Hello folks,
Since I've gone to the trouble of setting up this mailing list of Radiance
users, I might as well use it for something. (By the way, feel free to
send mail directly to the group yourself -- it's ray@hobbes.lbl.gov.)
This is the first edition of a digest of correspondence between myself and
other program users. I just take what I think might be interesting to
the general populace, not everthing. (You're welcome.) If you send me
mail that you don't want shared in this fashion, please tell me so at
the time.
I will start with mail I received (or sent) this year -- I think the older
stuff is too out of date now, anyway. In this digest, you will find
discussions on the following topics:
New options and programs (-p, -z, pinterp)
Penumbras and source sampling
Modeling software
Radiance display drivers
Pfilt and ies2rad light sources
Modifiying source for huge scenes
Radiance course possibility
--------------
OPTIONS
From greg Thu Jan 11 12:18:04 1990
Date: Thu, 11 Jan 90 12:17:59 PST
Subject: new programs and options
Dear Radiance users,
There is a new rview driver for X11 (written by Anat Gryberg) and a new
program for interpolating new views from images, called pinterp. There is
also a new option for rpict and pfilt to set the pixel aspect ratio of the
output picture, and rview no longer takes x and y resolution arguments.
Under X10 at least, you will notice that rview now responds to resize
requests.
Here are some release notes on the changes:
Added a -p option to rpict and pfilt to set the pixel aspect ratio for
output. Instead of giving the absolute x and y resolutions, the user
now gives rpict and pfilt the maximum desired resolution for x and y
and the pixel aspect ratio is used along with the given view to
calculate appropriate values within this boundary. This makes for
much more natural view specifications. For example, for a 512x400
device with a pixel aspect ratio of 1.0, the pfilt command:
pfilt -x 512 -y 400 -p 1
will always produce the appropriate output, regardless of the aspect
ratio of the input picture. If necessary, the x or y output resolution
will be reduced to accommodate the device's resolution. A square image
would occupy a region of 400x400 pixels.
View shift and lift options were added to the list of standard view
parameters, for specifying views for panoramas and holograms.
Rview no longer takes options for x and y resolution, but instead
gets them from the device driver along with the pixel ratio. This
makes it much easier to change the view aspect ratio (with the vh
and vv parameters).
A -z option was added to rpict to write out the distances for each
pixel in an image. This may be useful for z-buffer operations, and
is used by the new program pinterp, described below.
A program called pinterp, was added to the burgeoning list of
picture filters and converters. This program is designed primarily
to interpolate animated frames for walk-throughs of static scenes,
but it has a number of other useful functions besides. It takes as
its input one or more rendered pictures (with their corresponding z
value files) and desired viewpoint (hopefully not too far afield from
the given images). Pinterp then takes the input frames and moves the
pixels to their new computed location. Filling operations make sure
that the final image does not have large unpainted regions.
-Greg
From greg Tue Jan 16 08:48:57 1990
Date: Tue, 16 Jan 90 08:48:48 PST
To: dasilva%ced.Berkeley.EDU@jade.berkeley.edu
Subject: Re: recovering rpicts
Hello Deanan,
The -x and -y options have not been replaced in rpict, -vs and -vl are new
options. There is another new option, -p, which you will need to set to 0
to get your recovery operation to work. Simply add -p 0 to every rpict
command in your makefile. This wouldn't normally be required, except that
you are recovering files you created with the old rpict, which didn't pay
attention to pixel aspect ratios. Read the new manual page for rpict to
understand what I'm talking about.
The -z option of rpict is easy to use. Simply give it the name of the
file where you want to store the z buffer information, then stand back!
The z-file can be quite large, since it stores 4 bytes for every pixel.
For a 512x512 image, that's already 1 megabyte. You don't need to do
anything special to recover z-file information, just use the option as
you did in the aborted rendering.
Good luck, and let me know if you have any other questions.
