It has been a long time since I used the 2-pass method, and not yet with 
3.7, so what are the best settings nowadays? For my latest image  I have 
added a media, so 2-pass render becomes necessary I guess. At this 
moment and based on much earlier code, I use this, which is certainly 
not the most efficient I guess:

//start code
     radiosity {
       #if (RadSave)
         pretrace_start 0.08
         pretrace_end   0.004
         count 50, 1000
         error_bound 0.5
         always_sample on
       #else
         pretrace_start 1
         pretrace_end   1
         count 50, 100
         error_bound 1
         always_sample off
       #end  //of RadSave
       brightness 1
       recursion_limit 2
       nearest_count 10, 5
       adc_bailout 0.01/2
       low_error_factor .3
       minimum_reuse 0.015
       maximum_reuse 0.1
       gray_threshold 0.0
       normal on
       media off
       //max_sample 1.0
     }
//end code

During first pass, I switch off area light and any transparent textures. 
Can I switch off the media too? I don't remember about that.

Thomas

Post a reply to this message

> It has been a long time since I used the 2-pass method, and not yet with
> 3.7, so what are the best settings nowadays? For my latest image I have
> added a media, so 2-pass render becomes necessary I guess. At this
> moment and based on much earlier code, I use this, which is certainly
> not the most efficient I guess:
>
> //start code
> radiosity {
> #if (RadSave)
> pretrace_start 0.08
> pretrace_end 0.004
> count 50, 1000
> error_bound 0.5
> always_sample on
> #else
> pretrace_start 1
> pretrace_end 1
> count 50, 100
> error_bound 1
> always_sample off
> #end //of RadSave
> brightness 1
> recursion_limit 2
> nearest_count 10, 5
> adc_bailout 0.01/2
> low_error_factor .3
> minimum_reuse 0.015
> maximum_reuse 0.1
> gray_threshold 0.0
> normal on
> media off
> //max_sample 1.0
> }
> //end code
>
> During first pass, I switch off area light and any transparent textures.
> Can I switch off the media too? I don't remember about that.
>
> Thomas

An absorbing or scattering media does cast shadows. Emissive media 
illuminate it's surounding. It's not neutral. Same thing for any fog, 
it's not neutral.
If you use non-neutral media, you NEED media on in at least the first 
pass, preferably in both.
You can remove emissive media if you never use media on.
If you turn off your media for the first pass and use it for the second 
pass, then you realy need to do some pretrace steps in your second pass, 
and need to keep the same error_bound.

Don't switch off transparent textures unless they are totaly, or close 
to be, transparent, colourless (rgbt 1 or rgbf 1) and the object don't 
have any ior.

count should be the same for both passes.

Post a reply to this message

Thomas de Groot <tenDOTlnDOTretniATtoorgedDOTt> wrote:

>      radiosity {
>        #if (RadSave)
>          pretrace_start 0.08
>          pretrace_end   0.004
>          count 50, 1000
>          error_bound 0.5
>          always_sample on

In his Radiosity Voodoo articles (povray.text.tutorials), clipka recommends
setting always_sample to off.

>        #else
>          pretrace_start 1
>          pretrace_end   1
>          count 50, 100
>          error_bound 1

In 3.6, error_bound should be the same for both passes, otherwise the 2nd pass
would take eons.  I haven't tested this in 3.7.

>          always_sample off
>        #end  //of RadSave
>        brightness 1
>        recursion_limit 2
>        nearest_count 10, 5

In his Radiosity Voodoo articles, clipka recommends setting nearest_count to 1,
especially during the final render.  However, the current documentation suggests
that this would have poor results.  Although I can't recall the details, I
remember that I didn't get satisfactory results by tweaking this.

