"jr" <cre### [at] gmailcom> wrote:

> I've tested the code with a handful of different objects, and so far, so good.
> perhaps someone else will find use(s) for it too.  it is a work in progress
> though, and it would be nice if one or two wanted to be "guinea pigs", beta
> test, and help find bugs/problems, suggest improvements.

Sounds nice.
It brings to mind some of the stuff I've played with concerning isosurfaces,
parametrics, creating meshes from objects in SDL, and working with the camera
view frustum.

First, maybe add the option of making two wireframe AA-BB's - the original, and
the optimized so that the result can be seen in a render.

Second - this is dealing with axis-aligned bounding boxes, which immediately
brings up a million questions about what if I have a long, narrow box that's at
45 degrees to 2 axes .... is there some way to optimize the orientation of the
object itself.
Especially if you try to apply this to an infinitely long cylinder, or plane,
or....

I always get the feeling that we might be trying to reinvent some wheel that may
exist - somewhere, out there - in RenderMan or Blender, or Grasshopper, or Maya,
or....

I used to shim nuclear magnetic resonance spectrometers to get a good clean
spectrum and a high signal-to-noise ratio.  But there were lots of knobs for
strange "axes" of the magnetic field, and we only had a "trace" signal that went
up or down.
(28 shims available, which are on-axis: Z1, Z2, Z3, Z4, Z5, Z6 and off-axis: X,
Y, XZ, YZ, XZ2, YZ2, XZ3, YZ3, XZ4, YZ4, X2-Y2, XY, (X2-Y2)Z, XYZ, (X2-Y2)Z2,
XYZ2, (X2-Y2)Z3, XYZ3, X3, Y3, X3Z, Y3Z
So we were flying blind around this N-dimensional field, and you had to stay
aware of the fact that just because the signal went up didn't mean for sure that
the shim settings were getting better - because it might just be a local maximum
that wasn't as high as the global maximum, and you might want to traverse a
region in a local minimum in order to get over to that global maximum - so
sometimes a weaker signal was good - in the short run.

"So what, Bill, you incoherent maniac?"

;)  Well, - what if we try to calculate a bounding _sphere_ from a set of
trace() results, and then run another series of traces or insidedness tests when
we squish that sphere into a spheroid or ellipsoid?
Graph those results.   Look at them.   I'm sure they'll say something.
Just like the trace level on the NMR, we'd be looking for a global _minimum_ on
each of the axes of the ellipsoid, that tightly girdle the shape.
Then you'd have 3 axes of the ellipsoid, and could orient those axes to the
global axes, and THEN do the AABB.

I really need to go through things like the Graphics Gems series, (and IIRC
there is a GPU Gems series as well,) that may have some great algorithms that we
can use.

Post a reply to this message