Computer, does this shirt go with these pants?

Keeping up with the latest fashion trends and looking stylish can be difficult for even the most fashionable person. To stay in the know, you might consult a friend, social media or a publication.

But what if knowing how fashionable your outfit is was as easy as taking a picture?

That’s the idea behind an artificial intelligence system built by a computer science research team at the University of Texas in partnership with Cornell Tech, Georgia Tech and Facebook AI Research.

The tool, called Fashion++, uses visual recognition to analyze an outfit and suggest improvements on the spot.

Kimberly Hsiao, a graduate student studying computer science, came up with the idea for the system.

“Personally, I enjoy following fashion topics and as a computer science major I wanted to come up with something that can be useful and people like,” Hsiao said. “We wear clothing every day but before we go out, we don’t know if we look good or not, and it would be really useful if an AI system could give us an opinion on it.”

The AI tool can analyze various elements of a photo, including color, pattern and shape of the garments. Then it looks to where small adjustments to the outfit will improve it most and offers several options.

The system focuses on small tweaks to an outfit that could make a big impact from swapping in a new garment to tweaking its color, how it is worn, like rolling up sleeves, or its fit, such as making pants baggier.

Programmers taught the system about fashion trends by feeding it 10,000 pictures of fashion-centric outfits from influencers and fashion enthusiasts that were shared online.

Kristen Grauman, a UT professor of computer science and a Facebook AI researcher, said that while fashion can be subjective, the system is learning from data drawn from images, not opinions.

“We are not defining it. Neither is the system,” Grauman said. “Instead, it’s just emerging from data,”

For counter images, researchers had to get creative, making their own by taking pieces of outfits that drastically do not go together as the “unfashionable” outfits. This gave the program the data it needed to sort between the categories, researchers say.

“People don't actually share photos of themselves when they don’t look fashionable,” Hsiao said.

The resulting system lets a user submit a picture of an outfit, and the AI tool will send back several options of minor adjustments.

“So you can see like, ‘Oh I should tuck my shirt in,’” Hsiao said. “Or ‘I should roll my sleeves up or I should change from skirt to pants or to pants with another color.’”

Hsaio said users could also provide the AI with images from clothing from your closet or a particular merchandise catalog that would allow the suggestions to be more personalized based on your available options.

The system is also currently better at female-focused fashion because most fashion influencers in the photos used are women, the researchers say, but the algorithm can work for any outfit. Adding new trends or more diversity can be achieved by feeding it new pictures, they said.

“A year from now or two months from now if you want to update a model it’s just a matter of updating with new images of what people are wearing,” Grauman said.

Technology from the Fashion++ system is still a few steps from giving people the ability to pull out their phones, snap a pic and instantly knowing how trendy an outfit is.

To run the code people need a machine with AI framework supported, so it isn’t able to run on a smartphone or app yet, but the team is open sourcing the project to Github to let others build on or use the existing code.

Hsiao said she has no plans to turn it into an app, but does hope someone picks it up to make it one.

Grauman said AI technology has the potential to help out in other creative domains like cooking or music, where it’s not as “cut and dry” as traditional computer vision problems.

“We’re not trying to replace what a person can do, but we’re trying to augment or assist through methods that can scour lots of data and learn better and make suggestions ” Grauman said.

Want to hear Austin’s top news every weekday morning? Our Headline News Express podcast can be delivered to your home smart speaker or mobile device each weekday. Subscribe through Apple or Google Podcasts, Stitcher or Spotify (or wherever you get your pods) or enable your Amazon Echo or Google Home smart speaker for news from the Austin American-Statesman.

Get our podcast for Austin’s latest headlines every morning