I don't know if it's a bug or a (strange I must say) feature. The docs say
that 100 dispersion_samples or up are needed for very smooth results: but
how much RAM do you need?! I did some simple tests with ONE simple sphere
and ONE simple lightsource. The results were quite discouraging:

without dispersion: peak memory used: 607 kb
dispersion_samples 7: peak_memory used: 3.7 MB
dispersion_samples 20: peak_memory used: about 50 MB!

I tried with dispersion_samples 100, but had to stop after many minutes of
swapping.
It looks like the (real) photons count is somehow related to the
dispersion_samples: the greater is this value the greater the number of
photons stored.

What is more, as you increase the dispersion_samples value the caustics
start to look brighter and brighter (no matter how many-few photons you're
using) and thus quite unrealistic.

Here's the sample code I played with:

global_settings
{
max_trace_level 10
photons {count 200000 autostop 0}
  }

#declare Location = <-2, 3, -4>;
#declare Look_at = <0, 0, 0>;

camera
{
  location  Location
  look_at   Look_at
}

sky_sphere { pigment { rgb <0.6, 0.8, 1> } }

light_source
{ <4, 5, -5>
  rgb <1, 1, 1>*2.5
  fade_power 2
  fade_distance 5
 }

plane { y, 0
 pigment {color rgb <1, 1, 1>}
 finish { ambient 0 diffuse 0.5 specular 0.2 roughness 0.015 reflection
0.2 }
 }

sphere {
    y,1
    scale <0.5,1,0.5>
    pigment {rgbf 1}
    finish {ambient 0 diffuse 0.05 specular 1 roughness 0.003 reflection
{0.1,1 fresnel on} conserve_energy}
    interior {
        ior 1.5
        fade_power 1000
        fade_distance 1.5
        fade_colour color rgb <0.35, 0.42, 0.85>
        dispersion 1.1 dispersion_samples 20
    }
    photons {target reflection on refraction on collect off}
}


//-------
POV-Ray 3.5 beta 7 both compiles. Athlon Windows ME.

--
Jonathan.

Post a reply to this message

In article <3bfad725@news.povray.org>, jrg### [at] hotmailcom says...

> What is more, as you increase the dispersion_samples value the caustics
> start to look brighter and brighter (no matter how many-few photons you're
> using) and thus quite unrealistic.

Yep. I already reported this bug a while ago. 
I wonder why it hasn't been fixed yet. The solution is probably as easy 
as dividing a brightness/color value through the value of 
dispersion_samples?

Lutz-Peter

Post a reply to this message