Cornell researchers and Bloomsbury Publishing are partnering to assess the benefits of using artificial intelligence to enhance and improve the speed and quality of manual indexing for reliable fashion trend forecasting.

The aim of this project is to cultivate research connections between the computer vision and fashion communities, and thereby advance the state-of-the-art in fine-grained visual recognition for fashion and apparel.

Doctoral students Mengyun Shi and Menglin Jia from Human Ecology’s Department of Fiber Science & Apparel Design, will draw on images and metadata from the Bloomsbury Fashion Photography Archive to explore smart and rapid archiving and classification of fashion images through the application of machine learning and artificial intelligence.

Credit: Niall McInerney/Bloomsbury Publishing

“Success of this project will open a door to highly reliable trend forecasting, and help the fashion industry respond to changes in consumers’ need and fashion taste quickly,” said Huiju Park, associate professor of fiber science and apparel design, who is co-advising the project with computer science faculty members Serge Belongie and Kavita Bala.

The project is expected to be able to identify specific features of garments and styles and is likely to have applications well beyond the Fashion Photography Archive.

“The attribute dataset that this project will create will open new opportunities to recognize fine details in apparel, and will drive new problems and solutions in recognition of fashion and apparel,” Bala said.

Bloomsbury, an award-winning global independent publisher of fiction, nonfiction, children’s, specialist trade and academic publishing, acquired the archive of more than 750,000 images in 2011 and has been manually indexing each image by garment, color, person and theme for Bloomsbury Fashion Central, a suite of digital resources for the academic educational community.

Kathryn Earle, managing director of Bloomsbury Digital Resources, said, “We are tremendously excited to be in a position to supply a data sample that is sizable enough to assess whether machine learning can improve upon manual indexing.”

Subject indexingis the act of describing or classifying a document by index terms or other symbols in order to indicate what the document is about, to summarize its content or to increase its findability.Indexing completed by humans, is prone to error, subjectivity and knowledge limitations – something AI could prevent.

“We are anxiously awaiting the outcome of Cornell’s research,” Earle continued. “The Fashion Photography Archive contains priceless images that Bloomsbury has preserved for posterity, and we want to do everything we can to support them online. If AI can improve discovery for researchers and students, we will be thrilled.”

Stephen D’Angelo is assistant director of communications at the College of Human Ecology.