On 10/05/2018 03:04 PM, clipka wrote:
> Am 05.10.2018 um 20:33 schrieb Alain:
> 
>> I expected some message about the max_gradient found to be around 9 but
>> set at 1.0.
>> The resulting image show obvious holes.
>>
>> But, I don't get any message after the render is complete.
> 
> Can't reproduce:
> 
> - I only see an (almost perfect) clay-brown slab, even with the
> max_gradient 1 setting. No holes for me.
> 
> - Previous versions (tested with v3.7) do not produce a max_gradient
> warning for this scene either. [*]
> 
> - When I set max_gradient even lower (0.1), I do get holes, but I also
> do get a warning with alpha.9861167. [*]
> 
> - The reported maximum gradient found is 0.990, not around 9.
> 
> 
> [* Fun side note: At a max_gradient setting of 0.1, it is POV-Ray v3.7
> that fails to issue a warning.]
> 

I do see a few apparent holes (no AA) in the upper let corner for 
example, but otherwise agree with Christoph.

I think the bounding is off with z low being zero:

  contained_by {box {<-1,-1,0>, <1,1,1>}}

In other words, I think the function is mostly being clipped so we are 
seeing a slice of the function's "inside" and which on that slice is 
noisy at the edges. Change that zlow bounding 0 to -1, for example, and 
you'll get gradient warnings for the max gradient of 1.

Aside 1: Another way to clean up the edges in the clipped (zlow=0) case 
is to crank the accuracy way up so the clipped inside resolves more 
cleanly at the edges. Say maybe 0.00005 or less.

Aside 2: IIRC there is still a thread collision in the current 
implementation (which Thomas, in 3.8 no longer needs to be "naked" 
thanks to Christoph's updates) - meaning you are not guaranteed to see 
the very worst found gradient. In practice it's close. Practical 
implication is you need to bump any reported max a little for the 
setting not so much for result, but to be sure some later render doesn't 
report a slightly larger gradient just due the internal ordering of 
stuff.

Bill P.

Post a reply to this message