>        adc_bailout 0.01/2
>        low_error_factor .3
>        minimum_reuse 0.015
>        maximum_reuse 0.1
>        gray_threshold 0.0
>        normal on
>        media off
>        //max_sample 1.0
>      }

Post a reply to this message

Am 27.09.2011 23:52, schrieb Cousin Ricky:
> Thomas de Groot<tenDOTlnDOTretniATtoorgedDOTt>  wrote:
>
>>       radiosity {
>>         #if (RadSave)
>>           pretrace_start 0.08
>>           pretrace_end   0.004
>>           count 50, 1000
>>           error_bound 0.5
>>           always_sample on
>
> In his Radiosity Voodoo articles (povray.text.tutorials), clipka recommends
> setting always_sample to off.

Those recommendations are for using radiosity in a single render. When 
using a simplified scene to generate & save radiosity data, to be loaded 
& used later in a separate render, "always_sample on" is ok; the 
pre-render's main phase is likely to suffer artifacts, but it is 
discarded anyway, and this way it's still good for something (namely to 
serve as yet another radiosity pretrace step).

>
>>         #else
>>           pretrace_start 1
>>           pretrace_end   1
>>           count 50, 100
>>           error_bound 1
>
> In 3.6, error_bound should be the same for both passes, otherwise the 2nd pass
> would take eons.  I haven't tested this in 3.7.

It shouldn't be set /lower/ than in the first pass, but a higher value 
should be ok. Some care is recommended though: Increasing error_bound 
also increases the risk of light "leaking" through walls in corners.

>>           always_sample off
>>         #end  //of RadSave
>>         brightness 1
>>         recursion_limit 2
>>         nearest_count 10, 5
>
> In his Radiosity Voodoo articles, clipka recommends setting nearest_count to 1,
> especially during the final render.  However, the current documentation suggests
> that this would have poor results.  Although I can't recall the details, I
> remember that I didn't get satisfactory results by tweaking this.

I'm not so sure about this recommendation any longer. For the main pass 
in a 2-pass approach it doesn't actually matter much: In the main pass 
pretrace, a pretrace_start and pretrace_end setting of 1 will make sure 
that at most one single additional sample will be taken (if any 
pretracing is done at all; I'm not perfectly sure about that right now), 
which doesn't make much of a difference; and with always_sample off the 
nearest_count setting is irrelevant during the render proper anyway.

Post a reply to this message

Thanks guys! Basically I seem to be on the right track then. Also 
putting aa off in the first pass helps of course.

I was aware of Clipka's Voodoo, but was not sure about its use with a 
2-pass render.

Thomas

Post a reply to this message

On 27/09/11 16:19, Thomas de Groot wrote:
> During first pass, I switch off area light and any transparent
> textures. Can I switch off the media too? I don't remember about
> that.

   I would, for sure. I also make simplified versions of complex textures
wich render slow (heavily layered ones, for example).

--
Jaime

Post a reply to this message

On 28-9-2011 10:24, Jaime Vives Piqueres wrote:
> On 27/09/11 16:19, Thomas de Groot wrote:
>> During first pass, I switch off area light and any transparent
>> textures. Can I switch off the media too? I don't remember about
>> that.
>
> I would, for sure. I also make simplified versions of complex textures
> wich render slow (heavily layered ones, for example).

With what Alain says about scattering media, we could also use a very 
low sample value during first pass and a high one during second pass, 
couldn't we?

Thomas

Post a reply to this message

> On 28-9-2011 10:24, Jaime Vives Piqueres wrote:
>> On 27/09/11 16:19, Thomas de Groot wrote:
>>> During first pass, I switch off area light and any transparent
>>> textures. Can I switch off the media too? I don't remember about
>>> that.
>>
>> I would, for sure. I also make simplified versions of complex textures
>> wich render slow (heavily layered ones, for example).
>
> With what Alain says about scattering media, we could also use a very
> low sample value during first pass and a high one during second pass,
> couldn't we?
>
> Thomas

A low sample value for the media will miss most fine details, but, it's 
OK as the shadow will be difuse and effectively hide the fine details.

samples 3 to 5 should be correct for the first pass. It defaults to 10, 
and you may need more than 100 for the final render.

Post a reply to this message