Hi Hugo:

> I am very impressed by this! Both the lighting and modelling. There are
> many well defined objects. I suppose you've made them over the years..Your
> homepage has many objects too.. great stuff!  :o)  But 30 hours rendertime
> is scaring, a little..

  Thanks! Yes, most of the objects are "stock" objects, some other are 
still placeholders. And 30h is very fast, really. Without saved radiosity 
data it would take at least a 50% more. At first, it was even more slow 
because I've put the lights inside CSG diference objects. Now, with unions, 
it renders much faster (I must remember this).

> Will you explain the lighting technique for us, once
> you are satisfied with it? 

  Yes, it is an include file with a very short macro and some constants. It 
is mostly based on using very strong intensities and very short 
attenuation.  Not a try to get something "physically correct", but 
arbitrarily based on real color and lumens data to mantain a realistic 
relation between diferent lights in the same scene. Basically, you call the 
macro this way:

  lamplight(COLOR_INCANDESCENT,LUMENS_INCANDESCENT_60W)

  or 

  lamplight(COLOR_FLUOR_UNIVERSAL_WHITE,LUMENS_FLUOR_18W)

  and it returns a light_source properly adjusted, taking into account two 
global constants previously defined: REFERENCE_WHITE and 
MAXIMUM_LUMENS_ALLOWED. It's up to you to build the light container (bulb, 
lamp, etc...).

  Pretty simple in code, but I expended some months understanding some 
concepts about light. I will show it soon...

> Did you find out why radiosity artifacts were
> gone when using a saved rad file?  

  Oh! Yes... I followed the advice from Kari of using this setup for the 
load scene:

  radiosity{
    pretrace_start 1 pretrace_end 1
    always_sample off
    load "rad_file"
  }

  But as this is using default error bound, wich was greater than the one I 
used for the saved data, the final render "smoothed" the artifacts. Seems 
that when loading rad data error_bound is taken into account. For ideal 
results with high quality settings you should use also the same error bound 
as in the "dummy" render. But it also helps to get "quick-but-clean" 
results with radiosity.

> Does it really work to use a rad file
> taken from a lower-resolution image?

  I tried it with the "official" cornell-box scene done by Kari (very 
good!), and it shows very similar results with saved data from a previous 
render at 1/3 resolution. Substracting the original image from the "tricky" 
one shows an almost black image (I had to increase brightness to see the 
diferenced zones). Of course, the original image is better, much more 
smooth, but this is still a great trick for lazy people like me (or you...;)

  Bye!

-- 
Jaime Vives Piqueres

La Persistencia de la Ignorancia
http://www.ignorancia.org

Post a reply to this message