rtrace result varies for each calculation

Dear Radiance experts,

I found that the rtrace output may differ every time I execute this command,
even if the oct file and the sensor point file are the same.

The rtrace command I'm using is:

cat SENSORS.pts | rtrace -I -aa .2 -ab 8 -ad 1024 -as 512 -h -w -oov
scene.oct | rcalc -e '$1=$1;$2=$2;$3=$3;$4=179*(.265*$4+.670*$5+.065*$6)' >
RESUTLS_lx.dat

in which:
1) "SENSORS.pts" describes the sensors' xyz coordinates and xyz vector,
2) "scene.oct" describes the geometry, material, and environment and sky
conditions (overcast sky with zenith radiance value set at 22.86)
3) "RESULTS_lx.dat" is expected to contain the xyz coordinates and
illuminance value of each sensor point.

However, I found that the output values in the RESULTS_lx.dat file varies
each time I execute this rtrace command in the Terminal in Ubuntu 10.04.

Is it normal? or is there anything wrong with my rtrace command parameters?

Hope you can help me on this!

Thanks!

Ji

Ji.

You can always expect some variations in Radiance calculations. The
trick is to modify your ambient parameters (in general options
starting with "-a") until you're satisfied that the error has no
relevant impact on the results. The "right" ambient parameters depend
on your scene complexity and and the rendering time you can afford to
spend.

A few further comments on your parameters are below:

The rtrace command I'm using is:

cat SENSORS.pts | rtrace -I -aa .2 -ab 8 -ad 1024 -as 512 -h -w -oov
scene.oct | rcalc -e '$1=$1;$2=$2;$3=$3;$4=179*(.265*$4+.670*$5+.065*$6)' >
RESUTLS_lx.dat

-aa - (ambient accuracy) seems a bit high (although I don't know
anything about your scene). A value of 0.2 allows for 20% error in the
ambient calculation. If the sky in your scene is hardly visible this
will lead to significant errors. In general I prefer values of 0.05
and less. Reducing this value will increase your rendering time. If
you set it to 0 you will turn ambient interpolation (and the
associated errors) off but it will take much longer to calculate.

-ab - (ambient bounces) is generous. If you have a simple scene where
the sky is visible from most locations you can safely reduce this to 6
or even 5 which will reduce your rendering time to a quarter of the
previous run. Note that your average results will increase a bit with
each additional bounce. You should do a series of calculations with
increasing -ab values to estimate the effect of one bounce more or
less.

-ad - (ambient divisions) could be a bit higher. I use 4096 these
days. It doesn't have a big impact on rendering time.
In a simple room with one or two large windows and no external shading
1024 will do. If you have lots of small windows or external shading
devises use a larger number.

-oov - You will find that the origin (second "o") is reported
differently from your input file. This is an effect of the -I setting
of rtrace. Don't worry about it. If you want to convert it you have to
reduce the z-value by 1.

However, I found that the output values in the RESULTS_lx.dat file varies
each time I execute this rtrace command in the Terminal in Ubuntu 10.04.

As stated above, some variation is normal. For you can add "-u-" to
your rtrace command line which changes the random sampling algorithm
used to a "low-discrepancy sequence". That and setting "-aa" to 0
should give you the most reproducible results. However, that alone
does not guarantee that the results are correct.

Homework:

Reduce your set of sampling points to a line from the front (window)
of the room to the back. Then start with your command line above and
modify the parameters as discussed. Do more than one calculation with
the same set of parameters and plot and compare the results. Also keep
a record of the rendering times.

A good set of options generates a smooth DF curve from the front of
the window to the back of the room that does not change a lot between
separate calculation runs. If the curve is not smooth your error is
too high and you have to increase/decrease some values.

Regards,
Thomas

···

On Tue, Jul 13, 2010 at 5:41 AM, Ji Zhang <[email protected]> wrote:

-ar - (ambient resolution) This may also improve accuracy by setting the -ar parameter higher than the default. This would help mainly if you have a large scene with a small area of interest (eg a model of an entire large office tower but seeking high level of accuracy for points within one small private office).

-Chris

···

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Thomas Bleicher
Sent: Tuesday, July 13, 2010 4:40 AM
To: Radiance general discussion
Subject: Re: [Radiance-general] rtrace result varies for each calculation

Ji.

You can always expect some variations in Radiance calculations. The trick is to modify your ambient parameters (in general options starting with "-a") until you're satisfied that the error has no relevant impact on the results. The "right" ambient parameters depend on your scene complexity and and the rendering time you can afford to spend.

