An improved image model meanwhile aids the captioning system's powers of description, so that it not only identifies a dog, grass and frisbee in an image, but describes the color of grass and more ...