While the folks at Flickr would seem to have their work cut out for them — having just released a brand new iPad app chief among them — that doesn’t mean there’s no time for fun. Fun, for the Flickr team, appears to be rolling some image recognition technologies into a new Web app.
Aptly named Flickr Park or Bird, and inspired by the Web comic xkcd, the app simply tells you whether the image you shot was taken within the geographical confines of the US National Park System and whether that photo pictures a winged creature.
The easy part is the GPS location information already embedded in your photo, if the camera you’re using can record geolocation — smartphones and even some newer DSLRs can do this. The hard part is the vision technology that “sees” and interprets the content of the photo.
“This technology is working behind the scenes to improve Flickr search,” Hess told TNW. “Every photo that is uploaded to Flickr is run through the same image recognition technology that went into Park or Bird. This technology has the power to suggest more than 1,000 potential tags and we’re working to apply it more widely on Flickr to help people better organize and find photos.”
The vision team has been working for the last year on image recognition for more than 1,000 objects within images using convolutional neural nets technology and experimenting with scaled computer vision and deep learning algorithms using its image collection to expand search capabilities beyond metadata.
“We are using a technique called deep neural networks, often referred to as deep learning. We are still learning how to best balance human annotations with computer knowledge — we want to give Flickr users reliable, powerful search and discovery features,” Hess said.
In addition to birds, the technology also recognizes dogs, cats, sunsets, beaches, cars and airplanes. The technology’s effect can be viewed, for example, when searching Flickr for “ladybug” or “door.”
This computer vision technology will be applied across Flickr, on the web and on mobile and it is currently embedded in Flickr for iPad. “Our new search functionality lets you instantly explore your photos and videos across the site, without the requirement of tagging or adding metadata,” Hess said.