It might not be the most useful website ever designed, but Flickr’s new Park or Bird microsite serves a purpose: namely, it’s the Flickr Vision team’s response to a challenge by the popular webcomic XKCD.
This weird project began with the XKCD comic below:
Trying to get a computer to recognize objects in a photo is incredibly difficult, but it’s a nut that researchers have been starting to crack in earnest lately. And so, when Flickr saw this comic, they decided to take the unspoken challenge the comic presented, and show off just how well their system can differentiate between a national park, and a bird.
Some of us at Flickr saw that comic and thought, “… hey, we can do that!” Like the comic suggests, given a photo with geo data, it’s pretty straightforward to lookup whether that photo was taken in a US national park, and we already have the technology to answer the bird question (which took us less than 5 years to build, though it’s definitely a hard problem, and we’ve still got room for improvement (which we’re working on)).
So, all we had to do was whip up this spiffy website and put those technologies behind it! We think it’s a great way to show off some of the cool work we’re doing here at Flickr. We hope you think it’s fun!
The microsite, which you can visit here, will answer a simple question about any photo you drag into the box from your archives: is it a park, a bird, both, or neither?
Of course, answering the park question is pretty straight forward, just as long as your photo has GPS data in it. The bird question is a lot more difficult and involves something called deep convolutional neural nets.
Using these neural networks, Flickr’s Vision team says they can identify 1000+ plus things in photos. And as they ‘train’ their system with more images, the system will only get better and better.
We won’t get into the specifics, but you can by clicking here and reading the full writeup on deep convolutional neural nets over on Flickr’s code blog. Or, if you’re more interested in seeing the technology in action, head over to the Park or Bird site and start testing Flickr’s object recognition for yourself.