University student creates personal shopper technology
February 15, 2016
“They say pictures are a thousand words, and that’s exactly true for fashion,” said Bernard Ghanem, a graduate student at the University and the Beckman Institute,JT who co-developed the technology that allows users to search for an item of clothing through a catalog of pictures.
The technology, called Fashion Latte, was originally co-founded in 2008 by Ghanem and two of his colleagues who were in the same research lab with him, Sanketh Shetty and Esther Resendiz.JT
The team participated in the Illinois Ventures (iV10) program during their doctoral program, a competition where they were given ten weeks to put together a prototype for the idea they had proposed, which they then won.JT
“The good people at iV10, they gave us a place to develop our program and our idea, and they had a meeting with people who could help us put together a business plan,” Ghanem said. “Because we’re academics more or less, we didn’t really know anything about business development.”
Their idea was a way for users to use pictures to search for items of clothing.
“Given an item of clothing, finding things that look similar to it and finding attributes to that thing without those attributes being available in text,” Ghanem said.
The team made a website where they took pictures and descriptions from major store catalogs, such as Macy’s, and created a database focused on women’s clothing.
“We focused on women’s wear because the market analysis told us that, at that time, women buy ten time more than men online,” Ghanem said. “We enabled the user to explore all these catalogs together in a visual way.”
The technology has other features such as showing options of an item of clothing in a different neckline, similar items that are less expensive or of a different clothing brand.JT
“There’s filtering that you can change, not only on prices and sizes available but also on the visual content,” Ghanem said. “Things that don’t exist in the text that is provided by the online catalog.”
Fashion Latte also had an option that would help users pair certain items together.JT
“We also had a mix and match option where we would try to recommend to the user shoes and handbags that would go with the dress, or the other way around.”
After graduating from the University, Ghanem submitted a project proposal to the Advanced Digital Sciences Center (ADSC) in Singapore – a collaboration between the University and the Agency for Science Technology and Research (A*STAR).JT In 2010, Fashion Latte licensed out the technology to ADSC, which was the building block for the new version of Fashion Latte called FashionMatch. It was also where Ghanem met Vassilios Vonikakis, who headed up the ADSC side of FashionMatch.JT
Ghanem said FashionMatch was able to take it one step further and allow the user to use any picture to search.
“They did some really cool stuff, they showed me a phone application demo, and they improved the accuracy and the speed of the methods,” he said.
“When Bernard came here to Singapore to work in ADSC, he transferred the rights to ADSC in order to continue this project that he and the other two guys at the University had started,” Vonikakis said. “I started with Fashion Latte and started building on top of it.”
They then received a commercialization grant from the government of Singapore that allowed them to commercialize the product.
“The role of my team was to take what was existing there in Fashion Latte and devote into something that would be visible in real-life conditions,” Vonikakis said. “Fashion Latte evolved into FashionMatch, and the specifications changed to images that were not catalog images anymore so that means they were not perfect.”
He said FashionMatch evolved so that it works on images where the clothing items aren’t perfectly posed, as they typically are in store catalogs.
“Whenever you see someone out there take a picture, the system should work in these kinds of cases, the algorithm should work for clouded backgrounds,” he said.
Aishat Tepede, junior in AHS,JT said she has never used a technology like this before but that she could see herself using it in the future.
“I don’t think that it would be a bad idea,” she said. “I haven’t used it but I would, it makes it easier for me.”
The images currently take three to four seconds to process and are compared to 160,000 images in the database. The technology is also able to diminish the effects of shadows on images that are uploaded by a user, allowing newly uploaded images to be a part of the database.
FashionMatch has since licensed the technology out to five companies in Asia, including China, Korea and Singapore and is hoping to expand to America as well.