Computer scientists at Carnegie Mellon University are building a database of human torsos in order to teach computers about the topography of the human body, Atlas Obscura reports. This week, project creators Adam Milner and Ben Snell teamed up with a group of volunteers to organize and classify about 10,000 shirtless selfies.
The so-called Torso Computer Club has a few goals: Milner and Snell hope they can use their selfie collection to teach computers to differentiate between belly buttons and nipples—a task that may sound simple, but is still difficult for computers to achieve. The two are also interested in teaching computers to identify a range of physical characteristics (freckles and hair, for instance) and even poses.
Milner originally began collecting torso selfies on Grindr and other dating apps. By teaching computers to classify elements of the torso photos, they believe they can not only teach computers more about human physiology, but learn a little more about human behavior, themselves. Already, they’ve used a computer to identify a range of selfie archetypes, like profile shots and mirror selfies, based on their torso selfie database.
Part of the goal of the Torso Computer Club is simply to provide a dataset for other researchers. As Snell notes, “There aren’t open source data sets of torsos just lying around the internet.” However, both Snell and Milner hope to use their torso photo collection to pursue research of their own. Snell tells Atlas Obscura, “We want to continue mining this collection for the potential it has, though less as a tool and more as a form of provocation and speculation.”
Like #machinelearning? #torsos? #dumplings? Help label 1000s of nipples & navels this Tuesday 9/20 @creativeinquiry! pic.twitter.com/HGF26dV0qF
— Golan Levin (@golan) September 17, 2016
[h/t Atlas Obscura]
Know of something you think we should cover? Email us at tips@mentalfloss.com.
September 22, 2016 – 7:00am