Object Pose 6D Estimation 中文
← Back to the metrics index

BOP / 6D detection aggregation · aggregate

BOP AP

BOP Average Precision for 6D Detection

Uses precision–recall curves to evaluate detection and pose jointly in 6D detection tasks where the objects present in the image are unknown.

Unlike localization AR, the model must both estimate poses correctly and suppress incorrect classes and duplicate detections. The current BOP detection evaluation is based on MSSD and MSPD.

Explore principles and formulas View related papers 5

Principle

What error does this metric penalize?

The formulas define the computation. The diagrams show whether the error concerns a 3D model, visible surfaces, the image plane, or aggregation across thresholds.

Localization is given the object identities and mainly asks whether their poses are correct; detection must also find all objects without false positives.

AP therefore penalizes poor confidence ranking, missed detections, and duplicate predictions, making it generally stricter than AR.

AP=APMSSD+APMSPD2AP=\frac{AP_{MSSD}+AP_{MSPD}}{2}
Each AP_e comes from a confidence-ranked precision–recall curve; follow the matching and threshold rules for the corresponding BOP challenge year.
recall / precisionThreshold / confidenceAggregate across settings
Illustration created for this site to explain the metric's error space; it does not show measured data or a figure from the original paper.

How to compute

Computation steps

  1. 01

    Output all candidates with object id, pose, and confidence.

  2. 02

    Match predictions to ground truth in confidence order and use MSSD/MSPD to determine true positives.

  3. 03

    Compute precision–recall curves and AP_e, then aggregate them into AP.

Strength

What it is useful for

  • More closely reflects end-to-end use of real multi-object systems.
  • Evaluates detection quality and pose quality within one protocol.

Caveat

Cautions when reading results

  • Do not equate detection AP with AR evaluated using given instance identities.
  • Subtasks and AP variants may differ between BOP challenge years.

Paper atlas

Which papers report this metric?

Links are derived by normalizing the metric names used in each paper. Directly reported metrics are distinguished from metrics associated through a broader protocol.

Filter the paper library
5Related papers
5Directly reported
0Linked through a protocol
2026 Directly reported

3D-Object Perception Transformer (3PT)

CVPR · Agastya Kalra, Tim Salzmann, Guy Stoppi et al.

Original terminology: 6DoF pose AP-mm / BOP ranking
2026 Directly reported

XYZ-IBD: Benchmarking Robust 6D Object Pose Estimation under Real-World Industrial Complexity

ECCV · Junwen Huang, Jiaqi Hu, Peter KT Yu et al.

Original terminology: BOP AP
2025 Directly reported

ZeroPose: CAD-Prompted Zero-Shot Object 6D Pose Estimation in Cluttered Scenes

IEEE TCSVT · Jianqiu Chen, Zikun Zhou, Mingshan Sun et al.

Original terminology: BOP AP
2024 Directly reported

SAM-6D: Segment Anything Model Meets Zero-Shot 6D Object Pose Estimation

CVPR · Jiehong Lin, Lihua Liu, Dekun Lu et al.

Original terminology: BOP AP
2024 Directly reported

SD-Net: Symmetric-Aware Keypoint Prediction and Domain Adaptation for 6D Pose Estimation in Bin-Picking Scenarios

IROS · Ding-Tao Huang, En-Te Lin, Lipeng Chen et al.

Original terminology: BOP AP

Next

Compare related metrics

Compare metrics at the same level to understand why one score improves while another stays unchanged.

MSSD

Maximum Symmetry-Aware Surface Distance

error
Minimizes the maximum 3D surface deviation over all allowed global symmetry transformations.

MSPD

Maximum Symmetry-Aware Projection Distance

error
Measures the worst model-point displacement in image pixels after accounting for symmetry.

BOP AR

BOP Average Recall

aggregate
Computes VSD, MSSD, and MSPD recall at multiple thresholds, then aggregates them into BOP's localization score.
Sources for definitions and protocolsExpand to view original papers, surveys, and official evaluation pages.