The company announced that Facebook’s Automatic Alternative Text (AAT) system can now detect and recognize 10 times more objects and concepts than when it was first launched. The description will also be more detailed, and AAT will be able to identify activities, landmarks, animal types, etc. Facebook announced improvements to AAT. The system was introduced by the company in 2016. The system uses object recognition as a visually impaired person to generate photo descriptions on Facebook and Instagram.

Facebook said in a blog that the improved AAT can recognize more than 1,200 concepts. This is more than 10 times the possible number of the original version launched in 2016.

Facebook said that AAT will also include information about the location and relative size of elements in the image. For example, AI does not need to say “may be an image of 5 people”, but can specify that there are two people in the center of the photo, and three people are scattered on the edge.

Although by default, the new AAT will provide a compact description of all photos, users can also choose to obtain detailed descriptions of photos that may be more interesting. When they choose this option, a panel will be displayed with a more comprehensive description of the content of the photo.

Facebook said that it consulted users who rely on screen readers and found that they wanted more information about images from friends and family members, and less information from non-friends and relatives. The detailed description will also include a comparison of location information and relative prominence of the object.

See also  Paypal opens encryption service to millions of qualified account holders in the U.S.

Facebook said that the updated AAT now represents a number of technological advancements that can improve the user’s photo experience. In addition, the concept that it cannot be reliably identified is omitted to reduce the error range.


Will WhatsApp’s new privacy policy end your privacy? We discussed this on the weekly technical podcast Orbital, you can subscribe via Apple Podcast, Google Podcast or RSS, download the episode, or click the play button below.