TED had a very cool demonstration of 3D photo aggregation from a company recently acquired by Microsoft.
The image here is a composite of all the photos on Flickr tagged with the words "Notre Dame". The software looks at these photos and does three things. It works out what Notre Dame looks like and builds a composite 3D image from bits of the photos themselves (represented here by the ghostly blue - white image). It also works out the vantage points and viewing angles of the photographers in the Parvis Square outside (the orange triangles). It then wraps them both together in a 3D zoomable interface that feels as though you're flying, and enables you to easily view any of the photos that make up the image. Breath-taking.
Computer algorithms are not great at understanding what photos depict: (e.g.is it a sunset, a honeymoon, a particular beach in Spain, or all three?). The current generation of photo sites has overcome this by enabling people to add descriptive words to their photos ('tags') so that other people can discover them. Flickr does this brilliantly, and it's underpinned its explosive growth.
What photosynth does, cleverly, is to take this to the next level. If you get an algorithm to look at photos tagged with the same words (e.g. Notre Dame) it'll be able to figure out that the photos are of the same object, and then reaggregate those photos in interesting ways.
It'll be interesting to see where this leads, but an easy and obvious application is to develop 3D maps based on actual photos that you can walk or fly around. Microsoft's already on that case. It also points a way forward for navigating user generated images once there are unfathomable quantities available.
You need to see the demo to really get this, and it's at the following livelabs link (and it's worth waiting for all the relevant plug-ins....)



