~s Radiance Digest, v2n8 (Part 2)
Dear Radiance Users,

This is the second part of the Radiance Digest (v2n8), and it contains
only one topic.  If you got it before the first part, it may be a quirk
or it may be that your mailer rejected the longer message.  If you suspect
the latter, send me e-mail asking me to resend it to you.

As usual, please do not respond to this message.  If you have something
to tell me, write to "GJWard@lbl.gov".  If you wish to have your name
removed from the list, or someone else's name added, send e-mail to:

	radiance-request@hobbes.lbl.gov

With some human-understandable message like:

	Please take my name off the Radiance Digest list.

I receive these messages, and a single word like "subscribe" means nothing,
since there are two lists, a moderated digest list (the one you are certainly
on) and an unmoderated discussion list (radiance-discuss@hobbes.lbl.gov).

-Greg

=======================================================================
IRRADIANCE GRADIENT CALCULATION

From: hwj@henrik.igk.dth.dk
Subject: Irrad. grad. and BRDF
To: greg@hobbes.lbl.gov
Date: Thu, 17 Nov 1994 15:15:52 +0100 (GMT+0100)

                                       Henrik Wann Jensen
                                       Institute of Graphical Communication
                                       Technical University of Denmark
                                       Building 116
                                       2800 Lyngby
                                       Denmark
                                       Phone: +45 93 12 12 (tone) 16 64
                                       e-mail: igkhwj@unidhp.uni-c.dk

Re: Irradiance Gradient method on surfaces with complex BRDF's

Hi Greg,

I have been working with rendering/ray tracing for a while and I think
that Radiance is a very impressive program (the most impressive rendering
package on the net).

I have a question regaring the Irradiance Gradient method. As far as
I know it is currently limited to Lambertian surfaces. Has any work
been done towards extending it so that it could be used on surfaces
with more complex BRDF's (e.g. like Sillion's spherical harmonics 
on surfaces with directional diffuse properties)? If not then how
are these complex surfaces dealt with - by plain recursive Monte
Carlo sampling - or via some Lambertian assumption?

I have a "little" extra question that I have been wondering about.
How are caustics from refractive objects calculated. I know that
windows has been taken into account. But how about a pair of glasses
lying on a desk slightly focusing the light. I know that radiance does
not emit light rays from the light sources but is there a way around this 
problem (e.g. polygonalization of the refractive object) or are the 
caustics simulated using Monte Carlo sampling perhaps combined with 
high lighting during refraction?

Best regards

- Henrik 

Date: Thu, 17 Nov 94 09:30:00 PST
From: greg (Gregory J. Ward)
To: hwj@henrik.igk.dth.dk
Subject: Re:  Irrad. grad. and BRDF

Hi Henrik,

The "indirect irradiance" calculation which makes use of gradient
approximations in Radiance does not consider the directional reflectance
of the surface on which the calculation takes place.  It considers only
the diffuse component of its reflectance function, which in some cases
is small (even negligible).  It does of course consider the directional
reflectances of other surfaces sampled over the hemisphere, but only
inasmuch as they affect the incoming "indirect" radiation.

I know of no one who has extended the gradient work to more detailed
BRDF's, but the idea has merit.  Remembering the trouble I had just
dealing with diffuse incidence with a simple sampling pattern, I'm not
sure I would want to tackle such a problem myself.  Also, you must
remember that the principal problem with non-diffuse surfaces is that
their effect changes with the OUTGOING direction.  Therefore, any work
you do on computing the gradient must be duplicated somehow for each
outgoing direction.

Caustics are not handled properly in Radiance.  There are two reasons
for this.  The first is that they simply are not very important in most
architectural environments.  If light redirection is used, it is usually
done with mirrors or prisms rather than lenses.  Radiance does handle these
cases.  Second, caustics through refractive media are very difficult in
general to deal with.  Radiance "cops out" by following shadow test rays
through the refractive surfaces (be they flat or curved), and if they hit
the source, great.  If not, we consider it a "shadow", even though it's not.
This halfway approach is exactly right for glass panes, but further from
the truth the further the glass is from a plate.

I hope this answers your questions.
-Greg

From: hwj@henrik.igk.dth.dk
Subject: A few academic ideas...
To: greg@hobbes.lbl.gov
Date: Fri, 18 Nov 1994 03:15:43 +0100 (GMT+0100)
 
Hi Greg,

Thanks, for your fast reply...

I agree that modifying the irradiance gradient method to account for
surfaces with directional reflection is a complex task. (The following
is just a thought...) A simple idea could perhaps be to use the old
irradiance technique without the gradient information (comp. graph. 1988).
Instead of storing only the "Lambertian intensity constant" one could
use Sillion's spherical harmonics (comp. graph. 1991) or perhaps Wavelets
to store a complete intensity distribution and then compute the
interpolation/extrapolation of the intensity from this distribution.
This would allow an efficient simulation in models containing surfaces
with directional diffuse properties.

You are right that caustics aren't particularly important in practice
but still... I believe that caustics could be integrated rather
easily in Radiance as an external tool that would apply the light
ray tracing method introduced by Arvo in 1986. The tool or program
could be used in a preprocessing step and it should calculate the
caustics values using illumination maps. These illumination maps
could then be integrated into the rendering pipeline as texture
maps with illumination information. This would unfortunately be
limited to Lambertian surfaces but it would allow the simulation
of caustics caused by the reflection of light from a (small) water
bassin. I guess this effect is also used by some architects.
This kind of calculation must be somehow similar to the skyillum-program (?)
that calculates the indirect contribution through windows. The tool
would also fit nicely into the store and reuse idea exploited in Radiance
since the illumination maps are view-independent and as such reusable.

A little extra question. Has anyone implemented fractal/procedural objects
in Radiance (e.g. plants) without approximating these with polygons or
other simple objects?

- Henrik

PS: I don't hope my questions/ideas are to academic...

Date: Thu, 17 Nov 94 15:50:57 PST
From: greg (Gregory J. Ward)
To: hwj@henrik.igk.dth.dk
Subject: Re:  A few academic ideas...

Hi Henrik,

The main problem with storing information linked to directional reflection
is the amount of RAM it takes up.  The diffuse+gradient information stored
in Radiance requires about 75-85 bytes/value, which when you add up all the
values to be computed is quite substantial.  Multiply that by a factor of
at least 20 for spherical harmonics or the like and you're all out of memory.
Swapping becomes the rule, as in most radiosity calculations (even the ones
just handle Lambertian surfaces).

Your idea for integrating caustics is not too far-fetched, but again, the cost
of applying texture maps to EVERY surface is simply not practical.  Especially
when you consider that many Radiance models contain upwards of 100,000
surfaces, you can see why this might be a problem.  I feel it is essential to
always look at the practical side of the computation.  Otherwise, you end
up with something that no computer can run.

I have not implemented any primitives other than polygons, spheres and cones.
It is possible to do so, but you may find that the expense both in terms of
programming and in terms of rendering time is not worth the effort.  You can
save some RAM by implementing higher-order primitives, and maybe even improve
the quality of your results, but you will generally incur a large expense
during rendering unless you are VERY clever with your intersection procedure.

I recommend instead that you make use of the functionality already built into
gensurf, and check out the "genpine.csh" program in the ray/src/gen directory
as an example of building a fractal tree using hierarchical instancing.

-Greg

From: hwj@henrik.igk.dth.dk
Subject: Re: A few academic ideas...
To: greg@hobbes.lbl.gov (Gregory J. Ward)
Date: Fri, 18 Nov 1994 19:00:42 +0100 (GMT+0100)

Hi Greg,

I agree that spherical harmonics does indeed use a lot of memory. However
they should only be used on the (few) surfaces in the scene that has
directional reflection properties. Furthermore one might even cut the
memory requirements further down by using sp.ha. only at intersections
with primary rays or rather intersections seen by the eye (that includes
rays that have had one or more specular reflections).

The caustics idea could also be optimized a lot. Only surfaces illuminated
by caustics need a texture map (these surfaces could be specified by the
user).

One of the things I really like about Radiance are the rather advanced
materials and light sources. In order to use these properly I have
been looking for some material-database containing reflection-data
for common materials (metals, plastics etc.) but I have not been
able to find it. I was very pleased to see your 92-paper on
Anisotropic Reflection especially the tables of data measured
from a range of common materials. Since the article also presented
equipment for measuring reflection-properties I have been
looking forward to seeing some database containing measurements for
different materials. I think that this would be very useful since
it could eliminate some of the tentative part of designing scenes --- 
the creation of materials. Does such a database exist?

- Henrik

Date: Fri, 18 Nov 94 11:14:34 PST
From: greg (Gregory J. Ward)
To: hwj@henrik.igk.dth.dk
Subject: Re: A few academic ideas...

