Research map / Detection and segmentation / CNN features
PSRR-MaxpoolNMS: Pyramid Shifted MaxpoolNMS With Relationship Recovery
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| replace GreedyNMS with PSRR-MaxpoolNMS | Faster-RCNN ResNet-50 · PASCAL VOC 2007 test | mAP (%): 78.1 -> 77.6 | Hurt |
| replace GreedyNMS with PSRR-MaxpoolNMS | SSD VGG-16 · PASCAL VOC 2007 test | mAP: 77.3 -> 76.1 | Hurt |
| replace GreedyNMS with PSRR-MaxpoolNMS | Faster-RCNN ResNet-152 · PASCAL VOC 2007 test | mAP: 78.7 -> 78.4 | Hurt |
| replace MaxpoolNMS with PSRR-MaxpoolNMS | Faster-RCNN ResNet-50 final detection stage · PASCAL VOC 2007 test | mAP (%): 33.0 -> 77.6 | Helped |
| add spatial and channel recovery (ablation) | MaxpoolNMS score-map construction · PASCAL VOC 2007 test | mAP (%): 41.2 -> 71.1 | Helped |
| add spatial recovery (ablation) | anchor-box score-map projection · PASCAL VOC 2007 test | mAP (%): 41.2 -> 44.4 | Helped |
| keep the highest-scoring box per score-map cell instead of random assignment (ablation) | Relationship Recovery · PASCAL VOC 2007 test | mAP (%): 75.1 -> 77.6 | Helped |
| keep the highest-scoring box per score-map cell instead of sum assignment (ablation) | Relationship Recovery · PASCAL VOC 2007 test | mAP (%): 76.5 -> 77.6 | Helped |
| use multiple sequential channel-combination pooling stages rather than single-scan pooling (ablation) | Pyramid MaxpoolNMS · PASCAL VOC 2007 test | mAP (%): 63.9 -> 77.6 | Helped |
| execute pyramid pooling in reverse order (ablation) | Pyramid MaxpoolNMS · PASCAL VOC 2007 test | mAP (%): 77.6 -> 77.4 | Hurt |
| add shifted max pooling to alleviate edge effects (ablation) | Pyramid MaxpoolNMS · PASCAL VOC 2007 test | mAP (%): 74.2 -> 77.6 | Helped |
| replace GreedyNMS with PSRR-MaxpoolNMS | Faster-RCNN ResNet-50 · KITTI test | mAP (easy to hard): 94.7\quad 89.8\quad 83.0 -> 93.4\quad 88.5\quad 82.8 | Mixed |
Closest papers by meaning
- MaxpoolNMS: Getting Rid of NMS Bottlenecks in Two-Stage Object Detectors (CVPR 2019 · similarity 0.94)
- Learning Non-Maximum Suppression (CVPR 2017 · similarity 0.88)
- Accelerating Non-Maximum Suppression: A Graph Theory Perspective (NeurIPS 2024 · similarity 0.88)
- End-to-End Object Detection With Fully Convolutional Network (CVPR 2021 · similarity 0.86)
- GrooMeD-NMS: Grouped Mathematically Differentiable NMS for Monocular 3D Object Detection (CVPR 2021 · similarity 0.86)
- SCNN: A General Distribution Based Statistical Convolutional Neural Network with Application to Video Object Detection (AAAI 2019 · similarity 0.86)
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.