-Greg
-----------
PENUMBRAS
From greg Fri Jan 19 11:36:32 1990
Date: Fri, 19 Jan 90 11:36:25 PST
To: anderdla@cs.uoregon.edu
Subject: penumbras
Dear Darren,
Thank you for your letter.
You don't need to change your scene description at all to generate
penumbras, only the options to rpict (or rview). Set -dj to some
value less than 1. For most scenes, a value of .5 will do nicely.
If the value is too close to 1, strange things may happen for
irregularly shaped light sources (see the BUGS section of the rpict
manual page). Spherical light sources work best, but rings and nearly
square polygons work also as area light sources. Note that if your
light source is small in relation to the ratio of the obstruction-
shadow vs source-shadow distance, then the penumbras will not be
very pronounced (ie. the shadows will be sharp).
I am glad to hear that you are using the software!
-Greg
From greg Fri Jan 19 11:39:57 1990
To: anderdla@cs.uoregon.edu
There is one other thing I should mention. When you generate penumbras,
image sampling may start to cause problems. You may need to set the -sp
option to 1, which will result in longer rendering times. This is why
direct jitter is not turned on normally (rpict -defaults).
-Greg
From anderdla@cs.uoregon.edu Mon, 22 Jan 90 13:07:12 PST
To: greg@hobbes.lbl.gov
Subject: More questions!
Well, thanks for telling me about -dj! There is a real nasty side
effect,though. It seems the as I increase shadow jitter, the picture
becomes increasingly grainy. I have tried changing the -sj, and -sp
parameters to overcome this. I have also tried using pfilt to no
avail. Do you know how to overcome this problem??
Thanks.
Darren Anderson
From greg Mon Jan 22 13:26:05 1990
To: anderdla@cs.uoregon.edu
Subject: Re: More questions!
Are your light sources very long and narrow? This can cause big
problems for the -sj algorithm. You must break up any long sources
into smaller, more square pieces. Even if your light sources have
an aspect ratio around 1 (ideal), you still should not use a value
for -dj greater than about .7 .
A modest amount of graininess is to be expected of any Monte Carlo
sampling technique. In this case, it is caused by the angle between
the surface and the light source direction varying with the source
sampling. To reduce the amount of graniness, either lower -dj
(cheap) or raise the resolution for rpict and use pfilt to antialias
down to the final picture resolution (accurate):
rpict -dj .5 -sp 1 -x 1024 -y 1024 octree > rpict.pic
pfilt -x 512 -y 512 -r .7 rpict.pic > pfilt.pic
The -r option of pfilt uses a Gaussian filter, which looks slightly
better than the default box filter. If you have adjusted your
light sources to give you the picture brightness you want straight
out of rpict, you can use the -1 option of pfilt to speed it up.
If you don't want to produce an intermediate file (rpict.pic), you
can pipe the output of rpict directly into pfilt.
-Greg
-----------
MODELS
From hchen@gumbo.age.lsu.edu Tue Apr 10 07:34:56 1990
To: gjward@Csa1.lbl.gov
Subject: CAD programs
Dear Mr. Ward,
1. We read your RADIANCE Tutorial within RADIANCE package, it says that
the input model may contain many thousands of surfaces, and is often
produced by a separate CAD program'. We would like to know what kind of
CAD programs can be used in this situation. Can we use AutoCAD as a mean
to produce a model? If so, how?
2. I try to run the examples under examples/conf subdirectory using make
command. It give me the error message of "chair1.oct not found". The
original Makefile is as:
#
# Makefile for the conference room
#
VIEW = -vf vf/current
SCENE = test
#DEV = X
DEV = sundev
AMB = -av .02 .02 .02
OCTOPTS = -f
view: $(SCENE).oct
rview $(VIEW) -o $(DEV) $(AMB) $(SCENE).oct
Actually, octree file 'chair.oct' sits under current directory. I don't
know why the programs couldn't find it.
Thank you for your help.
