[Radiance-general] What's causing these artifacts???

[Greg Ward]

Hi Marcus,

There's rather a lot to respond to here, but I'll give it a go...

> From: "Marcus Jacobs" <marcdevon at hotmail.com>
> Date: December 23, 2005 10:25:01 AM PST
>
> Dear Group
>
> Thanks for all of the suggestions regarding my rendering problems.  
> I know its taken me a while to reply back about this issue to the  
> group. I have spent some time investigating these issues that I am  
> having. This way, I at least try to solve my problems myself before  
> bringing to this group.
>
> First, the dark splotches:
>
> http://i35.photobucket.com/albums/d178/marcdevon/ 
> AFrame2_Camera6_final.jpg
> http://i35.photobucket.com/albums/d178/marcdevon/ 
> AFrame2_Camera03_final.jpg
>
> I think I have some idea as to what is causing this issue.  
> Generally when I render, I perform a small overture calculation at  
> a 100 x 100 pixel resolution. The ambient settings are generally at  
> -aa 0.1 and -ar 0. The next step is to take a medium resolution  
> overture usually at 500 x 500 pixels. The settings are generally  
> set at -aa (.15 - .2) -ar = maximum scene size as determined by  
> getinfo -d in inches (basically a 1 inch grid).  The final render  
> is performed at -aa (.25 - .3) and -ar = maximum scene size as  
> determined by getinfo -d in inches/(2 to 10) (basically a 2 - 10  
> inch grid) depending on the illumination contrast. The -ad and -as  
> settings are generally set based on the illumination contrast in  
> the scene. For these renderings, I set -ar very low since I had  
> already rendered it at the same resolution with higher -ar  
> settings. Essentially, I wanted to force it to interpolate using  
> the existing ambient data. The splotches usually come in as the -ar  
> parameter degrades. The number of splotches depends on how low this  
> parameter has been set. The reason why I ask is because this seems  
> to me to have started occuring recently even though I have used the  
> aforementioned technique for some time now.

Your analysis makes sense.  As noted in discussions earlier this  
week, it is probably a bad idea to change the -ar setting between  
runs.  This is most likely the source of your problems.

> For the bright splotches here:
>
> http://photobucket.com/albums/d178/marcdevon/ 
> th_room2a_view2_render1.jpg
>
> it looks like mkillum saves the day again. I tried Greg's  
> suggestion of setting an illum at skylight openings at the surface  
> level of the ceiling. This seems to have done the trick. It's funny  
> but I ran dozens upon dozens of renderings and for some reason I  
> thought these artifact were due to the bright spots on the floor  
> rather than the skylight openings. Here are the results:
>
> http://i35.photobucket.com/albums/d178/marcdevon/room2_view3.jpg
> http://i35.photobucket.com/albums/d178/marcdevon/room2_view2.jpg
>
> May I ask, why is the brightness of the skylight openings more  
> prone to causing these artifacts rather than a bright spot on the  
> floor or wall?

It's difficult to say for certain, but it probably has to do with the  
solid angles visible from the problem areas.  Also, if the wall  
surface on the coves of your skylight are smooth and specular, this  
could greatly amplify their indirect contribution.  I believe  
specularities are at play in some of the artifacts you mention below.

> When initially attempting to find a solution to this problem, I  
> tried to go back to issues that Rob Guglielmetti  was experiencing  
> a couple of years back with a museum that he was modeling. At one  
> time he documented his progression on his blog but the page has  
> since been deleted. I did remember a couple of threads from July/ 
> 2004 which discussed whether to place illums inside or outside the  
> scene:
>
> http://www.radiance-online.org/pipermail/radiance-general/2004-July/ 
> 001871.html
> http://www.radiance-online.org/pipermail/radiance-general/2004-July/ 
> 001953.html
>
> The reason why I have been hesitant to using illums inside is  
> because I thought that mkillum could only simulate illumination  
> that's a smooth function with no discontinuities. I thought that  
> this may be an issue when windows have lites. I did a little  
> experiment with another old scene of mine. It is a window that I  
> knew would receive direct illumination from the sun. I wanted to  
> place the illums on the inside to see if mkillum would pick up the  
> lites. Here is the result:
>
> http://i35.photobucket.com/albums/d178/marcdevon/room2_camera1_tm.jpg
>
> I am impressed that mkillum has able to detect the presence of  
> lites in the windows as you can see their occluding effect by the  
> direct sunlight that is on the floor.  By placing the illums on the  
> inside of this room, I was also able to eliminate some bright  
> splotches (although less conspicuous than the previous scene) from  
> this room.

