Facebook’s Director of Applied Machine Learning Joaquin Quiñonero Candela announced that Facebook has developed an AI that can recognize untagged photos or photos without descriptions.
Candela said that Facebook has “built a search system that leverages image understanding to sort through this vast amount of information and surface the most relevant photos quickly and easily.” The Director of Applied Machine Learning at Facebook added that the new system allows Facebook users to search images based on the image content instead of depending on the surrounding text or tags. According to Candela, Facebook users can search and retrieve for instance a “black shirt photo” even if users have forgotten to tag or put description in photos with black shirts in their posts.
Facebook’s new system was developed using the company’s Lumos platform – described by Candela as a system that “keeps improving all the time, both through all the newly labeled data we feed it, and through the annotated data from the applications our teams build.” In developing Facebook’s image recognition AI, Candela further said that his team used pioneering deep learning techniques to sort out billions of photos and understand the meaning behind each photo.
Facebook’s image recognition AI, according to Candela, can be used to better describe photos for the visually impaired. In addition to providing better search results for posts with images, the newly developed image recognition AI can be used for videos and other immersive formats, the Director of Applied Machine Learning at Facebook added.