Huaiming Chen
From greg Tue Apr 10 12:04:25 1990
To: hchen@gumbo.age.lsu.edu
Subject: Re: CAD programs
The conference model is probably not working because you haven't set your
RAYPATH variable to include the current directory ".". Radiance uses this
environment variable to determine where to look for auxiliary files (incl.
instance octrees). The default value is ".:/usr/local/lib/ray", which
includes the current working directory.
There is currently no translator from AutoCAD, but we expect to have one
sometime in the near future. For it to work, the model would have to
have been created with surfaces, rather than lines.
We have a translator for GDS (from McDonnell Douglas) and may have one
for MacArchitrion soon as well.
Right now, genbox, genrev and gensurf are the most useful surface description
generators (oh, not to forget genprism, one of my personal favorites).
-Greg
----------
DRIVER
From mb@cs.albany.edu Thu Jun 21 12:50:30 1990
To: GJWard@Csa1.lbl.gov
Subject: RADIANCE
Dear Mr Ward,
We have a network of sun3's and sun4's running sunos 4.0.3. We
just installed RADIANCE on both architectures but we're having some
problems getting it to run. The installation process seemed fairly
simple - a matter of placing binaries and library files in the right
places, and so we did not recompile anything. Following the tutorial
given we get error messages when tyring to invoke rview. These are
the messsages:
rview: cannot open X-windows; DISPLAY variable set?
rview: fatal - cannot initalize X
The display variable is set to unix:0.0 for any user, it was not clear
that this had to be changed in any way. And these messages occurred
while in X (we run XllR4). If you could give some help in this matter
we would appreciate it, as we would very much like to use this software.
Thank you,
Michele Buselli
State University of New York - Albany
(518) 442-4279
...some back and forth, then:
From greg Wed Jun 27 11:23:12 1990
To: mb@cs.albany.edu
Subject: Re: RADIANCE
Michele,
Since you are already linking the X11 driver into rview directly, there
is no need to compile the separate driver program x11dev. Just remove
it from the DRIVERS definition in your Makefile. (If you were going to
build x11dev, you would have to add one more special compile similar to
that for x10.o, but as I said, it would be redundant in your case.)
If you don't use X10 at all, you should remove the line for x10dev from
devtable.c. Rview will still compile with it in, but without building
x10dev, this driver would not function.
Perhaps I should better explain how drivers work in rview. A driver
is an interface to the rview program that provides a few basic graphics
input and output functions, which are described in driver.h in some
detail. There are two basic driver types, drivers that are linked to
rview directly, and standalone programs that talk to rview via a pair
of UNIX pipes. Due to efficiency considerations, linked drivers are usually
preferred, but there are a few reasons for having standalone drivers
instead:
1) The libraries used by the driver are incompatible with
other program requirements or drivers. (Eg. sunview
libraries prevent the use of UNIX signal facilities,
and X10 and X11 calls interfere.)
2) The libraries are only supported on certain machines.
3) The driver's libraries result in a huge program. (Eg. when
I attempted to link rview to sunview in the past, the
compiled program quadrupled in size!)
4) Standalone drivers can be compiled without changing any of
the code for rview, thus avoiding the need for source
recompilation.
In your case, you will still need to compile sundev as a standalone
driver, but you can link to x11 directly (as your Makefile does already).
Compile x10dev only if you are still using X10 on some machines.
-Greg
From mb@cs.albany.edu Wed Jun 27 14:54:11 1990
To: greg@hobbes.lbl.gov
Subject: RADIANCE
Hi Greg,
Thanks very much. The Makefile compiled just fine. I'm in the process
of looking through the rest to see if I need to make any changes. At
first glance, there doesn't seem to be a need for this. Just out of
curiosity, what kind of environment do you run Radiance under?
Michele
From greg Wed Jun 27 15:05:57 1990
To: mb@cs.albany.edu
Subject: Re: RADIANCE
Hi Michele,
The environment here is sunny most of the summer, although we do get some
fog in the mornings (which is nice because things cool down then).
Oh -- I guess you mean what kind of computer environment, huh?
