Fashion is a strange, fickle beast. It changes with the seasons, following new styles and the hottest celebrities' latest looks. What's cool now could be dorky next week. But it may soon get much easier to keep up, provided you're willing to listen to the advice of an unfeeling machine – specifically, a computer algorithm that crunches the numbers to give you an assessment as to how fashionable your outfit is and how it might be improved.
The algorithm was produced by researchers from Spain's Institute of Robotics and Industrial Informatics, along with colleagues at the University of Toronto. It's built on a detailed analysis of a dataset containing 144,000 user posts from fashion website chictopia.com.
UPGRADE TO NEW ATLAS PLUS
More than 1,500 New Atlas Plus subscribers directly support our journalism, and get access to our premium ad-free site and email newsletter. Join them for just US$19 a year.UPGRADE
To turn the data into a well-defined mathematical model for trendiness, the researchers used a deep neural network coupled with a conditional random field, which helps the computer learn how random elements such as geographic location, background scenery, and ethnicity do or do not predict how fashionable a post is. The learning process considered not only what was visible in the photos but also the number of "likes" received, the tags, and any accompanying comments or descriptions.
From the many levels of analysis, correlations, and evaluation, the algorithm learned how fashion varies by season and city and even age group. Armed with this knowledge, it can now predict with reasonable accuracy whether a given outfit would be deemed fashionable, and offer suggestions for how to dress more fashionably.
The research leaves three big questions unanswered, however: firstly, how quickly could a system such as this adapt to new trends? Secondly, how would it cope with a wider range of body sizes and shapes than appear in a dataset drawn from a fashion-focused photography social network – which would have a tendency, through self-selection, to be dominated by thin young women?
And finally, most crucially, would anyone even consider fashion advice from a machine? Clothing is such a personal, individual thing that, even accounting for the desire to be trendy, it's hard to believe something like this ever catching on. Fashionistas are too far ahead of the game to benefit, style-conscious people will likely take the word of friends, magazines, and celebrities over that of an algorithm (however accurate it may be), and the rest of us probably don't care enough to bother.
That said, we're curious to see how much further the research could go. Just imagine a world where new trends are dictated not by the whims of a few but by the mathematical reasoning of a machine that studies the many. Would that be liberating? Or suffocating?
A paper describing the research was presented at the 2015 Conference on Computer Vision and Pattern Recognition in June.