Microsoft's latest iPhone app can identify your dog's breed based on a photo

Advertisement

Microsoft introduced a new app on Thursday that anyone with a dog should play with because it's a lot of fun.

Advertisement

It's called Fetch!, and it's available for iPhones and on the web. It uses artificial intelligence techniques to classify images of real-world dogs into breeds. On the web, users can upload a photo of a dog, or you can take a picture of your pet using your phone's camera.

If you upload a picture of, say, a Rhodesian ridgeback, Microsoft should be able to confirm the dog's breed.

Complimentary Tech Event
Transform talent with learning that works
Capability development is critical for businesses who want to push the envelope of innovation.Discover how business leaders are strategizing around building talent capabilities and empowering employee transformation.Know More

It's the latest in a line of fun, silly apps released by Microsoft Gargage, an "outlet for experimental projects" that are designed to show off creative and unexpected ways to apply Microsoft's expertise in artificial intelligence. In the past year, Microsoft has released apps that detect and measures mustaches in photos or guesses your age, for example.

Like Microsoft's other AI apps, Fetch! should become more accurate as users upload more photos and data. More technical information is available here.

Advertisement

Fetch! is already fairly accurate. Here it identified the breed of a dog belonging to one of BI's reporters.

havanese

Kif Leswing

But it thinks my cat is a dog with a "99% match." Incorrect answer, Fetch!

cat is not a dog

Kif Leswing

And the app has a sense of humor. When it's fed an image that's clearly not a dog, such as a flower, it will try to identify it as well as possible. If you upload a photo of a person, it'll make a joke:

Advertisement

IMG_3297

Kif Leswing

Dog classification is a classic challenge for AI researchers. Well-organized datasets for dog breeds are free and widely available. The prestigious ImageNet competition today requires entrants from places like Stanford and Google to identify many different types of images, but back in 2011, it only had one category: dog classification.