Under the hood: Building accessibility tools for the visually impaired on Facebook

What this post added

Introduced automatic alternative (alt) text for photos on Facebook for iOS, providing text descriptions for visually impaired and blind users using object recognition technology. This feature leverages a deep convolutional neural network within Facebook's computer vision platform to detect a curated set of ~100 concepts. The system constructs sentences by categorizing detected elements (people, objects, scenes) and prefacing the description with 'Image may contain' to manage uncertainty. The goal is to provide scalable, algorithmically generated descriptions without user intervention.