Remove prediction suppression

[czaloom]

Issue

Average Precision and Average Recall are currently not computed for labels that do not appear in the ground truth set. This is potentially problematic as a model could be highly confident in a label that is present in the real-world but not in the evaluation set leading to the suppression of certain labels.

Severity

LOW - This is likely not going to affect any existing evaluations as label sets are generally well-represented in training and validation sets. However, this could greatly affect one-off evaluations over smaller sets of data.

Solution

Remove suppression and update AP and AR derivative metrics.

• AP, mAP, APAveragedOverIOUS, mAPAveragedOverIOUs
• AR, mAR, ARAveragedOverScores, mARAveragedOverScores