Computer learns to give fashion advice via AI

The Fashion ++ program recognises colours and patterns, allowing it to suggest minor changes to improve outfits.

Spotted: US-based Facebook AI Research has created a computer system that recognises fashion faux pas. The program shows the potential of deep image learning for the fashion industry, according to its creators. 

The system overcomes several programing challenges, according to the researchers. For instance, they had to teach the computer to recognise fashion rules, regardless of changing styles. 

To create Fashion ++, researchers fed the computer over 15,000 fashion images. The computer then used an algorithm to code the shape, colour and pattern of pieces of clothing, and compared variations of the combinations. Based on its calculations, it proposes small changes to improve outfits. 

Some tweaks are minor, like changing the colour of a shirt. Other suggestions are more substantial, for instance swapping out a skirt for trousers. 

AI researchers from Cornell, Georgia Tech and Facebook AI Research worked on Fashion ++. Their research was recently published and the team plans to continue working on the system.