In the readme, there is this graph:
My question is: what dataset of prompts was used to generate this? Is it COCO or something else?