Using the Illum Material for Smoother Renderings in Radiance

Header Image

IES lights, my first time

The following images are the result of playing around with the ambient parameters in Radiance, to try and eliminate the "blotchy" rendering artifacts so commonly experienced by novice users of this software (me).

Problem Statement This is a fairly detailed model of a gallery space, with a coffered ceiling which will house a multitude of T5 striplights, concealed behind a metal baffle. The idea is that light spills out from the slots between the coffer's edges and the metal baffle. The model was created with ACIS solids; therefore it was exported to 3DS format, back into AutoCAD, then converted to Radiance scene descriptions with Georg's Radout utility.

Initially the entire space was filled with lights, but when this rendering artifact problem was discovered I eliminated all but one 4' strip light so that the test renders would complete faster.

Even after setting QUALITY=H, the blotchy patches on the walls remained. About that same time, Greg Ward replied to a question on the radiance-online.org mailing list that was very similar to the problems I was having, so I tried to apply some of those techniques to this rendering. Read this, then come back here. I'll wait.

Back so soon? OK, so I tried Greg's tip of dividing the octree size by the smallest important detail, to better judge the -ar parameter. Then I tried increasing -ar again, then -ad, then -as (after realizing that should have been set to 1/2 the -ad in the first place). In the interest of speed, I also changed QUALITY=M, but increased DETAIL and VARIABILITY to H, for the latter runs.

The results are shown below, with the settings used for each:

First Attempt: Quality=H Detail=M Variability=M render= -ar 500

First Attempt: Quality=H Detail=M Variability=M render= -ar 500

Second Attempt: Quality=M Detail=H Variability=H render= -ar 700

Second Attempt: Quality=M Detail=H Variability=H render= -ar 700

Third Attempt: Quality=M Detail=H Variability=H render= -ar 700 -ad 2000

Third Attempt: Quality=M Detail=H Variability=H render= -ar 700 -ad 2000

Fourth Attempt: Quality=M Detail=H Variability=H render= -ar 700 -ad 2000 -as 1000

Fourth Attempt: Quality=M Detail=H Variability=H render= -ar 700 -ad 2000 -as 1000

As you can see, the blotchies get better, but still remain (P.S. it should go without mentioning, the renderings were taking progressively longer and longer to complete /rpg). Am I looking in the wrong place for the source of my problems here? Is the model itself a problem? In the parameter troubleshooting information I've seen, blotchy renderings generally are the result of bad ambient settings, but I can't seem to "smooth out" these images. The exposure is not being set, but there are two ambient bounces (and three in the case of image 1) which I would have thought would minimize any problems with that.

I'm hoping I can get some clues from you all, since these parameters don't seem to help fully and these renderings take a couple of hours to produce now, with only the single light. I'd like to be able to produce a high quality image with the full lighting layout in it, which is likely to contain a couple hundred light sources, so if there's a less expensive way to achieve a better result, I'm all ears!
-Rob Guglielmetti


Progress Well, after several quick replies from some of the list gurus (and lunch), I was able to try some of the suggestions. Clearly, the direct calculation is the way to go, and mkillum is an amazing tool to help you use it more often.

Fifth Attempt: (now, we're using the illum primitive, finally) Quality=M Detail=H Variability=H

Fifth Attempt: Quality=M Detail=H Variability=H

While the "blotchies" are gone, and an amazing amount of detail has bubbled to the surface even at the lower rendering parameters (note the skinny cables the art is hanging from), there seem to be some new artifacts. Note how the floor beyond a certain radius of the light fixture is mottled. There are also a couple of strange irradiance boundaries on both of the side walls. I guess it's time to study the direct calculation parameters...


This process of creating a box to "catch" the light emitted from the coffer may still be in error. Refer to the section below for the geometry of the problem:

Coffer Section (1)

From Greg & Jack's description, I need to create a box that defines the sides and bottom of the coffer, and make it tight to the aperture. What I did was this (illum surfaces depicted in red; arrows indicate surface normals):

Coffer Section (2)

... which in retrospect, seems wrong. I think they meant for me to do this:

Coffer Section (3)

Note how all the illum surfaces are below the ceiling plane and point into the space. But for that matter, why couldn't I simply do this:

Coffer Section (4)

In this example, there is just one illum surface "plugging up" aperture. Is there a minimum distance the illum needs to be from the geometry above that blocks the light in order for this to be accurate? Is that why I was told to drop everything below the ceiling plane?

