So some researchers at the University of Washington have made a programme which digs up photos from the internet and turns them into videos, showing how landmarks change over a period of years. You can watch glaciers melt, or skyscrapers rise; rocks erode, and lakes dry out. They call this process, which is entirely automated, “time-lapse mining”.
We’ve been trying to come up with an intelligent comment on the value or meaning of such an exercise, and the best we’ve come up with is, “Cooooooooool”.
Here’s how it works. The researchers downloaded some 86m geo-tagged photographs. Because people love to take pictures of the same sorts of things, these will be disproportionately clustered around certain spots: stunning views, tourists attractions and so on.
The “mining” process then collects together photographs taken from roughly the same position, sorts them by date, and warps the images slightly so that they’re all facing the right way. After that it “regularises” them, to smooth out things like flickering lighting and so on.
Once all that’s happened, you’re left with videos showing how places and landmarks change over a period of years. You can see plaques gradually become unreadable due to water stains, then be suddenly cleaned up again. You can see archaeological sites rise from the ground in seconds as excavations proceed, or the greenery of San Francisco’s Lombard Street steps bloom and then die off again.
We’re not really doing this any justice here. Look, just watch the video:
The process doesn’t always work. Sometimes things moves too much, and everything is a blur. Other times lighting is inconsistent, leaving the world in a weird twilight. Oddest of all, the Wall Street Bull moves around, as if it’s about to come to life like that bit in Ghostbusters. Don’t ask why, we have no idea.
The researchers plan to publish further results, and the code of their mining programme shortly. In the mean time, you can read more about the project, and read their full paper, here.This article is from the CityMetric archive: some formatting and images may not be present.