"Carl Hoff" <hof### [at] wtnet> wrote in message news:401a73af@news.povray.org...
> > You might try setting the threshold very small - say 3/256. That way,
> > when the rays come back virtually identical (e.g., in the middle of
> > those big flat grid squares ;-) POV-Ray won't bother with AA. But
> > it should still do it for the bright red lines. Experiment!
> (Theoretically,
> > there's no point setting it below 1/256 - unless you increase the
> > number of bits per pixel. Not sure if any hardware out there can
> > display it though...)
>
> Lets take the simple case.  If we have a flat blue background and our only
> object is a very thin red cylinder, all the rays will come back blue
except
> the ones that the cylinder passes through.  Correct?  So if the test ray
> comes back from a pixel that contained the cylinder put the test ray
misses
> the cylinder it will not know that it needs to perform the AA even if the
> threshold is set at 1/256.  At least that was my thinking so I'm not sure
I
> follow your theoretical argument above.  I suspect I could render an image
> with a threshold of 1/256 and again with 0 and with the right image you'd
be
> able to see a difference without any fancy hardware.

OK, I see your point.

What I ment was... if you render an image with AA=1/256 and then again with
AA=1/512, you won't see a difference - unless you increase the bits/pixel
setting. However, you're quite right - setting AA = EXACTLY ZERO will make a
difference in the case of tiny objects (and other miniscule ray colour
variations).

>  But yes I do need to
> do more experimenting.  Its just slow going when it can take a week to
> render one picture.

Mmm... tell me about it! lol

> Ideally what I'd want to do I think is put a bounding box around the
> cylinder that is atleast several pixels wide (at whatever the final
> resolution is) and have a way to tell POV-Ray that any pixel that has a
ray
> that passes through that bounding box needs to use a given type of AA.
I'm
> rather sure that isn't doable in atleast the current version.  It might be
a
> nice thing to suggest to the powers that be though as something to
consider
> adding.  As I've only been ray tracing a few months I really don't know if
> that's a good suggestion or not to be honest.

Mmm... sounds good to me! But no, I don't think it's currently doable. Might
be an interestnig way to fix it though...

> > The thing to really watch out for is intersections of infinite planes,
> > where the intersection is finite. POV-Ray currently can't work that
> > out; if an intersection is between only infinite objects, POV-Ray
> > assumes the resulting shape is infinite too - i.e., EVERY ray is
> > tested against it! It the resulting object is actually small, manually
> > bounding it will probably help quite a bit. (On the other hand,
> > plane intersection tests aren't that slow to do... But they still add
> > up. ;-)
>
> I only have one infinite plane, which is the grid, and I don't have it
> making any intersections with anything be it another plane or a finite
> object.  However unless I'm missing something isn't the intersection
between
> two infinite planes always an infinite line.  And I don't think trying to
> render that line would work.  You'd have to give it some width for it to
> show up at all I think.

True - the intersection of *two* infinite planes would always be infinite.
But the intersection of, say, 4 planes could be finite. (A pyrimid, for
example!) Personally, a lot of my scenes involve generating irregular
"facetted" shapes by intersecting a few hundred planes. Slows things down a
touch if there's no manual bounding... (And then you have to figure out how
to get it RIGHT! :-S)

I don't know how your tank is constructed - if there are no infinite objects
making it up, the automatic bounding *might* be ok...

(Question to anyone else reading this... What does POV-Ray do with the
intersection of two large spheres which only slightly overlap?)

> > Next time you render, try using the Draw_Vistas=on
> > option. (You can read about it in the online help.)
>
> Thanks, I will.  This sounds like a very useful feature.  Thanks for
> pointing it out to me.

Well, you never know. ;-)

Thanks.
Andrew.

Post a reply to this message