As always, insights appreciated. This has been educational.


A breakthrough (of sorts) More answers found their way into my inbox, so I tried anew:

I reconfigured my illum surfaces to drop below the ceiling plane, and made sure they face into the room, no overlapping surfaces, yadda yadda yadda. Massive improvement. This image took about a minute to compute:

Getting There:

Getting There!

Now, the glow is obviously very much in error (there are 15 40W biax lamps up there; the coffers should be totally washed out), but it was my first time guessing how to accurately set one of those "falsie fixtures" up. Next I will use the lampcolor utility to come up with a better guess.

More disturbing are the strange glowing surfaces exactly where my illums are, in certain places. I double checked, and there is only one surface for each illum face (I used "option three" from the diagrams above, in the previous discussion). Hmmm....

Interested in seeing what the whole room would look like, I quickly arrayed my illums around all the coffers and re-ran the sim:

Getting There(er):

Getting There(er)

Note that I did not array the glows, so the coffers appear dark; I was simply interested in seeing the room lit at this point. I also decreased the DETAIL parameter to medium, in the interest of time, which explains the missing parts of the artwork cables, and the rougher door moldings. I was pleased to see that this quality image only took about 12 mins! All hail the direct calculation!

Every breakthrough yields more questions. Here are some:

  • What the heck are those bright artifacts near the coffers?
  • For the glow parameters, I'm wondering what the best way to define the surface is. A 40W biax lamp, the PL-L envelope, is essentially two 22" long T5 tubes.
  • When using lampcolor, should I say it's a 44" long 5/8" diameter cylinder, or a 2"x22" polygon (which is the way it's been modelled)?

A bigger breakthrough Now we're getting somewhere! After using a square test room to sort out some problems with the IES files I was using, I got a decent looking coffer assembly and arrayed it around the space. These renderings were completed in mere minutes, although I had the DETAIL and VARIABILITY parameters on low. These images were also tonemapped with pcond -h:

Tonemapped Image

Tonemapped Image

The room looks great, but the coffers do not. I replaced the lights after the mkillum run with glow polygons, guesstimating 170 for the luminance value (it's a 40W Biax lamp). They looked pretty bad with the glow radius set to 5, so I increased it to 12, which is what is was for the images above. I'm guessing that maybe increasing the DETAIL & VARIABILITY parameters will help smooth them out; those are test rendering right now.

So, I'm not quite satisfied yet, but thanks to the Radiance-online list gurus, I'm a whole lot happier with these renderings than I was a few days ago. Thanks everyone!


Is that all there is?

So close and yet so far! Arggh!

Well, I completely redid my model (after much experimentation, I find things get really fragmented and the directory becomes quite a mess, so now that I had a method for the illums I thought I'd start anew), and the room is looking even better still. I even played a bit with picture.cal and was able to get Edward Hopper up on the wall!:

Hopper

But, alas, as you can see my coffer still looks like crap. I have a handful of glow polygons up there, and I had hoped that they would illuminate the inside of the coffer properly, but they seem to leak, or something. Maybe a different source shape will help? (as I'm typing this, the ever-generous Greg Ward emails me a tip: try excluding the coffer materials from the ambient calculation. I will try.) Just to clarify, I am using the "inverted box" option for the ilums, because some of the coffer geometry drops below the ceiling plane. Here is a more accurate section of the coffer and my illum configuration:

Coffer Section (5)

So, I'm close.


Yippeee! Success. I got tired of looking at those artifacts around the coffers, and simply changed the geometry so that everything was above the ceiling plane. This allowed me to use a single polygon to close up the hole, and it seems to have done wonders. I even threw a 150W PAR38 spot on the Hopper painting, just for laughs.

These renderings were created with rad, using Q=M D=H V=M, and two ambient bounces (-ab 2). The first rendering took about 30 minutes, including the overture calculation. The second one only took about two minutes since there was a lot of overlap with the first, so the ambient cache was used heavily.

I did not exclude any materials from the ambient calculation, so the coffer appearance is all brute force ambient hunting for the glows. It actually doesn't look all that bad/off.

So, this was very informative, and thanks again to everyone who helped me with this problem. I'll see some of you next week; beers all around, and a root beer for Greg. =8-) I have attached the latest renderings below.

Raw Images


Tonemapped Images


Falsecolor Illuminance images


Well, I'm glad we got that sorted out.
/rpg