Social-Media
The new Facebook AI tool detects the products for sale automatically
Facebook is introducing a 'universal product recognition model' which uses artificial intelligence to recognize consumer products, ranging from furniture to fast-mode vehicles.
It is the first step in the future to identify and potentially shopping for the products in every single image on your site. Manohar Paluri, head of the Applied Computers Vision on Facebook told The Verge that "we 're going to make anything and everything on the site shoppable, whenever the experience is right. "The vision is great."
The business says that product recognition is the first in a series of new AI updates to come soon to its e-commerce platforms. They ultimately incorporate IT, increased reality and even digital assistants to create what it calls a "virtual first." It also launched today a feature called Shops which enables small businesses to set up Facebook and Instagram free stores.
Fashion would be a key element in this, as a potential Facebook "AI fashion stylist" might provide tailored shopping tips based on their wardrobes and regular recommendations for weather and schedule-specific outfits.
Tamara Berg, research scientist for Facebook, told Verge that "this is anything I wanted to build since I watched Clueless. "In 1995, they really imagined everything, but now the technology I believe is ready to bring it to life."
However, these features are not necessarily new ones, as the Clueless reference implies. They have been tried and tested for years even outside Hollywood, but with mixed results.
With the now little heard-of Echo Look, Amazon already has his own AI-powered fashion assistant. And it has been a reality since at least the Amazon Fire Phone using machine vision to identify and store products. In the meantime, Amazon is one of a number of companies that have started its own "Shazam clothes" using machine training. Online shopping platforms like eBay use AI already for speeding up the process of listing items to sell.
Facebook claims that the reach and accuracy of its tools are distinctive. GrokNet can recognise hundreds and thousands of different attributes in a image using the company's new product recognition tool. This include other labels, such as color and scale.
GrokNet has been deployed on the Facebook marketplace, where it allows users to list items for sale easily, recognize their products and create short descriptions. For example, you can upload the picture of your couch, which is quoted by the marketplace as "black, leather, sectional couch."
A version of this tool designed for companies is also being tested. The AI program automatically labels and connects to shopping sites when it uploads images to its homepage of its own items.
Facebook is helped to build these tools through access to marketplace user photos. GrokNet is educated on a gigantic magnitude database of about 100 million pictures, most of which are taken from the market. Facebook claims that this data is essential in building a vision system for a computer that is able to distinguish items from complicated lighting and dodgy angles — a part of the shopping experience that doesn't go anywhere.
Nevertheless, the exact nature of GrokNet is uncertain. This notes that 90% of the consumer photos in the Home and Garden category can be defined, but that it does not have comparable figures for other types of product categories.
As is always the case for resources like this, we have to wait and see what the Groknet reaction from Facebook users will be between marketed features and the actual user experience.
A graphic of how Facebook’s product recognition tool works, generating metadata from an image. Image: Facebook