I have a single Sun-3/60 running SunOS 3.5 and X10R4, and it hasn't
changed much in the two years since I bought it. We recently received
a grant from Apple and have been running the programs on a MacIntosh
IIcx running A/UX 1.1.1 and X11R3. (We have just ordered A/UX 2.0.)
The architecture department at UCB, which has been using Radiance quite
a bit, is running mostly Sun computers, although they were recently
given about 10 Silicon Graphics IRIS workstations and I am in the
process of getting drivers up on those machines.
I don't use sunview much myself, though I have easy access to it. By far
the environment Radiance has been used and tested in most heavily is Sun-3's
running SunOS 3.5 or 4.0 and X10. I wish I had better contact with people
using the software on different systems, so I could incorporate their
modifications and additions back into the distributed code, but I don't
communicate much with folks outside of UCB.
-Greg
----------
LIGHT
From emo@cica.indiana.edu Wed Aug 15 07:41:29 1990
To: greg@hobbes.lbl.gov
Subject: why luminance intensities so low???
Why is it the case that the radiance values output from ies2rad
seem to be woefully low? It's not unusual for the initial values
to have to be increased by factors of 3-10. Is there some trick
I'm missing that can be played with the '-dX' option to ies2rad?
For instance, if one were to set up an actual IES lighting device
10 feet from a white wall the visual impact of that illumination
is much more profound than that obtained by simulating the same IES
light source in Radiance projected onto a white wall 10' away.
Any clues/suggestions?
eric
From greg Wed Aug 15 07:54:00 1990
To: emo@cica.indiana.edu
Subject: Re: why luminance intensities so low???
Hi Eric,
Thanks for spotting the inconsistency in func.c! I will fix it on
future distributions. (I think only one other went out with the wrong
version.)
As far as the low light levels are concerned, you must specify the
same units to ies2rad with the -dX option as you are using in your
scene. Other than that, you must also realize that the image you
get from rpict is not exposure-adjusted, and you will probably have
to use pfilt to get a nice picture. The pixel values in the file
correspond to radiance, which is not always in the right range for
display. Pfilt fixes that.
-Greg
From emo@cica.indiana.edu Wed Aug 15 09:06:00 1990
To: greg@hobbes.lbl.gov
Subject: using pfilt
Could you send me a bit more info on using 'pfilt' to obtain
an exposure-adjusted image? Is the 'one-pass' option better
in this regard? What about using the other 'filtering' functions?
eric
From greg Wed Aug 15 18:22:53 1990
To: emo@cica.indiana.edu
Subject: Re: using pfilt
Pfilt without any options just does an automatic exposure adjustment.
The -1 option is faster, but only works if you know already what
exposure to set. If you had run pfilt before, then getinfo printed
a line from the final picture saying:
EXPOSURE=3.52
then you could run pfilt -1 -e 3.52 the next time and get the same
picture a little bit faster.
The other options are for anti-aliasing and rely on turning a big,
high-resolution picture into a smaller, anti-aliased picture. The
-r option (with a value of .6 or so) produces a nicer image at a
slightly higher processing cost.
-Greg
---------
SIZE
From emo@cica.indiana.edu Sun Sep 2 14:55:54 1990
To: greg@hobbes.lbl.gov
Subject: large polyhedra
Greetings Greg.
Another of the projects I am working on involves some astronomers who
produce large-ish data sets, on the order of 65x65x15 (symmetric about
x-z plane). We then use a modified marching-cubes 3D contouring algorithm
to resolve certain polyhedra of interest, e.g. surfaces w/ specific gas density.
These polyhedra are sometimes composed of 40,000+ discrete polygons
(actually, triangles). Converted to Radiance 'polygon' format,
this file becomes ~9+ Mb and 'oconv' crashes when producing the .oct
file, indicating that it is out of 'object space'.
Thus, I have two questions:
1. by modifying MAXOBJBLK in object.h and recompiling 'oconv' can
I increase the available space for object storage and thereby permit
the loading of my 9+ Mb polyhedra specification?