A few further comments on your parameters are below:

On Tue, Jul 13, 2010 at 5:41 AM, Ji Zhang <[email protected]> wrote:

The rtrace command I'm using is:

cat SENSORS.pts | rtrace -I -aa .2 -ab 8 -ad 1024 -as 512 -h -w -oov
scene.oct | rcalc -e
'$1=$1;$2=$2;$3=$3;$4=179*(.265*$4+.670*$5+.065*$6)' > RESUTLS_lx.dat

-aa - (ambient accuracy) seems a bit high (although I don't know anything about your scene). A value of 0.2 allows for 20% error in the ambient calculation. If the sky in your scene is hardly visible this will lead to significant errors. In general I prefer values of 0.05 and less. Reducing this value will increase your rendering time. If you set it to 0 you will turn ambient interpolation (and the associated errors) off but it will take much longer to calculate.

-ab - (ambient bounces) is generous. If you have a simple scene where the sky is visible from most locations you can safely reduce this to 6 or even 5 which will reduce your rendering time to a quarter of the previous run. Note that your average results will increase a bit with each additional bounce. You should do a series of calculations with increasing -ab values to estimate the effect of one bounce more or less.

-ad - (ambient divisions) could be a bit higher. I use 4096 these days. It doesn't have a big impact on rendering time.
In a simple room with one or two large windows and no external shading
1024 will do. If you have lots of small windows or external shading devises use a larger number.

-oov - You will find that the origin (second "o") is reported differently from your input file. This is an effect of the -I setting of rtrace. Don't worry about it. If you want to convert it you have to reduce the z-value by 1.

However, I found that the output values in the RESULTS_lx.dat file
varies each time I execute this rtrace command in the Terminal in Ubuntu 10.04.

As stated above, some variation is normal. For you can add "-u-" to your rtrace command line which changes the random sampling algorithm used to a "low-discrepancy sequence". That and setting "-aa" to 0 should give you the most reproducible results. However, that alone does not guarantee that the results are correct.

Homework:

Reduce your set of sampling points to a line from the front (window) of the room to the back. Then start with your command line above and modify the parameters as discussed. Do more than one calculation with the same set of parameters and plot and compare the results. Also keep a record of the rendering times.

A good set of options generates a smooth DF curve from the front of the window to the back of the room that does not change a lot between separate calculation runs. If the curve is not smooth your error is too high and you have to increase/decrease some values.

Regards,
Thomas

_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general
____________________________________________________________
Electronic mail messages entering and leaving Arup business
systems are scanned for acceptability of content and viruses

Dear Thomas,

Thank you very much for your detailed explanation! We'll try out the
parameters according to your suggestions later.

May I also ask:
1) how are these parameters relates to model size?
2) The formula (6) in Axel's tutorial (p.33) is about the radius of the
"splotches". If the model is fairly large, e.g. 100m, then what would you
suggest for "aa" and "ar"?

Thank you again!

Ji

···

On Tue, Jul 13, 2010 at 4:40 PM, Thomas Bleicher <[email protected]>wrote:

Ji.

You can always expect some variations in Radiance calculations. The
trick is to modify your ambient parameters (in general options
starting with "-a") until you're satisfied that the error has no
relevant impact on the results. The "right" ambient parameters depend
on your scene complexity and and the rendering time you can afford to
spend.

A few further comments on your parameters are below:

On Tue, Jul 13, 2010 at 5:41 AM, Ji Zhang <[email protected]> wrote:
> The rtrace command I'm using is:
>
> cat SENSORS.pts | rtrace -I -aa .2 -ab 8 -ad 1024 -as 512 -h -w -oov
> scene.oct | rcalc -e '$1=$1;$2=$2;$3=$3;$4=179*(.265*$4+.670*$5+.065*$6)'
>
> RESUTLS_lx.dat

-aa - (ambient accuracy) seems a bit high (although I don't know
anything about your scene). A value of 0.2 allows for 20% error in the
ambient calculation. If the sky in your scene is hardly visible this
will lead to significant errors. In general I prefer values of 0.05
and less. Reducing this value will increase your rendering time. If
you set it to 0 you will turn ambient interpolation (and the
associated errors) off but it will take much longer to calculate.

-ab - (ambient bounces) is generous. If you have a simple scene where
the sky is visible from most locations you can safely reduce this to 6
or even 5 which will reduce your rendering time to a quarter of the
previous run. Note that your average results will increase a bit with
each additional bounce. You should do a series of calculations with
increasing -ab values to estimate the effect of one bounce more or
less.

-ad - (ambient divisions) could be a bit higher. I use 4096 these
days. It doesn't have a big impact on rendering time.
In a simple room with one or two large windows and no external shading
1024 will do. If you have lots of small windows or external shading
devises use a larger number.

-oov - You will find that the origin (second "o") is reported
differently from your input file. This is an effect of the -I setting
of rtrace. Don't worry about it. If you want to convert it you have to
reduce the z-value by 1.

> However, I found that the output values in the RESULTS_lx.dat file varies
> each time I execute this rtrace command in the Terminal in Ubuntu 10.04.

