Research map / Detection and segmentation / Open-set and few-shot detection
One Metric to Measure Them All: Localisation Recall Precision (LRP) for Evaluating Visual Detection Tasks
Kemal Öksüz, Barış Can Çam, Sinan Kalkan, Emre Akbaş
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| evaluate with oLRP Error instead of AP50 to account for localisation | Faster R-CNN X101, Libra R-CNN, Guided Anchoring · COCO 2017 val | detector ranking: equal AP50 63.7 | Improved |
| use oLRP Error instead of APC to compare NAS-FPN and GHM | oLRP Error · COCO 2017 val | performance gap: 1.1% -> 0.4% | Improved |
| use oLRP components to separate ATSS and Faster R-CNN with equal APC | oLRP Error · COCO 2017 val | oLRP Error: oLRP prefers Faster R-CNN by 1.1% | Improved |
| add PR curve interpolation to AP on the rare toaster class (ablation) | APC · COCO toaster class | APC: 2.2% on average, up to 6.2% | Hurt |
| switch evaluation API from Pascal API to COCO API (ablation) | AP50 · Pascal 2007 test | AP50: 81.3 -> 84.0 | Hurt |
| increase detections per image from 300 to 1K (ablation) | AP / oLRP Error · LVIS v1.0 | AP: 2.6% AP vs 1.4% oLRP difference | Mixed |
| increase detections per class from 1K to 20K (ablation) | AP / oLRP Error · LVIS v1.0 (fixed AP) | AP: 3.7% AP vs 1.5% oLRP difference | Mixed |
| use stronger backbone R50 to X101 | Mask R-CNN · LVIS v1.0 | oLRP Error: 80.7 -> 77.5 | Improved |
| reduce RPN proposals from 200 to 30 (ablation) | Faster R-CNN R101 · Open Images val | oLRP Error: 79.2 -> 79.9 | Hurt |
| add repeat factor sampling with t=0.01 (ablation) | Mask R-CNN R50 · LVIS v1.0 | oLRP Error: 86.2 -> 78.8 | Improved |
| tune NMS IoU threshold to 0.50 (ablation) | Faster R-CNN R50 · COCO | oLRP Error: 90.5 -> 69.1 | Improved |
| remove FN errors from LRP computation (ablation) | LRP Error · COCO panoptic with COCO-stuff | LRP Error: 77.5 -> 62.4 | Improved |
Closest papers by meaning
- Localization Recall Precision (LRP): A New Performance Metric for Object Detection (ECCV 2018 · similarity 0.91)
- A Ranking-based, Balanced Loss Function Unifying Classification and Localisation in Object Detection (NeurIPS 2020 · similarity 0.89)
- KaLOS finds Consensus: A Meta-Algorithm for Evaluating Inter-Annotator Agreement in Complex Vision Tasks (CVPR 2026 · similarity 0.84)
- Generalized Focal Loss V2: Learning Reliable Localization Quality Estimation for Dense Object Detection (CVPR 2021 · similarity 0.84)
- A Delay Metric for Video Object Detection: What Average Precision Fails to Tell (ICCV 2019 · similarity 0.83)
- How to Evaluate the Generalization of Detection? A Benchmark for Comprehensive Open-Vocabulary Detection (AAAI 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
