expert assistance/guidance please?

Hi all,

Thanks for helping me to get up and running with Radiance up to this point. I have gotten to the point of being able to produce compelling images, and setting some variables to get results that I THINK I want. In other words, I think I know enough to be naively dangerous. Thus, I was wondering if some of the experts out there would be willing to give me some assistance.

I am about to start rendering a series of images for different times of day throught out the year. Accuracy is tantamount, speed, as long as it is reasonable, is not as important.

So, I have created a generic enclosed office space, assigned materials, etc. After using the rad to create an .oct file, I gave the rpict command with the following options:

rpict -t 60 -vf 1.vf -x 3000 -y 3000 -ps 1 -pt .04 -dp 4096 -ar 128 -ms 1.5 -ds .01 -dj .6 -dt .05 -dc .75 -dr 3 -sj 1 -st .01 -ab 7 -aa .08 -ad 1024 -as 512 -av 0 0 0 -aw 0 -af 6_21_8s.amb -lr 12 -lw .0005 office.oct > 6_21_8s.pic

It took about 13 hours, but I got an image.

Are there any blatant newbie mistakes I have made that may compromise the accuracy of the rendering? I decided not to use the illum material, and am relying completely on ambient bounces. At this point, there are no electric illumination in the space, just purely daylight. I have uploaded the resulting .pic images to http://homepage.mac.com/jsan/, as well as a .pic image that I used pfilt -x /2 -y /2 on.

I would really appreciate it (and am sure I will learn a lot) if someone would tell me what I may be doing incorrectly. Individually, I know all the information is in Rendering with Radiance. However, because of the many variables, I feel a bit unsure as to whether I have gotten the settings correct or not.

Another point I have been unsure about is the material properties of somewhat complex materials. I understand that for simple materials (plastics, metals) the reflectance is determined strictly by the r g b values (each value multiplied by a specific coefficient). However, if I apply a texture to the material, does the texture impact the reflectance value? For instance, I want a floor with 20% reflectance, so I gave it r g b values of .2 .2 .2. Does the fact that I applied a carpet texture further reduce the value? In Lightscape, I know that textures don't impact the characteristic of the material; I have a suspicion that in Radiance, it does.

Sorry for the rambling.

Thanks.

PS
I also wanted to give an update on the availability of the book Rendering with Radiance. My order with amazon.com was cancelled by the seller (saying that he had run out). However, just yesterday, I checked amazon.com again, and found a retailer selling new and used copies for around $64. I'm hoping this one doesn't get cancelled.

From: John An <[email protected]>
Date: Sun May 11, 2003 9:32:13 PM US/Pacific
To: [email protected]
Subject: [Radiance-general] expert assistance/guidance please?
Reply-To: [email protected]

Hi all,

Thanks for helping me to get up and running with Radiance up to this point. I have gotten to the point of being able to produce compelling images, and setting some variables to get results that I THINK I want. In other words, I think I know enough to be naively dangerous. Thus, I was wondering if some of the experts out there would be willing to give me some assistance.

I am about to start rendering a series of images for different times of day throught out the year. Accuracy is tantamount, speed, as long as it is reasonable, is not as important.

So, I have created a generic enclosed office space, assigned materials, etc. After using the rad to create an .oct file, I gave the rpict command with the following options:

rpict -t 60 -vf 1.vf -x 3000 -y 3000 -ps 1 -pt .04 -dp 4096 -ar 128 -ms 1.5 -ds .01 -dj .6 -dt .05 -dc .75 -dr 3 -sj 1 -st .01 -ab 7 -aa .08 -ad 1024 -as 512 -av 0 0 0 -aw 0 -af 6_21_8s.amb -lr 12 -lw .0005 office.oct > 6_21_8s.pic

It took about 13 hours, but I got an image.

