Edouard schrieb:

> Ahh, I didn't know that - I hadn't seen that kind of artefact before.

They seem to rarely look like this; usually they're more prominent near 
edges where sample density is particularly high, and then accumulate 
there to give the whole area a distinctively jagged noisy look. It is 
probably only when the sample density is low (due to comparatively flat 
geometry and/or a high error_bound) yet at the same time pretrace is 
insufficient for some reason (e.g. due to some sparse geometric or 
fake-geometric features, or always_sample set to "on") that they exhibit 
this particular look.

> always_sample off seemed to fix it, but I have to admit that I still find the
> exact behaviour of many of the radiosity parameters a bit of a mystery.

This does not really come as a surprise to me :-)

I still intend to get rid of at least /two/ of the parameters: 
nearest_count, which is partially redundant with low_error_factor when 
used with always_sample off, and always_sample which I find no reason 
whatsoever to set this to anything other than "off" (with the default 
value unfortunately being "on"), but work has stalled a bit on this.

Unfortunately, all the other parameters appear to be essential (or would 
need to be replaced with other ways of parameterizing the same thing). 
There are even some hard-coded parameters in the algorithm, a few of 
which I'm not perfectly happy hiding them from the user.


>> The radiosity statistics may give you a hint how good your pretrace
>> settings are. I recommend making sure that no more than half the total
>> number of samples is taken during the main render.
> 
> Ahh, OK - what am I looking for in the stats?

There should be a section like this:

--------------------------------------------------------------------------
   Pass     Depth 0    Depth 1    Depth 2           Total
--------------------------------------------------------------------------
   1             81         73          5             159
   2            156         36          -             192
   Final        256          4          -             260
--------------------------------------------------------------------------
   Total        493        113          5             611
   Weight     0.362      0.212      0.087
--------------------------------------------------------------------------

In this example, 81 samples were taken during first pretrace step, 156 
during the second, and another 256 samples were taken during the final 
render (at recursion depth 0; the other depths are not a problem as they 
cannot cause visible artifacts), so that 256 of 493 samples are taken 
during final render. This is a pretty ok ratio.

Post a reply to this message