Skip to main content

mAR

Measure detection recall with limits of 1, 10 and 100 detections per image.

Join waitlist
Testing group
Performance
Applies to
Detection
App version
2.0.0
Channel
stable

Required inputs

A model, task-compatible labelled evaluation data and the prediction settings used for the run.

Test procedure

Run the metric on the chosen evaluation split. Keep the task, label mapping, thresholds and aggregation settings with the result.

Interpreting results

Read the score against the task and the errors that matter. Inspect class or slice results before relying on a single aggregate.

Limitations

Evidence record

Retain predictions, reference labels, metric settings and the task-level result so comparisons use the same conditions.