Are there any blatant newbie mistakes I have made that may compromise the accuracy of the rendering? I decided not to use the illum material, and am relying completely on ambient bounces. At this point, there are no electric illumination in the space, just purely daylight. I have uploaded the resulting .pic images to http://homepage.mac.com/jsan/, as well as a .pic image that I used pfilt -x /2 -y /2 on.

I would really appreciate it (and am sure I will learn a lot) if someone would tell me what I may be doing incorrectly. Individually, I know all the information is in Rendering with Radiance. However, because of the many variables, I feel a bit unsure as to whether I have gotten the settings correct or not.

You can get more accurate/smoother results by performing an "overture calculation" -- running rpict with the same view and parameters and saving the ambient with a low-resolution image (-x 64 -y 64 is what I usually use) and discarding the result. This "seeds" the ambient file, resulting in a smoother interpolation. Then, run your final rpict command. Your results look quite nice the way they are. You could get a much faster result by using mkillum to create an output distribution for your skylight and reducing some of your rendering parameter values. If this rendering time is acceptable to you, though, it's not really necessary.

Another point I have been unsure about is the material properties of somewhat complex materials. I understand that for simple materials (plastics, metals) the reflectance is determined strictly by the r g b values (each value multiplied by a specific coefficient). However, if I apply a texture to the material, does the texture impact the reflectance value? For instance, I want a floor with 20% reflectance, so I gave it r g b values of .2 .2 .2. Does the fact that I applied a carpet texture further reduce the value? In Lightscape, I know that textures don't impact the characteristic of the material; I have a suspicion that in Radiance, it does.

You're correct -- the pattern affects the total material reflectance/transmittance. To avoid this, run pfilt on your picture in advance, like so:

pfilt -e 2 orig.pic > normalized.pic

The "normpat" script will do this for you, and blend the edges for better tiling as well if that's what you have in mind.

Sorry for the rambling.

Thanks.

PS
I also wanted to give an update on the availability of the book Rendering with Radiance. My order with amazon.com was cancelled by the seller (saying that he had run out). However, just yesterday, I checked amazon.com again, and found a retailer selling new and used copies for around $64. I'm hoping this one doesn't get cancelled.

Rob and Chas and I are looking into print-on-demand publishers as a way for folks to order copies. We have the copyright back from the publisher, but need to figure out a way to publish it ourselves at a reasonable cost.

So, I have created a generic enclosed office space, assigned materials, etc. After using the rad to create an .oct file, I gave the rpict command with the following options:

... and then Greg Ward wrote:

You can get more accurate/smoother results by performing an "overture calculation" -- running rpict with the same view and parameters and saving the ambient with a low-resolution image (-x 64 -y 64 is what I usually use) and discarding the result.

Hi John,

I don't understand (nothing new there). You use rad to create an octree file, but not for controlling all the other aspects of the calculation? This overture calculation Greg mentions above is done automatically for you if you use rad to do everything. Yes, you get finer control of the rendering parameters if you call rpict manually, but you can override any switch you want in the .rif file. rad makes intelligent guesses about what the rpict parameters need to be based on how you answer three questions: Quality, Variability and Detail. And if you add the line OPTFILE=parameters.txt to the rif file, all the choices that rad made are saved to a text file called parameters.txt so you can view them at your convenience. You can reuse the settings too, like so:
# rtrace @parameters.txt octree...

Even if you want to stick with the manual method, playing with the rad settings and observing what rad does to the parameters is instructive.

···

John An <[email protected]> wrote:

----

      Rob Guglielmetti

e. [email protected]
w. www.rumblestrip.org

Let me begin by apologizing for the format of my emails. I only subscribe to
the digest of the list, so I don't have the original emails to reply to, so I
kind of have to haphazardly cut and paste it together. So, here I go...

