AI-Driven Image Captioning For Inclusive Productivity


July 17, 2019


Ed Cutrell, Danna Gurari, Kathleen McCoy, Walter Lasecki, Ece Kamar


Microsoft Research, University of Texas, University of Delaware, University of Michigan, Microsoft Research


Advances in hybrid intelligence, deep learning, and related artificial intelligence techniques have provided us with a remarkable opportunity to ensure the future of work will be even more inclusive to more people than ever before. Because the communication and products of work increasingly comprise images—photos, charts, maps, and the like—that are often not accessible, people who are blind or low vision face unique challenges. One promising technology is the automated understanding and captioning of images. Office 365 applications, for example, can use APIs from Microsoft Azure Cognitive Services to automatically add alt text to images. But there remain many hurdles to making these captions truly useful and usable. In this breakout session, we will explore the state of the art and potential for advancement in automated image captioning, including data capture and curation for training, caption presentation and interactivity, and computer vision.