Towards VQA Models That Can Read

Amanpreet Singh, Vivek Natarajan, Meet Shah +5

Studies have shown that a dominant class of questions asked by visually impaired users on images of their surroundings involves reading text in the image. But today's VQA models ca…