I am a little confused by the term "lite," which my dictionary  
defines as "beer with relatively few calories."  I assume you mean  
the mullions.  Bear in mind that mkillum does not attempt to model  
light sources already in the scene, such as the sun, and this light  
passes without intervention.  Mkillum only takes over for ambient  
contributions and glow's, such as the sky.

> In my investigation of course there are more questions/issues that  
> has arisen.  From another test room, I wanted to see how much the  
> smoothness the indirect calculation would improve by placing an  
> illum inside its skylight. This is what's happened:
>
> http://i35.photobucket.com/albums/d178/marcdevon/ 
> room1_12_amb_view4.jpg
> http://i35.photobucket.com/albums/d178/marcdevon/ 
> room1_12_amb_view5.jpg
> http://i35.photobucket.com/albums/d178/marcdevon/ 
> room1_12_amb_view6.jpg
> http://i35.photobucket.com/albums/d178/marcdevon/ 
> room1_12_amb_view8.jpg
>
> The indirect illumination in the corner I do not believe is  
> correct. I know that this problem isn't due the presence of the  
> illum in the skylight at the surface level of the vaulted ceiling.  
> This seems to occur sort of haphazardly. Here is the same room with  
> the same day/time settings, albeit with different materials, and  
> this problem didn't occur:
>
> http://i35.photobucket.com/albums/d178/marcdevon/ 
> room1_b_view4_render2.jpg
> http://i35.photobucket.com/albums/d178/marcdevon/room1_b_view5_2.jpg
> http://i35.photobucket.com/albums/d178/marcdevon/room1_b_view6.jpg

What are the specularities and roughness values of your floor  
surfaces?  And which version of Radiance are you running?  This looks  
like it could be a bug, or it could just be mirror-like specular  
reflections off your floor.  I suspect you have a low roughness value  
-- do you?

> This problem that I am receiving looks similar to issues that I am  
> having with this rendering (right side near the corner):
>
> http://i35.photobucket.com/albums/d178/marcdevon/ 
> KountryKitchen_view2_render3_3.jpg

This looks like mirror reflections off your counter top, which could  
be real but are poorly rendered by the indirect calculation.  (You  
could use the "mirror" type to correct this, though I'm not sure it's  
worth it.)

> Also, for the previous rendering and the next, I do not know what's  
> causing the gray artifacts on the floor nearby the dishwasher and  
> lower cabinets:
>
> http://i35.photobucket.com/albums/d178/marcdevon/ 
> KountryKitchen_view1_render3_99.jpg

These artifacts are more worrisome.  I'm not sure what's causing  
them, unless it's more of the low-angle specular reflections off your  
skylight cove.  Are these surfaces purely diffuse, or do they have a  
small specular component that might get blown out of proportion at  
low angles?  (This is a problem that was discussed earlier this month  
under the heading "How to increase the rendering speed?" and fixed in  
the latest HEAD release.)

> Generally, what I wish to be able to acquire is achieving a higher  
> level of predictability so that I may take a scientific approach to  
> rendering rather than constantly tinkering. Sometimes when I think  
> I got it, I later realize that I don't. That's pretty much it in a  
> nutshell.

A worthy goal.  I think we can all benefit from what we learn in this  
exploration, and I appreciate your sharing your results.

Happy Holidays!
-Greg