I followed Kari Kivisalo's suggestions and rendered a 320*240 version of the
scene without ior, without normal and with both diffuse 0 and ambient 0. I
used save_file inside the radiosity block and then I reload the file as you
can see below:

 radiosity {
      pretrace_start 1
      pretrace_end   1
      count 400
      nearest_count 8
      error_bound 0.15
      recursion_limit 3
      low_error_factor 0.5
      gray_threshold 0
      minimum_reuse 0.015
      brightness 2.5
      adc_bailout 0.01/2
      always_sample off
      load_file "still_rad_data"
}

Well, the result isn't *perfect* (you can see several artifacts, esp on the
table planks and at the lowest edge of the room), but render time was much
more friendly.
The 320*240 image took 22 minutes to render, while the final version took
about 7 hours and 45 minutes (vs 3 days the other one would have taken).
Yet there are many aspects that puzzle me:

1) Why using load_file and always_sample off the render time is still much
slower than without radiosity?
2) Shouldn't you get the same results in spite of the settings used? (When
you're using load_file).
3) Why stopping and then resuming the render caused that discontinuity (with
pretrace_start 1, pretrace_end 1, load_file and always_sample off this
shoudn't occur)
4) What caused those artifacts on the table (there weren't artifacts in the
version rendered for saving data)?

IMO some bugs are involved here, esp. in the saving/loading process. Once I
have practised enough with C I will have a look to the code and find a
solution... but at the moment I'm too busy with my ultra-sofisticated
*Hello* programs :-)
TIA for those who will have the patient to answer all my questions,

--
Jonathan.
<Computer, kill Ned Flander...>

Post a reply to this message

Attachments:
Download 'rad.jpg' (71 KB)

Preview of image 'rad.jpg'