Skip to content
EvalSuite
Documentation menu

Object detection

mAP@[.50:.95]

Implementeddetection.map50_95

Definition

Mean AP over classes and over IoU thresholds 0.50, 0.55, …, 0.95.

Formula

mAP = (1/10) Σ_{t ∈ {0.50, …, 0.95}} (1/K) Σₖ AP_k(t)

Range: [0, 1]

Inputs and outputs

  • ground truth: boxes [x1, y1, x2, y2] with class labels
  • predictions: boxes with class labels and confidence scores

Returns: float

Assumptions

No assumptions beyond valid, aligned inputs of the documented types.

Limitations

  • Follows the COCO protocol (crowd regions, area ranges, 101-point interpolation); all twelve numbers match pycocotools in the test suite.

Python API

PythonSince v0.3.0
import evalsuite as es

es.mean_average_precision(gt, preds)                     # COCO mAP@[.50:.95]
es.mean_average_precision(gt, preds, iou_threshold=0.5)  # mAP@.50
es.detection_report(gt, preds)                           # all 12 COCO numbers

References

  1. Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., & Zitnick, C. L. (2014). Microsoft COCO: Common objects in context. European Conference on Computer Vision (ECCV), 740–755.

Implementation status