With over a trillion photos created every year now, one feature that could help people make sense of their massive photo collections could be object recognition and automatic tagging. If your camera and photo management software can figure out what’s in your shots, it’ll make searching through old photos much more easy and intuitive.
Companies and researchers are working hard on pushing this field forward. Photo sharing services are already adding auto-tagging to their systems — Flickr and Google had to work out some early “racist” bugs — and now we’re getting a glimpse of what the technology could look like live, in cameras.
A company called TeraDeep is working on a product called the Learning Camera. By augmenting a digital camera with deep learning, the company gives the camera the ability to recognize people, pets, faces, and all kinds of objects that are found in scenes. In addition to using pre-trained detectors, the camera can be trained to recognize new objects as well.
Here’s a 2-minute demonstration of a TeraDeep camera being taken around a house. The words in the upper-left-hand corner show what the camera is detecting in the shot in real time:
The camera above was taught to recognize 1,000 different things using a training set of 10 million photos.
It’s the “first truly smart camera,” TeraDeep says. If you’d like to try out the technology for yourself, you can find the code for a demo app over on GitHub.