On Thu, 06 Jan 2005 22:18:02 -0500, Slime wrote:

> +AM2 starts by sampling the *corners* of pixels instead of the centers.
> Probably her eyebrows pass through the center of pixels but not the
> corners, so they're missed. This is a common problem (happens a lot with
> ropes or other thin objects in the distance) that happens with *both* AA
> methods, and has no good solution at this time.

Cone tracing. Instead of sending out rays, send out conical areas (one per
pixel), with no gap between them, and see which objects intersect the
cone and where. Now you know each object that should show in the color of
the pixel, and thus can ensure that you send rays to each and every one of
them (and that you won't be sending rays to empty space).

Of course, all this is likely to make current antialiasing methods seem
like greased lightning in comparison :(.

> Personally, I think it would be more convenient to have an object property
> that forces *higher* AA (more subsamples) for the current pixel when it's
> intersected. This could be used to solve the eyebrow issue.

Which, unfortunately, doesn't help with small objects, since the problem
is that the ray never intersects them...

...Unless, of course, this property would be numerical, and allow one to
make the bounding box of the object bigger (and simply intersecting the
objects bouding box would be enough to trigger higher AA).

Doesn't AA method 2 already do something like this ? It triggers more rays
when it finds a color edge. This new method 3 would need to trigger more
rays whenever it gets near a marked object (or all objects, with proper
scene settings or command line options).

Post a reply to this message