> and just one light
> light_source{
> 0, rgb 3
> area_light 20*x,20*z,4,4 jitter adaptive 1 orient circular
> fade_distance 100
> fade_power 2
> translate<-198.2784,250,405.9303>}
>

You can improve that area_light.
As it is, when you use antialiasing, you have a high likelyhood that 
each subsamples be different, pushing the subsampling to it's limit.

I think that it's beter to use a high density array instead:

area_light 20*x,20*z,65,65 jitter adaptive 1 orient circular

You can even use larger values, like: 129,129 or 257,257
Optimum values follow the powers of two incremented by one making them 
odd. This is due to the way the adaptive algorythm works. It adaptively 
splits the interval between samples by 2 up to the minimal length set by 
the setting.

Here, the jitter introduce a small noise that is probably under the 
antialiasing threshold. The densest the array, the smaller the 
differences an finer the noise.

Also, there are times when using more than one light is advisable.

Other points:
Do you use the metalic textures from metals.inc? Those have a LOT of 
ambient! They are not suitable for any radiosity scene. They used 
ambient as a cludge to look acceptable in mostly empty scene with mostly 
nothing to reflect.
I suspect that something similar may also affect the various glass 
textures from glass.inc and glass_old.inc.

In your radiosity block, you can easily get values that fall oustide the 
supported or acceptable or sane range.

With rad_quality = 1; you only get one radiosity sample. NOT a sane 
value! This will result in extremely splotchy render, that can change 
radicaly by moving the camera by a very small amount.
Even rad_quality = 5; gives only 25 samples, whitch is usualy to little, 
especialy with an error_bound 0.5/5 (0.1)

rad_quality = 10; gives 100 samples, error_bound 0.05 nearest_count 20 
(the largest possible value)
Such a setting can realy cause some artefacts. Not quite enough samples 
for the error_bound, and the averaging of up to 20 samples is probably 
not enough to compensate.

To get enough samples for the quality 10 setting, you'd probably need 
rad_quality = 20; for 400 samples, but error_bound 0.025 and 
nearest_count 30 (out of range, silently reduced to 20).

Unless you use version 3.7 beta 40, rad_quality = 40; is the largest you 
can use. samples 1600 (maximum value) error_bound 0.0125 nearest_count 20.

I agree with clipka analysis.



Alain

Post a reply to this message