From Pixels to Prose: A Large Dataset of Dense Image Captions