DataTerminology.comNEW
Also known as: Mean Average Precision
Mean Average Precision averages the average precision across queries or classes to assess ranked retrieval and detection quality.
DESCRIPTION
Mean Average Precision aggregates Average Precision across queries or classes, summarizing ranked retrieval performance. In object detection, mAP averages AP across IoU thresholds and categories. It captures both precision at various recall levels and ranking quality. Proper evaluation requires consistent IoU definitions, class weighting, and dataset splits. mAP complements other metrics like F1 and ROC AUC to provide nuanced model assessment.
ALTERNATIVE DEFINITIONS
Wikipedia
Mean average precision (mAP) is the mean of the average precision scores for a set of queries.
Concise definition grounded in information retrieval.
Microsoft Learn
Mean average precision (mAP) summarizes precision-recall performance by averaging average precision across classes or IoU thresholds in object detection.
Focuses on object detection context and IoU thresholds.
Wikipedia
Mean average precision (MAP) is the mean of the average precision scores for a set of queries.
Defines MAP succinctly in information retrieval context.
COCO
Average Precision (AP) is measured across IoU thresholds (0.50:0.95). The main challenge metric is AP averaged across categories and across IoUs: AP@[.50:.95] (mAP).
Defines mAP for object detection with IoU averaging specifics.
RELATED TERMS
Precision Recall Precision-Recall AUC Object Detection Information Retrieval Ranking Evaluation Metrics Precision-Recall Curve Intersection over Union Ranking Evaluation Object Detection Evaluation Confusion Matrix
COMPETING TERMS
NDCG MRR AUC-PR F1 Score ROC AUC
RELATIONS
- Narrower: Choropleth Map Feature Map Monad Map Saliency Map Theme Map
KNOWLEDGE AREAS
AI Data Science Testing & Validation
TARGET AUDIENCE
Data Scientist AI Engineer MLOps Engineer Information Analyst BI Specialist
EXTERNAL SOURCES
- Wikipedia - Mean average precision
- Wikipedia - General definition and context
- Microsoft Learn - Detection metrics including mAP
- COCO Dataset - COCO detection evaluation protocol
- TREC (NIST) - IR evaluation benchmarks and measures
- Papers with Code - Leaderboards using mAP metric
- scikit-learn - Precision-Recall, AP definitions
TOP QUESTIONS
Why is MAP used in ranking evaluation?
MAP reflects precision across ranks and queries, rewarding correct early results. It provides a comprehensive summary of retrieval quality.
Category: Relevancy
How is MAP computed?
Compute average precision per query or class, then take the mean over all queries or classes.
Category: Requirements
What is the difference between AP and MAP?
AP is per query or class; MAP averages AP values across many queries or classes to provide a global score.
Category: Misunderstanding
How does MAP apply to object detection?
mAP averages AP across classes and often across IoU thresholds, summarizing detection performance comprehensively.
Category: Example
What are common mistakes when using MAP?
Using inconsistent IoU thresholds, improper ranking, or mismatched ground truth formats can invalidate results.
Category: Mistakes
What risks exist when optimizing only for MAP?
Overfitting to evaluation specifics and neglecting fairness or latency are risks. Use multiple metrics and constraints.
Category: Risk
What are alternatives to MAP for ranking quality?
NDCG and MRR are common alternatives. AUC-PR focuses on the precision-recall tradeoff without rank emphasis.
Category: Alternatives
Where should beginners learn about MAP?
Start with the Wikipedia article, then study datasets like COCO and library implementations.
Category: Getting Started
What is a hot topic around MAP today?
Evaluations that consider bias, diversity, and user-centric metrics beyond MAP are gaining attention.
Category: Topic