A research team at UT has developed an artificial intelligence program to improve people’s clothing choices.
Fashion++ is a program with the goal of making minimal edits for outfit improvement, such as changing the color or fit of a piece of clothing, said Kimberly Hsiao, a computer science graduate student. She said she is leading the project with UT computer science professor Kristen Grauman and students and professors from Cornell Tech, Georgia Tech and Facebook AI Research.
“We wanted to come up with something that is useful in peoples’ lives, and clothing is how people make statements about themselves,” Hsiao said.
Hsiao said the program works by users uploading a photo of their outfit. The AI creates a series of suggestions to improve fashionability. She said the program was presented at the International Conference on Computer Vision in Seoul, South Korea on Oct. 31.
The code for Fashion++ is posted on the project’s page for anyone to use. To teach the AI what was in style, Hsiao said the research team fed Fashion++ 10,000 photos from Chictopia, a fashion website where bloggers can post pictures.
Hsiao said the team plans on continuously updating Fashion++ with new images and clothing for different body shapes.
“We do recommendations not only based on the user’s style preference, but also the user’s body by incorporating the user’s estimated model in the system,” Hsiao said.
Isay Katsman is a math and computer science senior at Cornell Tech working on the project. He said machine learning relies on massive data sets to perform well, but Fashion++ is currently limited by its smaller data set.
“The bigger question for me is how we can productionize research to the point where people can incorporate the fruits of our effort into their everyday lives,” Katsman said.
Danny Lopez, communication studies and human relations junior, said Fashion++ can help people who are tired of their style.
“There’s definitely a lot of people who care (about fashion) and I can see this being very helpful for people who are indecisive or have anxiety about what they should wear,” Lopez said.
Katsman said Fashion++ is just one way he sees artificial intelligence helping people guide their decisions.
“I see the entire Fashion++ framework being a stepping stone more generally for useful generative models that help us answer unintuitive questions,” Katsman said.