“VQA models need to be able to read and reason about text in an image to answer questions about them. But even today’s state-of-the-art models struggle to do this. To advance VQA and research, we're hosting a challenge: https://t.co/QXQX0jifcO. Enter by 5/18!”