February 13, 2017 weblog
Age is just a number but Amazon AI is after some smart estimates
(Tech Xplore)—Tech sites are talking about an Amazon tool to estimate a person's age powered by the giant's Rekognition platform.
Amazon can justify bragging rights in image recognition with its service called Rekognition. It makes use of the company's deep learning expertise. Amazon Rekognition was designed to work seamlessly with popular AWS services.
Nick Statt in The Verge noted how Rekognition is offered "to developers who are interested in implementing general object recognition, labeling, and other likeminded features for their products and services."
The Amazon Rekognition site page said, "With Amazon Rekognition, your applications can confirm user identities by comparing their live image with a reference image."
With its API you are enabled to add "deep learning-based visual search and image classification" to applications.
Deep neural network models are put to work. Now the service is going up a notch and a number of tech watchers highlighted the news about estimates of ages.
The latest development is under this umbrella of the Rekognition AI service. John Mannes in TechCrunch: "The team behind Amazon's Rekognition API dropped a nifty feature... that lets anyone upload their photos to get the company's best guess of their age."
Lucy Black on Saturday wrote about the neural network trained by Amazon. Writing in I Programmer:
"You simply give it a photo and it returns an analysis of what it is looking at. The data it can return varies from a bounding box to emotion, gender, eyes open, etc. Now it also includes an estimated age range for a person..."
Jeff Barr on the AWS blog zeroed in on the age-telling capability on Feb. 10. The blog inspired talk elsewhere about the tell-the-person's-age attention-getter. "Amazon Rekognition returns an array of attributes for each face that it locates in an image. Today we are adding a new attribute, an estimated age range. This value is expressed in years, and is returned as a pair of integers. The age ranges can overlap; the face of a 5 year old might have an estimated range of 4 to 6 but the face of a 6 year old might have an estimated range of 4 to 8."
"Looks like a face." "Appears to be male." "Age range." "Smiling." "Appears to be happy."
Barr said the feature was available now.
Mannes tried it out. How reliable is it in taking a stab at someone's age? "First up was Amazon—the web app pegged my friend at somewhere between 26 and 43. This was slightly disappointing, given that he is considerably younger than 26. Though props for the 99.9 percent, almost certainty, on gender and the addition of 'appears' next to happy. I'd have to fake the smile too if someone told me I looked 43."
Nonetheless, Mannes said, "they still did better than Microsoft at this game. Microsoft's How-Old.net tool guessed my friend to be 30. Ouch."
The tool has earned more than laughs, even if it does not always make the best call. Significance? Rob Williams in HotHardware said that "this is only what Amazon's service is capable of right now. We're still early in this deep-learning game, so there's no doubt that these guesses will improve over time, as Amazon continues to churn through its never-ending supply of data, feeding the machine."
Nik Statt in The Verge: "Microsoft tried the latter approach back in 2015 with its own AI tool, resulting in some hilariously bad estimates that exposed fundamental weaknesses in how these types of image recognition algorithms function. Still, these experiments are more for fun, and both companies' cracks at age-guessing algorithms are a good way to mess around with AI if you're so inclined."
© 2017 Tech Xplore