Time-lapse photography is fascinating because it can reveal changes that transpire too gradually to observe in real time. The problem is that, well, it takes a long time.
Researchers from Google and the University of Washington have found an elegant way around that, at least for some of the world’s most-photographed landmarks and scenes. In a paper published online, the researchers show how publicly available images shot by countless amateur photographers over a period of years can be algorithmically transformed into beautiful time-lapse videos. They call the process “time-lapse mining.”
The researchers started by gathering 86 million time-stamped images publicly uploaded by various users of photo-sharing sites such as Google’s own Picasa and Panoramio. They used image-recognition software to automatically pick out thousands of “clusters” of photographs that all showed the same landmark, such as the Salute in Venice or the Mammoth Hot Springs at Yellowstone National Park. Then they developed algorithms to warp a subset of photos in each cluster to a common viewpoint and scale, and ordered those by time stamp.
Throw in a few image-stabilization techniques and correct for lighting differences, and voila: an automatically generated time-lapse video of each landmark that looks almost as if it were shot with a single camera. At the top of this post is the full video that the researchers published in conjunction with their paper.
“Whereas before it took months or years to create one such time-lapse, we can now almost instantly create thousands of time-lapses covering the most popular places on earth,” the researchers wrote in their paper. (Here is the PDF.) “The challenge now is to find the interesting ones, from all of the public photos in the world.”
Figuring out what’s interesting, you see, is a task that’s still beyond the ken of machine-learning algorithms. The Google and UW researchers had to go through the time-lapse videos themselves to determine which were worth highlighting in their paper. They homed in on several categories of subject, including waterfalls, seasonal changes in vegetation, geological changes, construction projects, and city scenes. Sprinkled through this post are a few of our favorites, in GIF form, including the one of Las Vegas' changing skyline below.
And here is the full video that the researchers published in conjunction with their paper, including a slew of other impressive time-lapses. It's very much worth watching.