Hi Henrik,

Most of the surfaces are non-diffuse, unfortunately, so that won't help
reduce memory costs much.  Primary rays still intersect many, many
thousands of surfaces.  I don't really like these kinds of halfway,
compicated implementations, especially when my budget for Radiance
development has dropped to zero.

Your idea for texturing selected surfaces is exactly what I have in
mind for my next work on light source computations.  The mkillum program
is a first attempt at that, though it still has trouble with some
interactions since it still runs rays backwards w.r.t. light.

Your ideas for databases are appreciated, and I agree that this is sorely
lacking.  Ultimately, I believe it is the responsibility of surface
manufacturers to measure their own products, since there are way too
many for the research community to deal with them unaided.  My current
work in this area is to develop and calibrate a portable reflectance
measurement device, that would permit field measurements of surface
materials in situ.  I see this as the most practical short-term solution
to the problem, since the databases we both want don't seem to be
forthcoming.

-Greg

From: hwj@henrik.igk.dth.dk
Subject: Re: A few academic ideas...
To: greg@hobbes.lbl.gov (Gregory J. Ward)
Date: Mon, 21 Nov 1994 19:05:20 +0100 (GMT+0100)

Hi Greg,

Most real-world surfaces are non-diffuse, but I think that our computer
models could settle with only a few "important" surfaces being non-diffuse
and since reflectance-data aren't available this comes quite natural.

I agree that different materials ought to be measured by their manufactors.
However computer graphics is quite new and it has not yet proven useful
in enough areas to make this realistic. Instead I believe that your
measurement device could be very useful in obtaining reflectance data for 
common materials that would fit the isotropic/anisotropic models. I really
believe the computer graphics community could benifit a lot from this.
It would fx. make the creation of standard scenes more realistic since
we would be able to talk about the same materials - instead of only
Lambertian reflection. 

I was browsing through the scenes following Radiance and I found some 
reflectance data for gold, silver, brass, wood, etc. Are these materials 
measured or just created in order to look good?  

I am looking forward to seeing illumination maps in Radiance. In my own
rendering program I have quite succesfully implemented light ray tracing 
and combined it with the irradiance gradient method.

Some time ago I heard about a global illumination mailing list. Does this
mailing list still exist and if so, can I somehow join it?

- Henrik

Date: Mon, 21 Nov 94 10:51:01 PST
From: greg (Gregory J. Ward)
To: hwj@henrik.igk.dth.dk
Subject: Re: A few academic ideas...

Hi Henrik,

As much as I would like to create a materials database, the real work of
collecting and cutting up suitable samples is beyond me.  I wouldn't know
how to go about it.  Perhaps you or someone could organize a "sample drive,"
where interested people sent in appropriate samples of various materials
for me to measure.  I don't want to do all the work of cataloging everything,
but I could probably handle the actual measurement and fitting if there
weren't too many samples to deal with.  This is a very tentative idea...
I'm nervous about taking on any new projects without funding right now,
as I have plenty to do and not enough time to do it already.

I'm glad that you've managed to implement and use the irradiance gradients
in your code.  You're the first I've heard from to do so.

Yes, the globillum list is still going.  To sign up, send a request to Paul
Heckbert <ph@cs.cmu.edu> and say you want on it.

-Greg

From: hwj@henrik.igk.dth.dk
Subject: Re: A few academic ideas...
To: greg@hobbes.lbl.gov (Gregory J. Ward)
Date: Tue, 22 Nov 1994 01:23:17 +0100 (GMT+0100)

Hi Greg,

I can see the problem of collecting material-samples. I would however
also think that measuring and fitting these materials would take a lot
of time. 

I could see if I could get my hands on some materials like:
Bricks, pieces of wall-paper, different types/colours of paint, 
plants? (grass, leaves, bark...) Of course a number of these 
would have to be accompanied by some texture mapping. Plants would
be quite difficult to sent - I guess the reflectance properties
would change during air-mail ;-)   
 
I wouldn't mind collecting a material database. I think the work of 
cataloging the materials could simply be done (initially) using a
text-file containing names of materials and the corresponding
reflectance model and measured data - just like the tables you
presented in your paper at siggraph 92. 

It surprises me that no one else has announced implementations of the
irradiance gradient method. I am very pleased with it. I started
initially with plain recursive Monte Carlo sampling which took
years (almost) to complete. Then I found your article on the store and
reuse technique in comp. graph. 1988 which I implemented and it was a large
improvement (It did however eat up a bit of memory). A few month
later I found your article on the irradiance gradient method which
was again a large improvement and it gave very good results. 

