[Question]

[GarryJAY502]

❓ Question

Hello！
”AP_IoU_0.10_MaxDet_100: is the main metric used for the evaluation in our paper. It is evaluated at an IoU threshold of 0.1 and 100 predictions per image. Note that this is a hard limit and if images contain much more instances this leads to wrong results.“

What does "100 predictions per image" mean, and why is it done?
Thanks in advance!  

[andres2631996]

Hello, 100 means that the metric is just computed on the 100 top scoring boxes for each inferred case. The different combinations of anchor sizes can lead to hundreds of thousands of boxes per case, most of them with a very low score. Hence, operating on such a high number of boxes may not be feasible.

Additionally, each case may be predicted with a different number of boxes, so it makes sense to consider the same number of boxes for each case, so that all cases contribute in the same way to the metric