MaP: Clarification on empty predictions and class metrics

[Chris-hughes10]

Hi,

I have been using MeanAveragePrecision but am struggling to find a definitive answer on a couple of things and was wondering if someone here can help.

Specifically, I am looking for examples of how to handle:

• cases where the model makes no predictions, but ground truth labels exist
• cases where the model makes predictions, but no ground truth labels exist

Additionally, if I specify to calculate class metrics, will the background class be returned as part of the map_per_class tensor?

Thanks!