Text this: Integrating visual memory for image captioning