Facebook launched what it calls a “universal model of product recognition” that uses artificial intelligence to recognize consumer goods, from accessories to fashion brands to fancy cars. This is the primary step towards a future in which the products can be identified and potentially purchased for in every image on its site.
To pursue this vision, a new, universal computer vision system called GrokNet is being deployed by Facebook. This new model of product recognition was designed to make virtually any photo shoppable. GrokNet currently features in the marketplace, the peer-to-peer trading platform on Facebook, for buyers and sellers. When a seller uploads a product to Marketplace, the program automatically recommends listing attributes, such as color or material for the product.
GrokNet is primarily developed for shopping and is capable of identifying the characteristics — color, shape, pattern or style — of the objects it sees. Machine vision systems are usually only able to distinguish one type of item and a device designed to recognize different shoe types would fail to recognize similar SUV models. Unlike the other machine vision systems, GrokNet is robust enough to do that whether you’re showing it an image of the newest Yeezys or a sectional red leather sofa.
Facebook claims that scope and accuracy are what make that software special. The company is also developing a version of this method, which is designed for businesses. When they upload images to their product page, the AI program will tag them automatically and connect them to shopping pages.
However, it’s unclear just how accurate GrokNet is. The company claims that it can identify 90 percent of marketplace pictures in section Home and Garden, but it did not provide similar figures for other types of products.
This is always the case for platforms like this; there can be a big gap between the proposed functionality and real user experience, so we’ll have to wait to see what feedback GrokNet receives from users on Facebook.