Greg Ward wrote:
You can get more accurate/smoother results by performing an "overture
calculation" -- running rpict with the same view and parameters and
saving the ambient with a low-resolution image (-x 64 -y 64 is what I
usually use) and discarding the result. This "seeds" the ambient file,
resulting in a smoother interpolation. Then, run your final rpict
command. Your results look quite nice the way they are. You could get
a much faster result by using mkillum to create an output distribution
for your skylight and reducing some of your rendering parameter values.
  If this rendering time is acceptable to you, though, it's not really
necessary.

My reply:
Thanks Greg. I have started using the overture rendering method to create a
seed .amb file. It doesn't seem to take too much extra time, as the second,
full resolution rendering seems to go MUCH faster.

You note that mkillum will speed things up, but I also recall reading in
Rendering with Radiance that using illum is not as accurate as just letting
the computer calculate the direct and indirect lighting. Is the difference
so negligible that using illum would be ok for accuracy? I am trying to
produce renderings for analysis that are as defensible in terms of accuracy
as possible.

I had written:

Another point I have been unsure about is the material properties of
somewhat complex materials. I understand that for simple materials
(plastics, metals) the reflectance is determined strictly by the r g b
values (each value multiplied by a specific coefficient). However, if
I apply a texture to the material, does the texture impact the
reflectance value? For instance, I want a floor with 20% reflectance,
so I gave it r g b values of .2 .2 .2. Does the fact that I applied a
carpet texture further reduce the value? In Lightscape, I know that
textures don't impact the characteristic of the material; I have a
suspicion that in Radiance, it does.

To which Greg Ward replied:
You're correct -- the pattern affects the total material
reflectance/transmittance. To avoid this, run pfilt on your picture in
advance, like so:

pfilt -e 2 orig.pic > normalized.pic

The "normpat" script will do this for you, and blend the edges for
better tiling as well if that's what you have in mind.

To which I ask:
I'm not sure I understand how adjusting the exposure of an already processed
image will correct the reflectance of a material. Could you help me
understand this a bit more?

Rob Guglielmetti wrote:
You use rad to create an octree file, but not for controlling all the other
aspects of the calculation? This overture calculation Greg mentions above is
done automatically for you if you use rad to do everything. Yes, you get
finer control of the rendering parameters if you call rpict manually, but you
can override any switch you want in the .rif file. rad makes intelligent
guesses about what the rpict parameters need to be based on how you
answer three questions: Quality, Variability and Detail. And if you add
the line OPTFILE=parameters.txt to the rif file, all the choices that
rad made are saved to a text file called parameters.txt so you can view
them at your convenience. You can reuse the settings too, like so:
# rtrace @parameters.txt octree...

Even if you want to stick with the manual method, playing with the rad
settings and observing what rad does to the parameters is instructive.

My reply:
I initially used rad to control everything, but I noticed that the -av
parameter when I chose High for Quality, Variability, and Detail was .01 .01
.01. While that's close enough to 0 0 0, I wanted to be absolutely sure that
the light levels were the result of the calculations, and not some a priori
assumption. Thus, the rpict parameters I chose were based on the output of
my rad file with some slight modifications. Is there a way to overide some
of rad's intelligent guesses within the .rif file itself (specifically, the
ambient value and ambient weight)?

Final point regarding the publishing of Rendering with Radiance:
I'm sure it is a conscious decision not to digitize the entire book into a
print disabled .pdf, but could some electronic version be an option?

Again, thanks for your help.

John An wrote:

Let me begin by apologizing for the format of my emails. I only subscribe to the digest of the list, so I don't have the original emails to reply to, so I kind of have to haphazardly cut and paste it together. So, here I go...

Why not switch over to "real time mode"? Despite the traffic lately, it's usually pretty reasonable, traffic-wise.

You note that mkillum will speed things up, but I also recall reading in Rendering with Radiance that using illum is not as accurate as just letting the computer calculate the direct and indirect lighting. Is the difference so negligible that using illum would be ok for accuracy? I am trying to produce renderings for analysis that are as defensible in terms of accuracy as possible.

