pure monte carlo rendering tips

Hi folks,

I'm experimenting with irradiance cashing turned off, and would appreciate
any tips. In a test scene, I have some curious results where surfaces meet
at a corner or edge. See partial image here (note the short partitions
meeting the floor plane, you also see it along the roof structure
components):

Here are my rpict settings (most generated by rad program, overrides given
for ambient parameters):
-dp 4096 -ms 0.25 -ds .2 -dt .05 -dc .75 -dr 3 -ss 16 -st .01 -lr 12 -lw
1e-5 -ab 4 -aa 0 -av 0 0 0 -ar 800 -ad 3000 -as 1500 -ps 3 -pt .04

I'm guessing -ar doesn't matter, there may be some optimization tricks
between -ad and -as, and probably some really important variable I have
less of a grasp on...

Many thanks in advance!

MM

Hi, Mike, many you can refer to the Axel Jacobs' Radiance tutorial in which
some combinations of Radiance's ambient parameters are proposed:
http://www.jaloxa.eu/resources/radiance/documentation/index.shtml

Joe

Mike Martinez <[email protected]>于2015年1月27日星期二写道:

···

Hi folks,

I'm experimenting with irradiance cashing turned off, and would appreciate
any tips. In a test scene, I have some curious results where surfaces meet
at a corner or edge. See partial image here (note the short partitions
meeting the floor plane, you also see it along the roof structure
components):
Dropbox - Error - Simplify your life

Here are my rpict settings (most generated by rad program, overrides given
for ambient parameters):
-dp 4096 -ms 0.25 -ds .2 -dt .05 -dc .75 -dr 3 -ss 16 -st .01 -lr 12 -lw
1e-5 -ab 4 -aa 0 -av 0 0 0 -ar 800 -ad 3000 -as 1500 -ps 3 -pt .04

I'm guessing -ar doesn't matter, there may be some optimization tricks
between -ad and -as, and probably some really important variable I have
less of a grasp on...

Many thanks in advance!

MM

Mike,

I had this same problem in March 2013, and the non-intuitive solution
was to crank up "-ar". The higher that went, the smaller those halos
became.

Wow, -aa 0 -ab 4 AND -ad 3000? You must have a lot of time on your hands.

Mark

···

On 1/26/15, Mike Martinez <[email protected]> wrote:

Hi folks,

I'm experimenting with irradiance cashing turned off, and would appreciate
any tips. In a test scene, I have some curious results where surfaces meet
at a corner or edge. See partial image here (note the short partitions
meeting the floor plane, you also see it along the roof structure
components):
Dropbox - Error - Simplify your life

Here are my rpict settings (most generated by rad program, overrides given
for ambient parameters):
-dp 4096 -ms 0.25 -ds .2 -dt .05 -dc .75 -dr 3 -ss 16 -st .01 -lr 12 -lw
1e-5 -ab 4 -aa 0 -av 0 0 0 -ar 800 -ad 3000 -as 1500 -ps 3 -pt .04

I'm guessing -ar doesn't matter, there may be some optimization tricks
between -ad and -as, and probably some really important variable I have
less of a grasp on...

Many thanks in advance!

MM

Mike,
Out of curiosity, at what image size (pixels) are you rendering, and are you using pfilt to filter down to the final image? I believe rendering many extra pixels with lower –ad and filtering down by /5, /10, etc. is the typical approach for the sake of rendering time when setting –aa 0. I’m not an expert in this, just recalling things I’ve seen in the distant past. I wonder if this approach would also reduce the size of those rendering artifacts.

···

____________________________________________________________
Electronic mail messages entering and leaving Arup business
systems are scanned for acceptability of content and viruses

Joe - indeed, Axel's resources are fantastic.

Mark, thanks for the tip - I'll see how it goes. So much for what I thought
I knew about -ar... It turns out you can get a pretty nice -aa 0 rendering
with today's newer machines in about the same time a "traditional" run
(complete with sampling artifacts) would take on 7-8 year old hardware...
It's a bit of an apples to oranges test, but interesting to see
advancements in hardware performance reducing the need for indirect
calculation optimization.

Chris, it's a 3000px rendering pfilted down to 1000px. I'll test out a
larger resolution too and see if the artifacts scale with the increased
iamge size.

Thanks!
MM

···

On Tue, Jan 27, 2015 at 6:10 AM, Christopher Rush <[email protected] > wrote:

Mike,

Out of curiosity, at what image size (pixels) are you rendering, and are
you using pfilt to filter down to the final image? I believe rendering many
extra pixels with lower –ad and filtering down by /5, /10, etc. is the
typical approach for the sake of rendering time when setting –aa 0. I’m not
an expert in this, just recalling things I’ve seen in the distant past. I
wonder if this approach would also reduce the size of those rendering
artifacts.

____________________________________________________________
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

Chris's points about smaller -ad and larger image size oversampling
are spot on. I ran some tests about a decade ago and found that very
large images with relatively low -ad (and -as) generated better
renderings for a given compute time. Results with images are here:

http://markjstock.org/radmisc/aa0_ps1_test/final.html

This won't affect the artifact sizes (-ar will), but it's definitely
related to your task.

Mark

···

On 1/27/15, Christopher Rush <[email protected]> wrote:

Mike,
Out of curiosity, at what image size (pixels) are you rendering, and are you
using pfilt to filter down to the final image? I believe rendering many
extra pixels with lower –ad and filtering down by /5, /10, etc. is the
typical approach for the sake of rendering time when setting –aa 0. I’m not
an expert in this, just recalling things I’ve seen in the distant past. I
wonder if this approach would also reduce the size of those rendering
artifacts.

____________________________________________________________
Electronic mail messages entering and leaving Arup business
systems are scanned for acceptability of content and viruses