- Henrik

From: hwj@henrik.igk.dth.dk
Subject: Re: A few academic ideas...
To: greg@hobbes.lbl.gov (Gregory J. Ward)
Date: Tue, 22 Nov 1994 22:32:03 +0100 (GMT+0100)

Hi Greg,

It could indeed be very interesting to arrange a sample drive. I can 
think about a number of materials that could be quite interesting to
measure: Bricks, different types/colours of wall-paper, different
types/colours of paint, plants (I don't know whether they fit into
the model). I wouldn't mind cataloging these materials. I think it can
be done (initially) simply by adding information to a text-file regarding
the material name and the measured reflectance data. 
 
It surprises me that no one else has announced implementations of
the irradiance gradient method. It has been very useful to me. I
started initially with plain Monte Carlo sampling of the indirect
light (it took years to complete). Then I found your article in
comp. graph. 88 concerning storing and reusing the indirect light
which I implemented with very good results. Later I found your
92 article with the gradient technique and it gave even better
results. I think that for instance Chen in comp. graph. 91 could
benifit from the irradiance gradients method instead of using path
tracing in the visualization step.
 
I have read in your articles that the results in Radiance are verified
from time to time (and they improve). How is this verification done.
Is it via a simple model that can be handled with exact calculations or
is it via a brute force Monte Carlo technique?

- Henrik

Date: Tue, 22 Nov 94 11:14:14 PST
From: greg (Gregory J. Ward)
To: hwj@henrik.igk.dth.dk
Subject: Re: A few academic ideas...

Hi Henrik,

Sorry not to respond to your last message, but there are so many problems
with creating a material database that I haven't solved yet, that I don't
even want to think about it right now.  (Material textures and colors are
BIG problems.)

Radiance has been compared to measurements of scale models, other lighting
simulations and measurements and qualitative comparisons of full-sized
environments.  Most of the good validation work has been done outside of LBL.

-Greg

From: hwj@henrik.igk.dth.dk
Subject: Re: A few academic ideas...
To: greg@hobbes.lbl.gov (Gregory J. Ward)
Date: Tue, 22 Nov 1994 20:24:10 +0100 (GMT+0100)

Hi Greg,

> Sorry not to respond to your last message, but there are so many problems

Sorry for resending it. But elm crashed just after sending the mail and I
was wondering whether it was sent at all.

> with creating a material database that I haven't solved yet, that I don't
> even want to think about it right now.  (Material textures and colors are
> BIG problems.)

That is true, but as a beginning textures could be ignored. Why are colours
problematic?

> 
> Radiance has been compared to measurements of scale models, other lighting
> simulations and measurements and qualitative comparisons of full-sized
> environments.  Most of the good validation work has been done outside of LBL.

Are these results somehow available? I would be very interested in validating
my own little rendering-program....

- Henrik

Date: Tue, 22 Nov 94 13:54:25 PST
From: greg (Gregory J. Ward)
To: hwj@henrik.igk.dth.dk
Subject: Re: A few academic ideas...

Hi Henrik,

The problem with colors and textures is that the device I currently have can
neither measure nor ignore them.  They interfere with the measurements,
especially surfaces with visible textures or patterns.

There are a couple of reports with Radiance validation data in them.  One
is by John Mardaljevic:

	Mardaljevic, John, K.J. Lomas, D.G. Henderson, ``Advanced Daylighting
	Design for Complex Spaces'' Proceedings of CLIMA 2000, 1-3 November
	1993, London UK.

Two others come from LBL:

	Grynberg, Anat, Validation of Radiance, LBID 1575, LBL Technical
	   Information Department, Lawrence Berkeley Laboratory, Berkeley,
	   California, July 1989.

	Papamichael, Kostantinos, Liliana Beltran,
	``Simulating the Daylight Performance of Fenestration
	Systems and Spaces of Arbitrary Complexity:  The IDC Method''
	"Proceedings of Building Simulation `93",
	Adelaide, Australia, August 16-18, 1993.

Don't ask me to send Anat's to you.  It's rather long, and the color
images don't copy well.  These papers are not the best resource if what
you want to do is validate your own software, however.  You would be better
off using the simulations that lie in the /pub/tests directory on
hobbes.lbl.gov.  That's what they're there for.

-Greg