Actually, when used properly there is potential for mkillum to be more accurate, as I understand it. Its primary limitation is with specular louvers in daylight redirection systems, but Georg Mischler's Raydirect product can help with those types of systems. Presumably so could Roland Schregle's photon map plug-in to Radiance. Others on the list can explain it better than I.

I initially used rad to control everything, but I noticed that the -av parameter when I chose High for Quality, Variability, and Detail was .01 .01 .01. While that's close enough to 0 0 0, I wanted to be absolutely sure that the light levels were the result of the calculations, and not some a priori assumption. Thus, the rpict parameters I chose were based on the output of my rad file with some slight modifications. Is there a way to overide some of rad's intelligent guesses within the .rif file itself (specifically, the ambient value and ambient weight)?

There sure is. Simply add a line like this:

render= -av 0 0 0

... to your .rif file. This adjusted parameter is then appended to the end of the @options file I mentioned the other day, overriding what rad put in there.

···

----

      Rob Guglielmetti

e. [email protected]
w. www.rumblestrip.org

From: John An <[email protected]>
Date: Wed May 14, 2003 11:04:41 AM US/Pacific

Rob G. did a nice job answering most of your questions, so I'll just touch on the ones he missed...

Greg Ward replied:
You're correct -- the pattern affects the total material
reflectance/transmittance. To avoid this, run pfilt on your picture in
advance, like so:

pfilt -e 2 orig.pic > normalized.pic

The "normpat" script will do this for you, and blend the edges for
better tiling as well if that's what you have in mind.

To which I ask:
I'm not sure I understand how adjusting the exposure of an already processed
image will correct the reflectance of a material. Could you help me
understand this a bit more?

The Radiance colorpict primitive uses the final pixel values after exposure adjustment to modify the material color. Therefore, adjusting the picture so that the average is 1 (as pfilt -e 2 will do) avoids biasing your average value.

Final point regarding the publishing of Rendering with Radiance:
I'm sure it is a conscious decision not to digitize the entire book into a
print disabled .pdf, but could some electronic version be an option?

That's a good suggestion. I have the PDF for the book, which Rob Guglielmetti assembled from the pieces, and it should be easy enough to disable printing on it. Personally, I think the book is much less useful in electronic-only form, but if that's all you can get, it's certainly better than nothing. I'll confer with my coauthors about distributing the PDF. We are looking into print-on-demand publishers, and hope to have it available again to book resellers (Amazon, etc.) before the Fall. Until then, we might distribute it to people in unprintable form, though that seems really annoying in some respects.

-Greg

Greg Ward wrote:

> Final point regarding the publishing of Rendering with Radiance:
> I'm sure it is a conscious decision not to digitize the entire book
> into a
> print disabled .pdf, but could some electronic version be an option?

That's a good suggestion. I have the PDF for the book, which Rob
Guglielmetti assembled from the pieces, and it should be easy enough to
disable printing on it.

With ghostscript, I can print *any* PDF, whether the author
thought he had protected it from being printed or not. This so
called "print protection" is just a friendly hint within the
file, that the original Acrobat Reader respects, but any other
software will easily ignore.

Of course, that may or may not matter to you. If nobody wants to
get rich off the potential print-on-demand proceeds, then it's
probably irrelevant whether the same thing is also available in
digital form for free. Ordering a demand printed copy will still
be a lot cheaper than printing it at home for most people, let
alone binding (or glueing) the resulting heap of paper together.
But I don't know about the exact situation with the rights, so
you better check that out in detail before deciding anything.

-schorsch

···

--
Georg Mischler -- simulations developer -- schorsch at schorsch com
+schorsch.com+ -- lighting design tools -- http://www.schorsch.com/

Rob Guglielmetti wrote:

Why not switch over to "real time mode"? Despite the traffic lately,
it's usually pretty reasonable, traffic-wise.

                                  ^^^^^^^^^^^^

You been watching "The Apartment", Rob? :^)

···

--
END OF LINE. (MCP)