I am wondering if it is possible to evaluate or compute mAP (or also called AP) based on the COCO Detection challenge. My understanding is that it computes the average for all object classes and 10 IOU thresholds from 0.5 to 0.95, also known as AP@[.5:.05:.95].
The COCO API provides the mAP implementation for the COCO challenge. There are differences in how average precision (used in the mAP) is calculated, which is discussed in this article.
I am wondering if it is possible to evaluate or compute mAP (or also called AP) based on the COCO Detection challenge. My understanding is that it computes the average for all object classes and 10 IOU thresholds from 0.5 to 0.95, also known as AP@[.5:.05:.95].