Research map / Detection and segmentation / CNN features
Localization Recall Precision (LRP): A New Performance Metric for Object Detection
Kemal Oksuz, Baris Can Cam, Emre Akbas, Sinan Kalkan
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 |
|---|---|---|---|
| add FPN to Faster R-CNN R50 (ablation) | Faster R-CNN (R50) · MS COCO val2017 (80 classes) | moLRP: 0.714 -> 0.689 | Improved |
| deepen backbone from R50+FPN to R101+FPN (ablation) | Faster R-CNN · MS COCO val2017 (80 classes) | moLRP: 0.689 -> 0.673 | Improved |
| switch backbone R101+FPN to X101+FPN (ablation) | Faster R-CNN · MS COCO val2017 (80 classes) | moLRP: 0.673 -> 0.663 | Improved |
| increase SSD input size from 300 to 512 (ablation) | SSD · MS COCO val2017 (80 classes) | moLRP: 0.854 -> 0.763 | Improved |
| use ResNeXt-101 instead of ResNet-50 backbone (ablation) | RetinaNet (R50+FPN) · MS COCO val2017 (80 classes) | moLRP: 0.711 -> 0.677 | Improved |
| use two-step Faster R-CNN (R101+FPN) instead of RetinaNet (X101+FPN) at equal mAP (ablation) | RetinaNet (X101+FPN) · MS COCO val2017 (80 classes) | moLRP: 0.677 -> 0.673 | Mixed |
| add box linking across frames with general threshold 0.5 (ablation) | B (still-image RetinaNet-50) · ImageNet VID val (346 videos, 15 classes) | mean AP: 0.619 -> 0.544 | Hurt |
| add box linking across frames with general threshold 0.5 (ablation) | B (still-image RetinaNet-50) · ImageNet VID val (346 videos, 15 classes) | mean oLRP: 0.669 -> 0.656 | Improved |
| replace general threshold with class-specific oLRP thresholds (ablation) | G · ImageNet VID val (346 videos, 15 classes) | mean AP: 0.544 -> 0.561 | Improved |
| replace general threshold with class-specific oLRP thresholds (ablation) | G · ImageNet VID val (346 videos, 15 classes) | mean oLRP: 0.656 -> 0.651 | Improved |
| replace general threshold with class-specific oLRP thresholds (ablation) | G · ImageNet VID val (346 videos, 15 classes) bus class | AP and oLRP: 8.7% in AP and 4.1% in oLRP | Improved |
| replace general threshold with class-specific oLRP thresholds (ablation) | G · ImageNet VID val (346 videos, 15 classes) | mean oLRP_FN: 0.424 -> 0.398 | Improved |
Closest papers by meaning
- One Metric to Measure Them All: Localisation Recall Precision (LRP) for Evaluating Visual Detection Tasks (TPAMI 2021 · similarity 0.91)
- A Ranking-based, Balanced Loss Function Unifying Classification and Localisation in Object Detection (NeurIPS 2020 · similarity 0.90)
- LocNet: Improving Localization Accuracy for Object Detection (CVPR 2016 · similarity 0.88)
- DeNet: Scalable Real-Time Object Detection With Directed Sparse Sampling (ICCV 2017 · similarity 0.87)
- Region-Based Convolutional Networks for Accurate Object Detection and Segmentation (TPAMI 2016 · similarity 0.87)
- Weakly Supervised Object Localization and Detection: A Survey (TPAMI 2021 · similarity 0.87)
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.
