I figure that subject line will get some attention  :-)

Well, to explain myself, I checked 3 renders against each other (by eye, not
extensively) concerning render time and appearance.  I was stunned to find a
16 X slower render without any visible benefit from it.  A low count of 50
and recursion of 1 took a mere 19 minutes and 11 seconds.  Then using count
100 recursion_limit 2 it went to 5 hours and 15 minutes!  Without radiosity
at all was a 5 min 42 sec render.
The peak memory amount really jumps too.  I mean, wow, is that really needed
when the 6 minute render looks okay?  At most the 20 minute one...
I have to add that the 3rd image actually did finish but I stupidly
rerendered using the same filename to make sure about the non-radiosity one.
Luckily I had screen captured it before that.  Believe me, no noticeable
difference between the recursion 1 and 2 except for a texture change on the
chimney I did in the script.

Following are excerpts from those stats in WinMegaPov 0.5a:
----------------------------------------------------------------------------
No radiosity, it was skipped using #if(), statistics even so?
----------------------------------------------------------------------------
Radiosity samples calculated:      38461 (5.26 percent)
Radiosity samples reused:         692156
----------------------------------------------------------------------------
Smallest Alloc:                 26 bytes   Largest:            49200
Peak memory used:          5463401 bytes
----------------------------------------------------------------------------
Time For Parse:    0 hours  0 minutes   1.0 seconds (1 seconds)
Time For Trace:    0 hours  5 minutes  41.0 seconds (341 seconds)
    Total Time:    0 hours  5 minutes  42.0 seconds (342 seconds)
Total cache size: 16880
----------------------------------------------------------------------------
----------------------------------------------------------------------------
Radiosity with count=50, recursion=1, error bound=0.36
----------------------------------------------------------------------------
Radiosity samples calculated:      38461 (5.26 percent)
Radiosity samples reused:         692156
----------------------------------------------------------------------------
Smallest Alloc:                 26 bytes   Largest:            49200
Peak memory used:         11965226 bytes
----------------------------------------------------------------------------
Time For Parse:    0 hours  0 minutes   1.0 seconds (1 seconds)
Time For Trace:    0 hours 19 minutes  10.0 seconds (1150 seconds)
    Total Time:    0 hours 19 minutes  11.0 seconds (1151 seconds)
Total cache size: 16880
----------------------------------------------------------------------------
----------------------------------------------------------------------------
Radiosity with count=100, recursion=2, error bound=0.321
----------------------------------------------------------------------------
Radiosity samples calculated:     410694 (4.31 percent)
Radiosity samples reused:        9126101
----------------------------------------------------------------------------
Smallest Alloc:                 26 bytes   Largest:            49200
Peak memory used:         77512206 bytes
----------------------------------------------------------------------------
Time For Parse:    0 hours  0 minutes   1.0 seconds (1 seconds)
Time For Trace:    5 hours 10 minutes   9.0 seconds (18609 seconds)
    Total Time:    5 hours 10 minutes  10.0 seconds (18610 seconds)
Total cache size: 16880
----------------------------------------------------------------------------

I know there are scenes that do wonderfully but this apparently was a flop.

Bob
--
omniVerse http://users.aol.com/persistenceofv/all.htm

Post a reply to this message

Attachments:
Download 'lhgroundlevelr2.jpg' (19 KB) Download 'lhgroundlevel.jpg' (20 KB) Download 'lhgroundlevelr.jpg' (20 KB)

Preview of image 'lhgroundlevelr2.jpg'

Preview of image 'lhgroundlevel.jpg'

Preview of image 'lhgroundlevelr.jpg'

Sorry, they posted out of order even though I tried to prevent that.
1st is the radiosity with recursion 2, 2nd is without any, 3rd is recursion
1.

Post a reply to this message

Bob Hughes wrote:
> 
> I figure that subject line will get some attention  :-)
> 
> Well, to explain myself, I checked 3 renders against each other (by eye, not
> extensively) concerning render time and appearance.  I was stunned to find a
> 16 X slower render without any visible benefit from it.  A low count of 50
> and recursion of 1 took a mere 19 minutes and 11 seconds.  Then using count
> 100 recursion_limit 2 it went to 5 hours and 15 minutes!  Without radiosity
> at all was a 5 min 42 sec render.
> The peak memory amount really jumps too.  I mean, wow, is that really needed
> when the 6 minute render looks okay?  At most the 20 minute one...
> I have to add that the 3rd image actually did finish but I stupidly
> rerendered using the same filename to make sure about the non-radiosity one.
> Luckily I had screen captured it before that.  Believe me, no noticeable
> difference between the recursion 1 and 2 except for a texture change on the
> chimney I did in the script.
> 

Recursion limit above 1 is not necessary in many situations, but there has to be
something wrong IMO, because you should get at least a strong bluish touch
because of the sky in every radiosity version.  It could help if you would
explain the changes made between the radiosity/no radiosity version.  

