TLDR #2 - If you’re interested in seeing what kind of tags are being generated, you can see a version of the site that displays them at http://allparel.tech
Hi everyone! My name is Sarah and I'm a senior studying computer science at MIT. Over the past 9 months I've been working on Allparel (http://allparel.io). Allparel uses tags that are auto-generated from clothing product images in combination with product descriptions to match search and filter queries.
I had the idea of using computer vision to improve text search a couple years ago while shopping on Urban Outfitters. Every time I searched something, almost everything (even products that should have been returned) would get filtered out. Because search didn't work, I instead had to browse pages upon pages of products to find what I wanted. I noticed that the reason search wasn't correctly finding matching products was because the product titles and descriptions were often missing information. Text search can only be as good as the textual information provided about a product, so I decided to train classifiers that could automatically generate search tags for a product given just its image.
Allparel is live right now, and lists hundreds of thousands of products from retailers like Nordstroms, Macy's, Maurices, and more! Mens' products, as well as additional retailers and product tags, will also be coming soon. In the meantime, let me know your thoughts and feedback!