As stated above, some variation is normal. For you can add "-u-" to
your rtrace command line which changes the random sampling algorithm
used to a "low-discrepancy sequence". That and setting "-aa" to 0
should give you the most reproducible results. However, that alone
does not guarantee that the results are correct.

Homework:

Reduce your set of sampling points to a line from the front (window)
of the room to the back. Then start with your command line above and
modify the parameters as discussed. Do more than one calculation with
the same set of parameters and plot and compare the results. Also keep
a record of the rendering times.

A good set of options generates a smooth DF curve from the front of
the window to the back of the room that does not change a lot between
separate calculation runs. If the curve is not smooth your error is
too high and you have to increase/decrease some values.

Regards,
Thomas

_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general

--
ZHANG Ji 张冀 (PhD) :: Research Fellow :: Centre for Sustainable Asian Cities
:: School of Design and Environment :: National University of Singapore :: 4
Architecture Drive, Singapore, 117566 :: Contact: 65-6516 5046 :: Email:
[email protected]

Thank you very much, Chris!

Ji

···

On Tue, Jul 13, 2010 at 9:54 PM, Christopher Rush <[email protected] > wrote:

-ar - (ambient resolution) This may also improve accuracy by setting the
-ar parameter higher than the default. This would help mainly if you have a
large scene with a small area of interest (eg a model of an entire large
office tower but seeking high level of accuracy for points within one small
private office).

-Chris

-----Original Message-----
From: [email protected] [mailto:
[email protected]] On Behalf Of Thomas Bleicher
Sent: Tuesday, July 13, 2010 4:40 AM
To: Radiance general discussion
Subject: Re: [Radiance-general] rtrace result varies for each calculation

Ji.

You can always expect some variations in Radiance calculations. The trick
is to modify your ambient parameters (in general options starting with "-a")
until you're satisfied that the error has no relevant impact on the results.
The "right" ambient parameters depend on your scene complexity and and the
rendering time you can afford to spend.

A few further comments on your parameters are below:

On Tue, Jul 13, 2010 at 5:41 AM, Ji Zhang <[email protected]> wrote:
> The rtrace command I'm using is:
>
> cat SENSORS.pts | rtrace -I -aa .2 -ab 8 -ad 1024 -as 512 -h -w -oov
> scene.oct | rcalc -e
> '$1=$1;$2=$2;$3=$3;$4=179*(.265*$4+.670*$5+.065*$6)' > RESUTLS_lx.dat

-aa - (ambient accuracy) seems a bit high (although I don't know anything
about your scene). A value of 0.2 allows for 20% error in the ambient
calculation. If the sky in your scene is hardly visible this will lead to
significant errors. In general I prefer values of 0.05 and less. Reducing
this value will increase your rendering time. If you set it to 0 you will
turn ambient interpolation (and the associated errors) off but it will take
much longer to calculate.

-ab - (ambient bounces) is generous. If you have a simple scene where the
sky is visible from most locations you can safely reduce this to 6 or even 5
which will reduce your rendering time to a quarter of the previous run. Note
that your average results will increase a bit with each additional bounce.
You should do a series of calculations with increasing -ab values to
estimate the effect of one bounce more or less.

-ad - (ambient divisions) could be a bit higher. I use 4096 these days. It
doesn't have a big impact on rendering time.
In a simple room with one or two large windows and no external shading
1024 will do. If you have lots of small windows or external shading devises
use a larger number.

-oov - You will find that the origin (second "o") is reported differently
from your input file. This is an effect of the -I setting of rtrace. Don't
worry about it. If you want to convert it you have to reduce the z-value by
1.

> However, I found that the output values in the RESULTS_lx.dat file
> varies each time I execute this rtrace command in the Terminal in Ubuntu
10.04.

As stated above, some variation is normal. For you can add "-u-" to your
rtrace command line which changes the random sampling algorithm used to a
"low-discrepancy sequence". That and setting "-aa" to 0 should give you the
most reproducible results. However, that alone does not guarantee that the
results are correct.

Homework:

Reduce your set of sampling points to a line from the front (window) of the
room to the back. Then start with your command line above and modify the
parameters as discussed. Do more than one calculation with the same set of
parameters and plot and compare the results. Also keep a record of the
rendering times.

A good set of options generates a smooth DF curve from the front of the
window to the back of the room that does not change a lot between separate
calculation runs. If the curve is not smooth your error is too high and you
have to increase/decrease some values.

