Json captioning described in the paper.

[SRagy]

In the paper, a captioning procedure is described, including 1) short description, 2) detailed description, 3) background etc... It is mentioned that this is given in a structured way in json format. How precisely is this done for if we want to replicate it for our own prompting?

1. Do we do, e.g. {"short description": "A cat in some grass", "detailed description": A short-haired Bengal cat ..., "background": ... etc? 2. Do we include the quotation marks?
2. Are the category names actually given in the json as in the list (short description, etc)?
3. Were all 7 items in the list given for every caption or were they mixed and matched?

Many thanks for any response you are able to give. The model works wonderfully even with ordinary language captioning, but I would like to experiment with the structured captioning you mention.