Company Facebook today launched a “comprehensive model of recognition products,” which uses artificial intelligence (AI) to identify a wide range of consumer goods from furniture and clothes to expensive cars. This is the first step to the future, when the objects on each image on the website of the social network will be to first identify, and then buy.
“We want everything on our platform was available for purchase, when will want” – said Paluri Manohar (Manohar Paluri), head of the Department of applied computer vision in Facebook. The possibility of recognition of goods-the first innovation on the basis of the AI in a series of updates to e-Commerce platform Facebook in the near future. In the end, the social network will combine artificial intelligence, augmented reality, and even digital assistants that can be expressed by the term “social” shopping. Facebook and Instagram also launched today a platform called Shops that allows small businesses to create a free showcase for their stores Facebook and Instagram.
Facebook was not the first in the field of e-Commerce through the means of AI. For example, Amazon has already created his own aid on the basis of AI the field of clothing/fashion, though now it is little heard. And the use of machine vision to identify and purchase goods has become a reality yet, at least since the Amazon Fire Phone. However, online trading platforms such as eBay, are already using artificial intelligence to speed the listing of goods for sale, and Amazon is one of many firms which launched their own “Shazam for clothing” with the use of machine learning technology.
Company Facebook assures that its own tools are different from competitors in its coverage and accuracy. A new tool of recognition of goods, which is called GrokNet can identify tens of thousands of different attributes on the image: from the color and size of items to determine the specific brands.
AI GrokNet has already been deployed at Facebook Marketplace, where he helps users quickly identify items for sale, and generates a short description. For example, you can upload to social network a photo of his sofa, and the online platform will offer to call it for sale as “black, leather, sectional sofa”.
In the development of these tools of Facebook, of course, helps direct access to the user photos in the Marketplace. GrokNet trained on a huge database of 100 million images. Facebook says that this data is vital to create a vision system that can identify products in photos with bad lighting and questionable angles.
It is not known how accurate is GrokNet. The company assures that the AI can identify 90 percent of the images in the Marketplace, in the category “House and garden”. However, Facebook does not disclose similar statistics for other categories products.