Photosynth (ted talk, demo page, our coverage), a project acquired by Microsoft, made for one of the coolest tech demos to grace the internet in a very long time. The experience that Photosynth provided by inferring 3d structures from collections of 2d pictures made for a very rich — and jaw dropping — user experience. But not rich enough, it seems: Microsoft has a competing project another such tool from some people on that team, developped jointly with the University of Washington. Introducing Photo Tourism (project homepage, /. coverage). Check out the video:
Like Photosynth, Photo Tourism assembles its 3d models from photos on flickr. Photo Tourism, however, allows you to not only add your own photos to a 3d set, but “walk” between the locations where your pictures were taken, virtually. Photo Tourism also allows a user to rotate their point of view around a landmark or object, and can even determine if photos of that landmark/object were taken during the day or at night, grouping them accordingly. This results in a very high “feels like you’re there” experience, but what’s the next step?
This isn’t a fatal blow for Photosynth, though. Microsoft, in what I consider to be a very smart move, has merged the Photosynth (also known as Seadragon) team into the Microsoft Virtual Earth team. Here is how I see this playing out: Photosynth is able to infer 3d structures from photographs and (should soon be able to) skin these structures dynamically from the very same photographs. If there was a large enough set photos, Photosynth could infer the shapes, sizes and positions of objects all over the world. Then, with Virtual earth, geotagging data, and the relations between photos, these 3d structures could be mapped into their Virtual Earth platform. Essentially, it should be possible to make a virtual copy of the earth, passively, by processing photos that already exist.
The user experience that this could provide would be even more mindblowing than the tech demos we’ve seen before. Imagine being able to “walk” through foreign locations that are always as current and up to date as the lmost recent picture uploaded to the master data set. With version control and date-tagging of photos, you could even watch how an arbitrarity bit of the planet changes over time, watching changes in season, volume of visitors, and even cleanliness and litter. We’ll have the world at our fingertips, and be free to peel back its layers and inspect it as deeply as we like.
Scary tangent: I see another application for this technology: people. If this technology can be used to infer 3d structures of buildings, why can’t the same be done for people? I have friends on facebook that are present in upwards of 1000 photos. That’s a lot of data from which to infer a model — very scary, especially given how realistic simulated facial animations have gotten. It’s pretty easy to foresee video-conferecing identity-theft for people who neglected to set their facebook photo privacy settings sufficiently high.
Edit: it already exists!