Research map / Detection and segmentation / CNN features
Soft-NMS -- Improving Object Detection With One Line of Code
Navaneeth Bodla, Bharat Singh, Rama Chellappa, Larry S. Davis
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 |
|---|---|---|---|
| replace NMS with Soft-NMS Gaussian | R-FCN · MS-COCO test-dev | AP 0.5:0.95: 31.1 -> 32.4 | Improved |
| replace NMS with Soft-NMS Gaussian | Faster-RCNN · MS-COCO test-dev | AP 0.5:0.95: 24.4 -> 25.5 | Improved |
| replace NMS with Soft-NMS Gaussian | Deformable-RFCN · MS-COCO test-dev | AP 0.5:0.95: 37.4 -> 38.4 | Improved |
| replace NMS with Soft-NMS Gaussian | Deformable-RFCN + multi-scale testing · MS-COCO test-dev | AP 0.5:0.95: 39.8 -> 40.9 | Improved |
| replace NMS with Soft-NMS | Faster-RCNN and R-FCN · PASCAL VOC 2007 test | AP 0.5:0.95: 37.7 -> 39.4 | Improved |
| replace NMS with Soft-NMS linear | SSD and YOLOv2 · PASCAL VOC 2007 test | AP: 0.5% | Improved |
| replace NMS with Soft-NMS Gaussian | R-FCN · MS-COCO test-dev | Recall@100: 43.6 -> 52.0 | Improved |
| replace NMS with Soft-NMS Gaussian | R-FCN · MS-COCO test-dev | AP small: 14.4 -> 15.2 | Improved |
| raise NMS threshold Nt from 0.5 to 0.7 (ablation) | R-FCN with NMS · MS-COCO minival | AP@0.5: 0.5227 -> 0.4894 | Hurt |
| raise Soft-NMS sigma from 0.5 to 0.9 (ablation) | R-FCN with Soft-NMS · MS-COCO minival | AP@0.8: 0.2669 -> 0.2710 | Improved |
| raise Soft-NMS sigma from 0.5 to 0.9 (ablation) | R-FCN with Soft-NMS · MS-COCO minival | AP@0.5: 0.5274 -> 0.5186 | Hurt |
| use best Soft-NMS sigma instead of best NMS threshold (ablation) | R-FCN · MS-COCO minival | AP: 1% | Improved |
Closest papers by meaning
- Accelerating Non-Maximum Suppression: A Graph Theory Perspective (NeurIPS 2024 · similarity 0.91)
- Learning Non-Maximum Suppression (CVPR 2017 · similarity 0.91)
- MaxpoolNMS: Getting Rid of NMS Bottlenecks in Two-Stage Object Detectors (CVPR 2019 · similarity 0.88)
- Oriented Tiny Object Detection: A Dataset, Benchmark, and Dynamic Unbiased Learning (TPAMI 2025 · similarity 0.88)
- GrooMeD-NMS: Grouped Mathematically Differentiable NMS for Monocular 3D Object Detection (CVPR 2021 · similarity 0.88)
- YOLOv10: Real-Time End-to-End Object Detection (NeurIPS 2024 · 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.
