Artificial Intelligence: The Future of Fashion

By Jürgen Kalwa   |   February 19, 2020   |   Provided By Avnet Silica

Designers and manufacturers of men’s and women’s apparel are exploring ways to predict trends faster and more accurately than ever before. They are helped by researchers who are looking for ways to utilize artificial intelligence. Smart Industry went to Cornell University in New York to talk to two bright young scholars who are teaching computers everything about garments, fabrics, colors, and patterns and how style is manufactured.

A few years ago, when Mengyun Shi entered the fashion industry and moved from China to take various posts in well-known arbiters of taste like Dolce & Gabbana and Giorgio Armani in Italy, he learnt a valuable lesson right away. While it was intriguing to give things a personal touch, he struggled mightily with some of the basics. It took him 30 minutes, for instance, to flush out a decent sketch of one of his product ideas but in order to develop things worth manufacturing he had to come up with many versions and variations.
He thought that technology, at least some time down the road, might be able to shorten the elaborate process of creating mass-market products with flair to successfully target fickle, trend-conscious consumers. In order to achieve such a lofty goal, he decided to leave Italy and the industry to enter the realm of American academia to make the right connections.

 “AI can reduce the design process to just a few minutes. The value of design work will be diminished.” Mengyun Shi, Cornell University

He moved to Ithaca, a small town on the banks of Cayuga Lake in upstate New York and the home of Cornell, one of the lesser known of the prestigious Ivy League universities. In 2014, he began his master’s studies there and, in 2016, entered a PhD program in the Department of Fiber Science and Apparel Design, a division of the university’s College of Human Ecology.
His ultimate goal sounds straightforward, even if it will take time to put all the puzzle pieces together. Mengyun wants to apply artificial intelligence to fashion forecasting and help the industry and its experts become more efficient. “It’s scary,” he admits. “The AI model can reduce the process of creating design sketches to just a few minutes and produce hundreds of sketches in a short time period. The value of design work will be diminished.” This mirrors what has happened everywhere when computers, with their enormous power to handle tons of data and trillions of permutations, start getting going. Even when they first need to learn all the difficult patterns of something as sensitive and personal as fashion items, something style-conscious people wear as their second skin in order to express personality, temperament, and status.
“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,” says Huiju Park, associate professor of fiber science and apparel design. Park is an advisor on the project together with computer science faculty members Serge Belongie and Kavita Bala.
The project will in all likelihood open more than a door – it has the potential to bulldoze and flatten a whole mountain between the people with ideas and the hundreds of millions of fashion-conscious consumers. Right now it takes a year and a lot of capital, plus risk-taking acumen, to produce a fashion line to be runway-ready and set to be shipped in bulk to interested retail outlets. Applied AI could shorten that period considerably.
Currently, the industry relies heavily on forecasting techniques to minimize risks. The overview of the people in charge has become more sophisticated over time and there is no need for fortune tellers or astrologists to predict the future. Although the methodology is rational, common forecasting can look like a mixture of alchemy, psychology, and a con man’s gambit. Often, it only succeeds because of massive marketing efforts powerful enough to convince middlemen, media, celebrities, retailers, and the general public to believe in what is coming down the pike.
Any prognosis worth its money needs to anticipate many different things correctly, including colors, fabrics, textures, print patterns, graphics plus accessories and footwear. Since the arrival of social media, predictions are now often based on analyzing social media trends from sites such as Instagram or Pinterest. So much of this guessing game could eventually be reduced as soon as artificial intelligence gets to do its magic. Nobody expects computer programs to be able to push aside legendary versions of high caliber designers like Coco Chanel, Christian Dior, or Yves St Laurent and their successors. Their work, socalled haute couture, is considered as valuable as fine art and it caters for a pretty-rich clientele, with expensive taste, who wouldn’t want to be caught shopping at H&M – but it’s here that artificial intelligence might well be right on the money in the ready-to-wear and mass-market segments of the industry where cost is king.
“In our vision, Artificial Intelligence can facilitate people’s work and help the fashion insider, designers, and buyers, but it can also help customers,” says Menglin Jia, research partner of her countryman Mengyun Shi. “We want to create something that helps all sides to make efficient and well informed decisions.”
Jia was born in mainland China and studied in Hong Kong, where she joined German lingerie manufacturer Triumph in its headquarters for the Asian market.

