Research map / Detection and segmentation / CNN features
Detection in Crowded Scenes: One Proposal, Multiple Predictions
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 one prediction per proposal with multiple-instance prediction trained using EMD loss (ablation) | FPN, ResNet-50 · CrowdHuman validation | AP: 85.8 -> 87.4 | Helped |
| Add Set NMS, retaining overlapping predictions from the same proposal (ablation) | FPN with MIP and EMD loss · CrowdHuman validation | AP: 87.4 -> 90.3 | Helped |
| Add the refinement module (ablation) | FPN with MIP and Set NMS · CrowdHuman validation | MR^{-2}: 42.2 -> 41.4 | Helped |
| Add the refinement module (ablation) | FPN with MIP and Set NMS · CrowdHuman validation | AP: 90.3 -> 90.7 | Helped |
| Increase prediction head count from K = 2 to K = 3 (ablation) | Proposed detector · CrowdHuman validation | AP: 90.7 -> 90.7 | No clear effect |
| Increase prediction head count from K = 2 to K = 3 (ablation) | Proposed detector · CrowdHuman validation | MR^{-2}: 41.4 -> 41.6 | Hurt |
| Use multiple-instance prediction with EMD loss and Set NMS | FPN, ResNet-50 · CrowdHuman validation | AP: 85.8 -> 90.3 | Helped |
| Use multiple-instance prediction with EMD loss and Set NMS | FPN, ResNet-50 · CrowdHuman validation | MR^{-2}: 42.9 -> 42.2 | Helped |
| Use multiple-instance prediction with EMD loss and Set NMS | FPN, ResNet-50 · CrowdHuman validation | JI: 79.8 -> 82.0 | Helped |
| Raise naive NMS IoU threshold from 0.5 to 0.6 | FPN baseline · CrowdHuman validation | MR^{-2}: 42.9 -> 45.4 | Hurt |
| Apply Soft-NMS instead of naive NMS | FPN baseline · CrowdHuman validation | AP: 85.8 -> 88.2 | Helped |
| Apply the proposed method | FPN, ResNet-50 · CityPersons validation reasonable | AP: 95.2% -> 96.1% | Helped |
Closest papers by meaning
- Progressive End-to-End Object Detection in Crowded Scenes (CVPR 2022 · similarity 0.90)
- Improving Crowded Object Detection via Copy-Paste (AAAI 2023 · similarity 0.88)
- NMS by Representative Region: Towards Crowded Pedestrian Detection by Proposal Pairing (CVPR 2020 · similarity 0.88)
- Looking Beyond the Known: Towards a Data Discovery Guided Open-World Object Detection (NeurIPS 2025 · similarity 0.84)
- Point-to-Region Loss for Semi-Supervised Point-Based Crowd Counting (CVPR 2025 · similarity 0.84)
- Union-over-Intersections: Object Detection beyond Winner-Takes-All (ICLR 2025 · 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.