Regards,
Thomas

_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general
____________________________________________________________
Electronic mail messages entering and leaving Arup business
systems are scanned for acceptability of content and viruses

_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general

--
ZHANG Ji 张冀 (PhD) :: Research Fellow :: Centre for Sustainable Asian Cities
:: School of Design and Environment :: National University of Singapore :: 4
Architecture Drive, Singapore, 117566 :: Contact: 65-6516 5046 :: Email:
[email protected]

From Rendering With Radiance there are a few tables describing how the rad command calculates the various ambient settings. The –ar setting is determined by a formula a*d. The value of ‘a’ ranges from 8 for low accuracy to 64 for high accuracy. The value of ‘d’ is largest scene dimension divided by dimension of zone of interest.

So for example trying to get high accuracy for a model that’s 50x50x50m and looking for detailed results in a 5x5x5m room:

a * d = ar setting
64 * (50/5) = 640

That should give you a starting point, and if you still see too much variation between calculation you can try setting parameters even more rigorously.

···

From: [email protected] [mailto:[email protected]] On Behalf Of Ji Zhang
Sent: Tuesday, July 13, 2010 10:09 AM
To: Radiance general discussion
Subject: Re: [Radiance-general] rtrace result varies for each calculation

Dear Thomas,

Thank you very much for your detailed explanation! We'll try out the parameters according to your suggestions later.

May I also ask:
1) how are these parameters relates to model size?
2) The formula (6) in Axel's tutorial (p.33) is about the radius of the "splotches". If the model is fairly large, e.g. 100m, then what would you suggest for "aa" and "ar"?

Thank you again!

Ji

On Tue, Jul 13, 2010 at 4:40 PM, Thomas Bleicher <[email protected]<mailto:[email protected]>> wrote:
Ji.

You can always expect some variations in Radiance calculations. The
trick is to modify your ambient parameters (in general options
starting with "-a") until you're satisfied that the error has no
relevant impact on the results. The "right" ambient parameters depend
on your scene complexity and and the rendering time you can afford to
spend.

A few further comments on your parameters are below:

On Tue, Jul 13, 2010 at 5:41 AM, Ji Zhang <[email protected]<mailto:[email protected]>> wrote:

The rtrace command I'm using is:

cat SENSORS.pts | rtrace -I -aa .2 -ab 8 -ad 1024 -as 512 -h -w -oov
scene.oct | rcalc -e '$1=$1;$2=$2;$3=$3;$4=179*(.265*$4+.670*$5+.065*$6)' >
RESUTLS_lx.dat

-aa - (ambient accuracy) seems a bit high (although I don't know
anything about your scene). A value of 0.2 allows for 20% error in the
ambient calculation. If the sky in your scene is hardly visible this
will lead to significant errors. In general I prefer values of 0.05
and less. Reducing this value will increase your rendering time. If
you set it to 0 you will turn ambient interpolation (and the
associated errors) off but it will take much longer to calculate.

-ab - (ambient bounces) is generous. If you have a simple scene where
the sky is visible from most locations you can safely reduce this to 6
or even 5 which will reduce your rendering time to a quarter of the
previous run. Note that your average results will increase a bit with
each additional bounce. You should do a series of calculations with
increasing -ab values to estimate the effect of one bounce more or
less.

-ad - (ambient divisions) could be a bit higher. I use 4096 these
days. It doesn't have a big impact on rendering time.
In a simple room with one or two large windows and no external shading
1024 will do. If you have lots of small windows or external shading
devises use a larger number.

-oov - You will find that the origin (second "o") is reported
differently from your input file. This is an effect of the -I setting
of rtrace. Don't worry about it. If you want to convert it you have to
reduce the z-value by 1.

However, I found that the output values in the RESULTS_lx.dat file varies
each time I execute this rtrace command in the Terminal in Ubuntu 10.04.

As stated above, some variation is normal. For you can add "-u-" to
your rtrace command line which changes the random sampling algorithm
used to a "low-discrepancy sequence". That and setting "-aa" to 0
should give you the most reproducible results. However, that alone
does not guarantee that the results are correct.

Homework:

Reduce your set of sampling points to a line from the front (window)
of the room to the back. Then start with your command line above and
modify the parameters as discussed. Do more than one calculation with
the same set of parameters and plot and compare the results. Also keep
a record of the rendering times.

A good set of options generates a smooth DF curve from the front of
the window to the back of the room that does not change a lot between
separate calculation runs. If the curve is not smooth your error is
too high and you have to increase/decrease some values.

Regards,
Thomas

_______________________________________________
Radiance-general mailing list
[email protected]<mailto:[email protected]>
http://www.radiance-online.org/mailman/listinfo/radiance-general

--
ZHANG Ji 张冀 (PhD) :: Research Fellow :: Centre for Sustainable Asian Cities :: School of Design and Environment :: National University of Singapore :: 4 Architecture Drive, Singapore, 117566 :: Contact: 65-6516 5046 :: Email: [email protected]<mailto:[email protected]>
____________________________________________________________
Electronic mail messages entering and leaving Arup business
systems are scanned for acceptability of content and viruses