"AI can help the fashion insider, designers, and buyers, but it can also help customers." Menglin Jia, Cornell University

Among other things, she designed a line of bras and panties featuring monkey motifs that were sold at the beginning of the Chinese Year of the Monkey (2016). After that she applied to Cornell to study for her master’s degree and decided to add a doctorate to dig deeper by combining her interest in fashion and Artificial Intelligence. Recently she added an internship in the Facebook AI department to her résumé. The research is still in an early phase. Machines will need to learn many things they have not mastered yet. “The goal is to advance fine-grained recognition in computer vision,” says Jia. The software needs to understand: “That’s a blue-striped shirt. That’s a button-down. What kind of fabric that is, what kind of color that is,” she explained. The honest answer to the question of how close they are to accomplishing that: “We are not there yet.”
But the two scholars are pushing hard to get there. One of the little steps needed happened in 2018, when Cornell announced a partnership with Bloomsbury Publishing. This came with the opportunity to draw on a large archive of images and metadata from the company’s fashion photography archive. The two PhD candidates used this to build on something they call a “Fashionpedia,” a methodology to annotate images with a tree-like classification criterion using finegrained attributes in particular.

^{Intelligent Design Nestled in the picturesque Finger Lakes District in upstate New York, Cornell University’s Department of Fiber Science and Apparel Design, a division of the university’s College of Human Ecology, is leading research into the transformation of haute couture through Artificial Intelligence (AI).}

One important element is to train and benchmark the next generation of computer-based models for the comprehensive understanding of fashion. This will be helped through the latest partnership with the American magazine publisher Hearst, which puts out flagship publications such as Harper’s Bazaar, Equipe, Elle, and Marie-Claire. Cornell is not the only place where people are reshaping the fashion design process. Amazon, the largest online retailer in the world, is developing machine-learning systems that, according to a recent report in the online edition of MIT Technology Review, could “provide an edge when it comes to spotting, reacting to, and perhaps even shaping the latest fashion trends.” The work is innovative because computers usually require extensive labeling in order to learn from visual information. The same dynamic seems to be taking hold outside academia. The fashion industry clearly has not stopped thinking about how to integrate forward-looking concepts. One product of ongoing innovation efforts is the “hybrid design algorithm,” which is used to help customers to build a wardrobe collection based on actual garments, using guidance provided by the algorithm’s analysis.

^{Hybrid Design For upmarket stores such as Nordstrom, which are feeling the pinch of losing business to e-commerce retailers, algorithms that help them better understand their customers could prove to be crucial.}

Start-ups like the subscription service Stitch Fix, an online personal styling service founded in 2011, have come up with systems to mix and match wants and needs of its millions of customers. Users complete a style profile but are also assigned a personal stylist who will then send a box with a curated selection of clothes, accessories, and shoes – also referred to as a “fix” – that fit within a person’s taste and budget. Using each client’s constant additional feedback, the stylist, assisted by the algorithm, aims to develop a better understanding of the particular sensitivities of the customer in question.

"A jacket or a pair of pants that will adapt to your style." Tim Oates, University of Maryland

This is what companies like Trunk Club (owned by Nordstrom) or cosmetics specialist Birchbox (owned by Walgreens) are also trying to do. They have the potential to upscale the business of high-end American department stores in upmarket malls, such as Neiman Marcus or Nordstrom, or specialty retailers like J Crew, which are all feeling the pinch of losing business to e-commerce retailers. In Mountain View, California, the Google Brain team is working on finding ways to enable computers to analyze visuals and create data sets that can be applied to the style of clothing. Other teams are exploring ideas that could end up profiting the consumer. A group from the University of Illinois at Urbana-Champaign has developed an algorithm for identifying fashion-focused social network accounts. According to MIT Technology Review, Tim Oates, a professor at the University of Maryland in Baltimore, is working on a system that makes the transfer of styles from one garment to another feasible. He envisions algorithms that have been trained “on your closet, and then you could say here’s a jacket or a pair of pants, and I’d like to adapt it to my style”. Thanks to computers and their artificial understanding of your interest in being part of the in-crowd, the future of fashion could simply mean that your individual fashion sense might finally prevail.

READ MORE ARTICLES

About Author

Jürgen Kalwa

Jürgen Kalwa is a German journalist living and working in New York....

View Bio