Am 28.09.2011 10:17, schrieb Jaime Vives Piqueres:
> On 28/09/11 09:42, Reuben Pearse wrote:
>> Here's my latest render.
>>
>> Not sure if I've got the gold colour quite right - I'm using "lg_tan"

>> I tried "lg_pearl_gold" but it looked to shiny.
>
> It's difficult to say withouth something to reflect...
>
>> Any ideas how I can get rid of the strange effect at the base of the
>> legs?
>
> I would say these white artifacts are radiosity artifacts: try a smaller
> pretrace_end, perhaps....

They /are/ radiosity artifacts. Here's the essentials of how they come 
into existence:

When POV-Ray computes the radiosity brightness at a particular point 
from previously taken samples, it interpolates between multiple samples, 
weighing them according to how well it /expects/ them to be suitable for 
that particular point. Two of the primary criteria playing a role here are:

- how close the sample is to any other objects, or protrusions on the 
surface (such as the foot protrudes from the leg), and
- how close the sample is to the point in question.

As a matter of fact it is the ratio between these distances that is a 
weighing factor.

Unfortunately there is no simple way for POV-Ray to tell the distance to 
the next object, so what POV-Ray does is that it uses the 
sample-ray-shooting process to also probe for nearby surfaces. As a 
consequence, POV-Ray will fail to notice nearby objects or protrusions 
if none of the sample rays happens to hit it.

Normally this is not a big deal: For points close to a protrusion, such 
odd samples are usually far away, getting a much lower quality ranking 
than numerous other samples that have been taken much closer, and are 
therefore less prone to entirely missing the protrusion. So those odd 
samples are "drowned" in good samples.

However, artifacts /do/ occur at places where POV-Ray did not take /any/ 
good samples at all. That is what has happened between the legs and the 
feet. (Notice how the bright splotches only appear on the legs, while 
the feet are free: This is because the feet are short, and there's no 
room on them to take a sample that could miss the legs, while there's 
enough room on the leg and breast from where the feet could be missed.)

As for ideas how to solve this, here are a few that /might/ help:

- Increase count; this will reduce the likelyhood of protrusions (such 
as the feet) being missed. You can cause POV-Ray to just test for 
distance in these added rays by making use of the radiosity importance 
mechanism (needs POV-Ray 3.7). I doubt whether this will help much 
though, unless you crank up count really high.

- Decrease pretrace_end; this will cause a more thorough pretrace, 
reducing the likelihood of POV-Ray not taking any good samples at any 
particular place.

- Increase error_bound; this will increase the effective radius of all 
samples, causing samples taken /somewhat/ close to the feet to also be 
good /extremely/ close to them. I /think/ a value of 1.0 or greater 
should eliminate the problem (at the cost of potentially introducing 
others :-P), but I might be wrong.

- Increase minimum_reuse; this will only increase the effective radius 
of the "finest" samples, again causing samples taken /somewhat/ close to 
the feet to also be used /very/ close to the feet.

- Reduce maximum_reuse; this will decrease the effective radius of the 
"coarsest" samples, including the odd sample. (Very close to the feet, 
POV-Ray will then not find /any/ samples at all during the main render, 
and will take more to fill the gaps.)

- Set always_sample on; this will cause POV-Ray to take additional 
samples during main render in spaces where it only finds the odd sample. 
I wouldn't recommend it though, as it is prone to introducing other 
artifacts.

Other radiosity parameters probably have no effect on this issue.

Post a reply to this message