The improvement of image-recognition software via deep learning could help usher in an era of image-based Web shopping/browsing, with Pinterest and Shoes.com testing such techniques.
Pinterest supports a new visual search tool in which the user draws a box around an item seen in an image to find visually similar items in a vast index. The system was trained to understand images by drawing on the text people attached to photos shared on Pinterest.
Shoes.com's approach also is powered by deep learning, but it differs from the approach taken by Pinterest. The footwear retailer is using image-processing software from Sentient, whose scheme involves users clicking on a "visual filter" button to summon a grid of 12 images that the software believes represents the most distinct clusters of styles from a catalog of approximately 7,000 boots. When a shopper chooses the item closest to their search request, the software will use the visual characteristics of the selection to refresh the grid to show 11 more items that resemble the chosen product.
"This is a category where it's very hard to describe in words to a search engine what you're looking for," says Sentient's Nigel Duffy. "We can get really granular preferences very quickly."
From Technology Review
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA