limited to completely diffuse  
> reflection
> and completely specular transmission.
>
> If you had semi-specular reflection, semi-diffuse transmission or  
> complex
> geometry raytracing is more likely to result in higher accuracy in  
> less time
> than radiosity.
>
> Andy