My empirical studies with MegaPOV radiosity have shown that it
works as advertised. There are some other factors that are more
important for realism than the radiosity settings.

1. assumed_gamma & Display_Gamma

   These parametrs relate only to the CRT and have nothing to do
   with simulating light transport and should be taken out of
   the simulation by using assumed_gamma 1.0. Display_Gamma should
   be set to a value correct for the particular monitor. 

2. Light source intensity distribution

   In real life there are no omnidirectional light sources. By this I mean
   that all ligh sources have some charachteristic intensity distribution.
   In this scene the light source is a flat square panel. It's projected
   area decreases by cos(angle) and so does it's brightness. For this reason
   just using area_light isn't enough because it still radiates in all
   directions with equal intensity (see fig.3). I used spotlight to simulate
   this cosine falloff. (tightness 0 produces cos(angle*2)/2+0.5 intensity)

3. fade_power

   In real life light source's apparent brightness follows 1/r^2 equation.
   The correct value for fade_distance seems to be 1.4*light_source_diameter.


All three factors must be taken into account if relism is the goal.
   
   
As proof I present 3 images (gamma 2.2):

  1) The Cornell reference image.

  2) MegaPOV 0.5 image rendered with with:
       assumed_gamma 1.0, Display_Gamma=2.2
       spotlight modifier
       fade_power 2

     This is a nice match with basically the default MegaPOV radiosity settings.
     All texture and radiosity parameters within nominal range.
  
  3) The same scene with same settings without:
       assumed_gamma
       spotlight modifier
       fade_power

     (Texture and light colors gamma corrected)

      This is a typical 1st attempt at radiosity. Too much color bleed,
      not enough diffuse light, flat shading.


Once again, the difference between images 2 and 3 is caused by
assumed_gamma, spotlight and fade_power. Source to binaries.scene-files
by Monday.

_______________________________________________________________________
Kari Kivisalo                                          www.kivisalo.net

Post a reply to this message

Attachments:
Download 'realistic.jpg' (26 KB)

Preview of image 'realistic.jpg'