Computer Vision Metrics: IoU, Precision-Recall & mAP by Hand
Evaluate an object detector cell by cell: box formats and YOLO coordinates, IoU with MAX and MIN, matching predictions to ground truth, precision and recall at thresholds, the PR curve, 11-point and all-point AP, mAP and mAP@[.5:.95], class confusion and Dice vs IoU for masks, every cell checked live.
The first lesson is free — no account needed.
About this course
A computer vision metrics course that evaluates one fictional warehouse robot's object detector cell by cell in a spreadsheet, for machine learning students, engineers and anyone who wants to understand what a mAP score means. You convert bounding boxes between corner, centre and normalised YOLO formats, calculate intersection over union with MAX and MIN and clamp boxes that do not overlap, match predictions to ground truth by confidence at IoU 0.5 to count true positives, false positives and misses, measure precision, recall and F1 at seven confidence thresholds, walk down the ranked detections to draw the precision-recall curve, calculate all-point and 11-point average precision, average it into mAP and COCO-style mAP@[.5:.95] and compare Dice with IoU for segmentation masks. The first lab is free.
Lessons
- Bounding Box Geometry: Corners, Centre & YOLO FormatFree
- Intersection over Union (IoU) with MAX and MIN Pro
- Matching Predictions to Ground Truth: TP, FP & FN at IoU ≥ 0.5 Pro
- Precision & Recall at Confidence Thresholds Pro
- The Precision-Recall Curve, Point by Point Pro
- Average Precision: 11-Point vs All-Point Interpolation Pro
- mAP Across Classes & mAP@[.5:.95] Pro
- Class Confusion & Segmentation Metrics: Dice vs IoU Pro