Photosynth of Osaka Castle in Osaka, Japan and interior of Saint Peter’s Basilica in Rome, Italy
About two years ago, Microsoft showed a preview of Photosynth at the internal company meeting event, and wowed us all, but we weren’t allowed to talk about it. Then, almost a year later, the beta software was made available to employees, allowing us to play with it, but we could discuss or even show this outside of Microsoft. Well, last month, Microsoft finally released Photosynth to the world as a Microsoft Live Labs project, so I can finally blog about what a kick-ass awesome product this is, and publish two of my really synthy photosynths.
For those who don’t know, Photosynth is a software application that takes a set of digital photographs, and builds a three-dimensional cloud out of these pictures just through pure computational analysis. Without the user having to prepare or do anything at all, the software analyzes and determines how each photo relates to one another, provided that there is some overlap. The resulting mesh is then presented in a 3D world that allows you to navigate and zoom in and out with no limit on resolution or zoom factor. It’s hella cool, and absolutely stunning!
So, let me present to you to Photosynth that I painstakingly did on my last two vacations.
This is the interior of Saint Peter’s Basilica in Rome, Italy. While PT had a headache and was resting on a bench, I wandered around in there for two hours to take tons of pictures. From those, 291 pictures were reconstructed in this Photosynth that is 97% synthy.
This is Osaka Castle in Osaka, Japan. I took this earlier this year on my return trip from Vietnam. This Photosynth comprises of 621 photos and is 99% synthy. That’s a lot of photos! You can see the front, side and back of the Castle (yeah, only 180 degrees, I couldn’t go on the other side of the castle) .
This is just the start of what’s possible and I predict that this is going to make giant leaps in the world of artificial intelligence! Imagine taking a video of an object (say a bicycle). A potential software could decompose the video into pictures, feed it into Photosynth so that 3D cloud could be reconstructed. As a result, recognizing that the video includes a bicycle is then merely the same problem of optical character recognition (OCR)--a problem set that computer science has already solved fairly well (at least in the 2D world). This could also help with automatically tagging videos based on computational recognition. The software could identify all bikes in online videos, making a video search on ‘bikes’ actually work based on the content of the video and not on the honesty of people who provide the keywords.
