Hello! Thank you for your excellent work on this research.
I have one question regarding the recall calculation: could you please elaborate on the specific implementation details of the recall metric? In particular, since each image is associated with multiple queries in your setup, I am curious about how you aggregated or computed the recall score across these queries.