the 3d recognition seems quite robust on the boulder. what algorithm does photosynth use? is it SIFT? or SIFT related? something unique? is there a public paper about it?
answered my own question. paper here: http://phototour.cs.washington.edu/ and it says SIFT is used. at least. is it still the basis for point matching? or have you improved it? or just built up systems to aid it for fewer mistakes
Dan, I've heard that Photosynth actually uses a Microsoft Research feature detector, rather than SIFT, but I wouldn't be surprised to find that their work builds on top of the ideas that SIFT pioneered. I don't know its name, but please let me know if you find out.
Here's a topic about the video that comes to mind that mentions this:
Swami Worldtraveler also found this topic interesting based on that same video. See his topic here:
Find a list of many different videos about Photosynth and related technologies here:
Lastly, although it's a different topic, I think you'd be interested in this as well: