There are millions and millions of photos under the public domain, and for some they are nothing short of a gold mine. For instance, some scientists could find them most useful to compare things like glacial retreat or deforestation with what we’re seeing today or with results generated by models. Shifting through such a catalog is no easy feat, though.
Almost mockingly, Google not only shows that this isn’t half as challenging as it may seem, but also manages to turn image processing and sorting code into spectacular works of art. Using millions of photos scrapped from social networks like Flickr or Picasa, Goggle engineers made an algorithm that stitches them together to make 10,000 timelapse videos. Some are so accurate that you wouldn’t think for a moment each frame is actually a photo taken by some random, different person. Quite amazing, and a nice demonstration of what can be achieved in the future using other, much older data sets.
First, the algorithm sorts the photos so they’re from the same viewpoint, then a warping step follows, a bit of retouching and voila! Automatic stop-motion videos of various landmarks. It’s quite better actually than leaving your camera to shot in some deserted place for a whole year, not to mention the technical challenges.
The project will be featured at the annual Siggraph conference in Los Angeles from Aug. 9 to 13.
https://www.youtube.com/watch?t=93&v=wptzVm0tngc