2 or 3 axes? I don't think so... I did some experimentation with booleans in Povray once and I think you really need a fairly large number of axes (i.e. photos) to get a reasonable approximation... Take a sphere, for instance. A sphere looks like a circle from all directions. Suppose you had 1 photo. You could end up with something as far off as a cylinder! 2 photos? OK, how about a cross shape made out of cylinders truncated in just the right way so that the curvature looks circular from those 2 particular angles. 3 photos? Same thing as 2, only with six projections instead of four. Get the picture?
Now, if this thingy manages to actually deduce contours from the shading on the physical object and construct the model accordingly, that is pretty nifty... that would greatly reduce the number of photos required, though it would be totally screwed up by multi-colored objects!
