Research map / Detection and segmentation / CNN features
Confidence Propagation Cluster: Unleash Full Potential of 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 NMS with CP-Cluster | YOLOv5 detectors · COCO test-dev | AP: 37.1 -> 37.4; 45.5 -> 45.8; 49.0 -> 49.3; 50.7 -> 51.1; 44.3 -> 44.6; 51.2 -> 51.5; 53.8 -> 54.1; 55.1 -> 55.5 | Helped |
| Replace NMS-based post-processing with CP-Cluster | MMDetection detectors · COCO val/test-dev | mAP: 29.5/29.6 -> 30.1/30.1; 38.4/38.7 -> 39.2/39.4; 42.7/42.8 -> 43.0/43.2; 37.4/37.7 -> 38.1/38.4; 33.5/33.5 -> 34.1/34.1; | Helped |
| Replace maxpool post-processing with CP-Cluster | CenterNet · COCO test-dev | AP: 37.3 -> 39.2; 41.7 -> 43.3; 40.2 -> 41.1; 45.2 -> 46.6 | Helped |
| Replace NMS with CP-Cluster | Mask R-CNN · COCO test-dev | Box AP: 41.5 -> 42.2; 43.1 -> 43.7; 44.6 -> 45.2 | Helped |
| Replace NMS with CP-Cluster | Mask R-CNN · COCO test-dev | Mask AP: 37.7 -> 38.1; 38.8 -> 39.2; 40.0 -> 40.2 | Helped |
| Use Soft-NMS instead of maxpool | CenterNet with multi-scale fusion · COCO test-dev | AP: 41.7 -> 40.6; 45.2 -> 44.3 | Hurt |
| Use Soft-NMS instead of NMS | YOLOv5 · COCO test-dev | mAP | No clear effect |
| Add weaker-friends aggregation (WFA) before suppression (ablation) | Soft-NMS · COCO val/test-dev | mAP: 29.8/29.9 -> 30.0/30.0; 39.0/39.2 -> 39.1/39.3; 42.7/42.8 -> 42.8/42.9; 37.5/37.9 -> 37.7/38.2; 33.8/33.8 -> 33.6/33.7; | Helped |
| Use two CP-Cluster iterations (ablation) | YOLOv5s · COCO val | mAP | Helped |
| Use different strong-neighbor selection criteria by iteration: confidence first, overlap second (ablation) | CP-Cluster · YOLOv5s ablation | mAP | Helped |
| Use a smaller incremental overlap-threshold increase (ablation) | CP-Cluster · YOLOv5s ablation | AP50 | Helped |
| Use a smaller incremental overlap-threshold increase (ablation) | CP-Cluster · YOLOv5s ablation | AP75 | Hurt |
Closest papers by meaning
- Learning Non-Maximum Suppression (CVPR 2017 · similarity 0.87)
- Accelerating Non-Maximum Suppression: A Graph Theory Perspective (NeurIPS 2024 · similarity 0.86)
- FSOD-VFM: Few-Shot Object Detection with Vision Foundation Models and Graph Diffusion (ICLR 2026 · similarity 0.86)
- Unbalanced Optimal Transport: A Unified Framework for Object Detection (CVPR 2023 · similarity 0.86)
- MaxpoolNMS: Getting Rid of NMS Bottlenecks in Two-Stage Object Detectors (CVPR 2019 · similarity 0.86)
- Unsupervised Object Detection with Theoretical Guarantees (NeurIPS 2024 · similarity 0.85)
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.