mAP

Why we included this project

Training a detector is only half the job; you also need a defensible way to grade it. This is a clean Python port of the mAP metric from PASCAL VOC 2012, which compares your model's detections against ground-truth boxes, matches them by label and IoU, builds precision/recall curves per class, and collapses them into a single score. Because it follows the official MATLAB reference implementation, the numbers line up with published results. Setup is just dropping ground-truth and detection text files into the input folders and running one script, with converters included for common formats like PASCAL VOC, YOLO, and darkflow. It's a small utility rather than a framework, so it slots into an evaluation step after your own pipeline has produced detections.