Mean Average Precision (mAP) is a metric (not loss function) to evaluate the performance of object detection models across all classes and at various IoU thresholds. It provides a single performance measure that accounts for both the precision $\text{Precision} = {TP}/{(TP + FP)}$ (i.e., the (in)correct detection out of all detection) and recall $\text{Recall} = {TP}/{(TP + FN)}$ (i.e., the (un)missed detection) of the model by averaging across all classes. The data points on the curves are sampled by changing a score threshold $\text{Pr}(\text{Class} _i|\text{Object})\cdot\text{Pr}(\text{Object})\cdot\text{IoU}$.

• True Positives (TP) are the correctly identified detections, where the predicted class matches the ground truth, and the IoU (Intersection over Union) with a ground truth box is above a certain threshold.
• False Positives (FP) are the incorrectly identified detections, either because the predicted class is wrong or the IoU with any ground truth box is below the threshold.
• False Negatives (FN) are actual objects that the model failed to detect or detected with an IoU below the threshold.

Object detection entails recognizing the presence, class, and location of numerous objects in an image, complicating the relevance of "True Negatives (TN)." mAP primarily considers the other three elements of the confusion matrix, evaluating model performance based on incorrect detections (Precision) and missed detections (Recall), thereby focusing on the accuracy and reliability of detecting and classifying objects without incorporating TNs.