Artificial intelligence image recognition is already powerful, but it’s getting better week by week. Now, Wolfram Research has opened up a new website that will try to identify what’s depicted in any image you toss its way.
The new tool takes an image and attempts to work out what it shows. The algorithms used by Wolfram use machine learning to get better over time: they look at more and more images which are tagged as showing certain objects — a cat, a hat or a mat, say — to learn what they look like. Wolfram uses neural networks to sift through what it sees and narrow the possible choice of objects displayed in the image down to the most likely. Each layer of the neural network works at a different level of detail: first identifying brightness and colour of individual pixels, then how they come together to form edges and shadows, and latterly how they come together to signify features like wheels or eyes or petals.
“It won’t always get it right, but most of the time I think it does remarkably well, explains Stephen Wolfram in a blog post. “And to me what’s particularly fascinating is that when it does get something wrong, the mistakes it makes mostly seem remarkably human.”
The service can be used on desktop or mobile, and the code behind the site can be used too — making it possible to embed Wolfram’s advanced image analysis into new software and apps. The applications are clearly limitless. But let’s not mess around: what does it make of your photographs? Post the results in the comments. [Wolfram Image Identification Project via 9to5mac via The Verge]