This repository provides a Python implementation of the mAP metric used in the PASCAL VOC competition to evaluate neural network performance on object recognition tasks. It calculates Average Precision per class and then the mean across all classes, sorting detection results by confidence and matching them against ground-truth boxes using IoU >= 0.5 as the true positive threshold. The tool was adapted from the official MATLAB implementation and is designed to work with output from object detectors like YOLO and Darknet.