cap F1 dataset and prompts

Hi,

I am trying to reproduce the cap F1 procedure described in the first paragraph of section C in the Molmo and Pixmo paper.

Is it possible to release the dataset used for that evaluation (the 1500 image and their transcripts at least), as well as the prompts used with GPT-4o to compute both precision and recall (enumerating atomic statements, matching them, and checking consistency) ?

Thank you!