Christoph

--
Christoph Hormann <chr### [at] gmxde>
Homepage: http://www.schunter.etc.tu-bs.de/~chris/

Post a reply to this message

"Christoph Hormann" <Chr### [at] schunteretctu-bsde> wrote in
message news:39B4F61C.19135F25@schunter.etc.tu-bs.de...
|
| Recursion limit above 1 is not necessary in many situations, but there has
to be
| something wrong IMO, because you should get at least a strong bluish touch
| because of the sky in every radiosity version.  It could help if you would
| explain the changes made between the radiosity/no radiosity version.

Isn't much changed, well that's a white lie.  The ambient is dropped to near
zero for the radiosity images but that's a well known fact (I just tend to
leave it non-zero though, very low).  And the lighting is changed somewhat
as well as the sphere (not sky_sphere) for the sky because of the drastic
differences when ambient is dropped.  In other words, ambient and diffuse
for the sky is .6, .6  in radiosity and .67, .33 without.
The radiosity itself is done like so:

#if (Use_MP & Rad = yes)
 ambient_light <.1,.05,.025>
 ini_option "+qr"
 radiosity {
pretrace_start .08
pretrace_end .01
         media off
         normal on
         count 100 // or 50 // may be too low
         nearest_count 5
         error_bound .321
         low_error_factor .321
         gray_threshold .2
         recursion_limit 2 // or 1
         minimum_reuse .0167
        // brightness 1.333
          }
#else
 ambient_light <1,1.025,1.05>*.75
#end
}

Hope that tells you what you were wanting to know.

Bob

Post a reply to this message

Bob Hughes wrote:
> 
> #if (Use_MP & Rad = yes)
>  ambient_light <.1,.05,.025>
[...]
> #else
>  ambient_light <1,1.025,1.05>*.75
> #end

That's probably the reason for the missing additional blue in the radiosity
version.  

For the rest: I think the balconies look much better in the radiosity version,
because it generates some structures in the shadows, while the conventional
lighting only shows the plain ambient. 

Christoph 

--
Christoph Hormann <chr### [at] gmxde>
Homepage: http://www.schunter.etc.tu-bs.de/~chris/

Post a reply to this message

I don't know about the radiosity side of things, but the buildings
have really come along. 

-- 
Cheers
Steve              email mailto:ste### [at] zeroppsuklinuxnet

%HAV-A-NICEDAY Error not enough coffee  0 pps. 

web http://www.zeropps.uklinux.net/

or  http://start.at/zero-pps

  4:33pm  up 14 days, 20:51,  2 users,  load average: 1.00, 1.00, 1.00

Post a reply to this message

Bob, what happens if you use "ambient_light 0" for both/all three?

Grim

Post a reply to this message

Doh! Nevermind. I went back and read everything you wrote.

I think what is going on is you have adjusted the scene for similarity (re:
rad/no rad). MP rad really kicks in when you have no ambient components.
Sky_spheres, of course, influence the scene none-the-less (in my experiments
anyway), but everything else should be defined as ambient 0, unless you want
them to glow/radiate. Adding ambient_light really messes with things.

You are, of course, free to use artistic interpretation at your own whims!
:) I'm always playing with such things.

Grim

Post a reply to this message

Bob Hughes wrote:
> 
> Isn't much changed, well that's a white lie.  The ambient is dropped to near
> zero for the radiosity images but that's a well known fact (I just tend to
> leave it non-zero though, very low).  And the lighting is changed somewhat
> as well as the sphere (not sky_sphere) for the sky because of the drastic
> differences when ambient is dropped.  In other words, ambient and diffuse
> for the sky is .6, .6  in radiosity and .67, .33 without.

Set the sky ambient to 1 and diffuse to 0.

> #if (Use_MP & Rad = yes)
>  ambient_light <.1,.05,.025>

Yeah, I'd kill that ambient. If you're doing this right, you really
shouldn't need it, it'll just wash out whatever deep shadows you DO end
up with.

> pretrace_start .08
> pretrace_end .01
>          media off
>          normal on
>          count 100 // or 50 // may be too low
>          nearest_count 5
>          error_bound .321
>          low_error_factor .321
>          gray_threshold .2
>          recursion_limit 2 // or 1
>          minimum_reuse .0167
>         // brightness 1.333

Unless hyper-accurate shadowing is a necessity, and I don't see why it
would be for this scene, error_bound will be just fine between 0.6 and
1. Count 50 is plenty; a high count is best when you have small, bright
objects, or a very "noisy" scene, but you have a relatively simple one
with an entire lit hemisphere. You could even go down to 30. Set the
low_error_factor to 1, you're not going to need it (in my testing, it's
pretty useless). Recursion_limit 1 is plenty, I wouldn't mess with the
minimum_reuse, and brightness could go as high as 1.667 or so.

You could even set pretrace_start to .01 if you want to shave an extra
several seconds off; I haven't found that multiple passes help, provided
that the one pass you DO use is fine enough.

Try that and see how it goes. :)

-Xplo

Post a reply to this message

"Steve" <ste### [at] zeroppsuklinuxnet> wrote in message
news:slr### [at] zero-ppslocaldomain...
| I don't know about the radiosity side of things, but the buildings
| have really come along.

Thanks Steve.  And just when I'm ready to move on.

Bob

Post a reply to this message