On 05/16/2017 08:52 AM, clipka wrote:
> Am 16.05.2017 um 13:43 schrieb William F Pokorny:
>
> Did you notice any new artifacts whatsoever?
>

No. Unless somebody turns up some issue I think we should go with your 
update.

>
>> 1) We still sometimes get noise in the end result which I think AA will
>> mostly clean up. The nosiest example is yours from Github #147:
>>
>> https://github.com/POV-Ray/povray/issues/147
>>
>> see the attached: LipkaFS81_Apr2010_Example.png image.
>
> Note that this example, too, is a "planar" sphere_sweep, with all center
> points on a plane perpendicular to the camera direction. I'd suspect a
> connection there.
>

I'm certainly not claiming to be an expert, but the way I currently see 
it is that our solver is capable of handling everything up to the 
ray/plane (I think also ray axis alignments in play(1)) alignments - if 
we adjust current hard coded internal controls to allow it. See the 
attached image LipkaFS81_SolverTune.png where I have done this.

(1) The axis alignment issues are why I think we sometimes see seams in 
polynomial based shapes. Changing camera position, scale etc can all 
"solve" these seam artifacts, but the fundamental issue is the solver 
cannot reliably handle particular "alignments."

LipkaFS81_NearPerpAsPlane.png is an image, using a hack based on the 
code I posted earlier, showing rays near-ish perpendicular as a, clipped 
by sphere sweep, plane. So yes, we do have rays getting near 
perpendicular, but I think there is no fundamental issue with finding 
the roots until we are ray / plane / axis aligned within numerical 
accuracy. At that point the polynomial is "ill-formed" and we HAVE to do 
something different if we want reliable roots.

Use of the orthogonal camera will tend to bring out the ill-formed 
polynomial issues.

I've been looking some at our solver code and reading up, but theory, 
algorithms and code all new & a big stretch for me. I've no idea at the 
moment how to fix the fundamental ray / axis alignment issue in our 
solver. Perhaps it must be handled by analyzing and pre-conditioning the 
polynomials used in some fashion. Basically I don't know what I'm doing. 
:-)

As an aside AA itself sometimes fixes things given it changes ray / 
"object" relationships.

Apologies to all for the gibberish I sometimes write.

Bill P.

Post a reply to this message

Attachments:
Download 'lipkafs81_nearperpasplane.png' (13 KB) Download 'lipkafs81_solvertune.png' (31 KB)

Preview of image 'lipkafs81_nearperpasplane.png'

Preview of image 'lipkafs81_solvertune.png'