2. alternatively, I would like to simply be able to load a .geom
file, composed of a vertex table and edge list for each discrete
polygon. Looking at the code in 'readobj.c', it's not clear how
I can go about implementing the code to interface to this new kind
of 'object'. What would you suggest?
As always, thanks for the support!
eric
From emo@cica.indiana.edu Sun Sep 2 14:58:24 1990
To: greg@hobbes.lbl.gov
Subject: polyhedra size
I just discovered that some of the contour surface have upwards of 80,000
polygons in their specification.
One more question: when approaching 100K polygons, am I going to run into
performance bottlenecks in Radiance's implementation. In other words,
is it realistic to expect rapid renderings, esp. using 'rview', of
such large polyhedra?
Thanks.
eric
From greg Sun Sep 2 16:16:10 1990
To: emo@cica.indiana.edu
Subject: Re: polyhedra size
Hi Eric,
I was wondering when someone would want to start working with huge models.
The main concern is, do you have enough memory? The performance of oconv
O(N), meaning 100,000 surfaces should take 100 longer than 1000 surfaces
to convert. That is actually going to be your main cost in terms of time.
Rview and rpict have an O(n^.33) intersection algorithm, so 100,000 surfaces
in general will take roughly 4.5 times longer than 1000 surfaces.
I don't recommend implementing a vertex sharing polygon structure. I have
considered such a model, and it doesn't save much space -- especially for
triangles. You are better off just changing the definitions as you suggested.
Besides increasing MAXOBJBLK in object.h (you might try 2047 or 4095 to start),
you will have to change the type of OBJECT from short to int (or long). Also,
you will probably need to increase MAXOBLK in octree.h to 8191 or more or you
will run out of octree space when running oconv. Also, for better performance,
you should probably increase OSTSIZ in objset.c a like amount, using a prime
number. (I suggest you start with 12329.)
I hope you have done a "back of the envelope" calculation to figure out how
much memory this is all going to take. You may find yourself in over your
head in a hurry. For example, I have 16M on my machine, and it starts to
choke on models of around 18,000 surfaces.
Let me know how it goes and if you run into any other errors.
-Greg
-------------
COURSE
From arthur@abies.cfnr.colostate.edu Mon Oct 8 23:04:48 1990
To: GJWard@Csa1.lbl.gov
Subject: RADIANCE
Hello Greg;
If you recall, I first made contact with you last winter.
I have recently attempted to tackle RADIANCE again. The Tutorial
available in the updated version gave me hope! I do find it very
cryptic, however. Do you offer short courses in RADIANCE ? I would
gladly fly out for instruction. I am unable to progress rapidly
enough. ANy suggestions?
D. Arthur Sampson
Dept. Forest and Wood Sciences
Colorado State University
Ph.D. Student
From greg Tue Oct 9 10:45:02 1990
To: arthur@abies.cfnr.colostate.edu
Subject: Re: RADIANCE
We are going to have a meeting on Radiance and Superlite (a daylighting
analysis program) this January at LBL, and might be able to work a short
tutorial into it. If that's too far away for you, and you have a little
money to spend, I might be able to recommend someone who has an excellent
background in using the software to serve up some private lessons.
-Greg
From arthur@abies.CFNR.ColoState.EDU Tue Oct 9 11:53:43 1990
To: greg@hobbes.lbl.gov
Subject: RADIANCE meeting
Hi;
I would be interested in seeing an agenda for the meeting,
or if a Tutorial is appropriate for my request (Would I be
welcome in the meeting?), that would work nicely. Also, of
interest now is this Superlite program you mentioned.
Arthur
------------
End of Radiance Digest v1n1
Let me know if this has been useful to you. It is not my intention to
flood people's boxes with unread mail.
-Greg
Back to Top of Digest
Volume 1, Number 1
Return to RADIANCE
Home Page
Return to RADIANCE
Digests Overview