"R. Suzuki" wrote:
> 
> That's impressive image!
> 

Thank you!

> 
> Have you tried "adaptive max_gradient" technique, which is already
> implemented in 3.5 beta?
> 
> I think "adaptive accuracy" is not so useful as "adaptive max_gradient",
> 
> The below code is an example.
> If you render it with accuracy=0.002 instead of 0.0001, you will
> see artifacts near the top of small peaks.
> On the other hand, if you render it with adaptive max_gradient
> (e.g. "evaluate 2, 1.2, 0.9" instead of "max_gradient 5") and with
> accuracy=0.0001, the image quality will be much better and the
> rendering speed will be comparable or rather faster than the above.
> 
> Note that this technique does not guarantee artifact-free image
> and requires experiences to determine the "evaluate" parameters.
> (So, this should be only for advanced users.)

Hmm, seems i don't understand completely how evaluate works, but i
remember in the past (with megapov) that scenes using 'eval' sometimes
were much faster than with a sufficient max_gradient value.  I will do
some more experiments on that...

NTL, what i meant with adaptive accuracy was adapting it to the distance
from viewpoint.  Of course it could be useful to adapt max_gradient in the
same way.

> BTW, have you tried normal function pattern like below
> (commented out) for the foreground?   I usually use it in such
> cases.
> 

No, but it's an interesting idea, anyway i suppose with 'normal on' in
radiosity this will be quite slow too.

Christoph

-- 
Christoph Hormann <chr### [at] gmxde>
IsoWood include, radiosity tutorial, TransSkin and other 
things on: http://www.schunter.etc.tu-bs.de/~chris/

Post a reply to this message