Mean Average Precision (mAP)

Mean Average Precision is a critical metric used to evaluate the performance of models in Computer Vision and Information Retrieval. It is predominantly used as the primary benchmark for Object Detection systems, such as those evaluated on the PASCAL VOC and COCO Dataset. The metric provides a single-figure summary of the Precision-Recall Curve across multiple classes.

To compute Mean Average Precision, one must first determine the Average Precision (AP) for each category. In Object Detection, a prediction is classified as a true positive if its Intersection over Union (IoU) with a ground-truth bounding box exceeds a predefined threshold. The Average Precision is then calculated by integrating the Precision over the Recall range. The final Mean Average Precision is the arithmetic mean of these AP values across all classes in the dataset.

Variations of this metric exist depending on the competition or framework. For instance, the COCO Dataset evaluation calculates mAP by averaging over various IoU thresholds from 0.50 to 0.95, rewarding models that achieve high localization accuracy. Detailed mathematical definitions and implementations can be found on the Wikipedia page for Evaluation Measures and the official COCO evaluation site.

• Precision-Recall-Curve
• F1-Score
• Confusion-Matrix
• Deep-Learning