Research map / Detection and segmentation / CNN features
MaxpoolNMS: Getting Rid of NMS Bottlenecks in Two-Stage Object Detectors
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 RPN GreedyNMS with MaxpoolNMS multi-scale pooling | Faster R-CNN+ResNet-101-V1 · COCO test-dev2017 | mAP@IoU 0.50:0.95: 30.1 -> 30.1 | No clear effect |
| Replace RPN GreedyNMS with MaxpoolNMS multi-scale pooling | R-FCN+ResNet-101-V1 · COCO test-dev2017 | mAP@IoU 0.50:0.95: 27.0 -> 27.3 | No clear effect |
| Replace GreedyNMS with MaxpoolNMS multi-scale pooling | Faster R-CNN+ResNet-50-V2 · KITTI val | mAP: 80.09 -> 80.42 | No clear effect |
| Replace GreedyNMS with MaxpoolNMS multi-scale pooling | R-FCN+ResNet-50-V2 · KITTI val | mAP: 78.76 -> 78.48 | No clear effect |
| Replace GreedyNMS with MaxpoolNMS multi-scale pooling | Faster R-CNN+ResNet-50-V2 · VOC2007 test | mAP: 74.28 -> 73.85 | No clear effect |
| Replace GreedyNMS with MaxpoolNMS multi-scale pooling | R-FCN+ResNet-50-V2 · VOC2007 test | mAP: 70.93 -> 71.01 | No clear effect |
| Add cross-scale multi-channel pooling to MS (ablation) | Faster R-CNN+ResNet-50-V2 · VOC2007 test | mAP: 73.85 -> 69.16 | Hurt |
| Add cross-scale multi-channel pooling to MS (ablation) | R-FCN+ResNet-50-V2 · VOC2007 test | mAP: 71.01 -> 69.03 | Hurt |
| Add cross-aspect-ratio multi-channel pooling to MS (ablation) | Faster R-CNN+ResNet-50-V2 · VOC2007 test | mAP: 73.85 -> 74.38 | Helped |
| Add cross-aspect-ratio multi-channel pooling to MS (ablation) | R-FCN+ResNet-50-V2 · VOC2007 test | mAP: 71.01 -> 71.31 | Helped |
| Replace GreedyNMS with MaxpoolNMS | Faster R-CNN+ResNet-152-V2 · KITTI timing, score threshold 0 (n=53280) | execution time (ms): 31.69(18.19/13.50) -> 6.67(6.22/0.45) | Helped |
| Use cross-aspect-ratio multi-channel pooling rather than MS (ablation) | Faster R-CNN+ResNet-152-V2 · KITTI timing, score threshold 0 (n=53280) | execution time (ms): 6.67(6.22/0.45) -> 3.44(2.75/0.69) | Helped |
Closest papers by meaning
- PSRR-MaxpoolNMS: Pyramid Shifted MaxpoolNMS With Relationship Recovery (CVPR 2021 · similarity 0.94)
- Learning Non-Maximum Suppression (CVPR 2017 · similarity 0.89)
- Accelerating Non-Maximum Suppression: A Graph Theory Perspective (NeurIPS 2024 · similarity 0.89)
- Exploit All the Layers: Fast and Accurate CNN Object Detector With Scale Dependent Pooling and Cascaded Rejection Classifiers (CVPR 2016 · similarity 0.87)
- GrooMeD-NMS: Grouped Mathematically Differentiable NMS for Monocular 3D Object Detection (CVPR 2021 · similarity 0.87)
- YOLOv10: Real-Time End-to-End Object Detection (NeurIPS 2024 · similarity 0.84)
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.