Research map / Detection and segmentation / CNN features
Improving Crowded Object Detection via Copy-Paste
Read the paperCheck 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 |
|---|---|---|---|
| Apply CrowdAug rather than the strong baseline | Faster R-CNN · CrowdHuman val | MR^-2: 42.46 -> 40.21 | Helped |
| Apply CrowdAug rather than the strong baseline | RetinaNet · CrowdHuman val | MR^-2: 50.65 -> 47.35 | Helped |
| Apply CrowdAug | Faster R-CNN · CityPersons val, heavy | MR^-2: 52.05 -> 48.50 | Helped |
| Apply CrowdAug | Faster R-CNN · KITTI val, cars, hard | AP@0.7: 79.44 -> 81.69 | Helped |
| Use crowdedness-oriented copy-paste rather than random copy-paste (ablation) | Faster R-CNN · CrowdHuman val | MR^-2: 42.01 -> 41.11 | Helped |
| Use crowdedness-oriented copy-paste rather than random copy-paste (ablation) | RetinaNet · CrowdHuman val | MR^-2: 49.75 -> 48.81 | Helped |
| Add consensus learning to CCP (ablation) | Faster R-CNN · CrowdHuman val | MR^-2: 41.11 -> 40.80 | Helped |
| Add consensus learning to CCP (ablation) | RetinaNet · CrowdHuman val | MR^-2: 48.81 -> 47.93 | Helped |
| Add overlay-depth prediction and OD-aware NMS to CCP + CL (ablation) | Faster R-CNN · CrowdHuman val | MR^-2: 40.80 -> 40.21 | Helped |
| Add overlay-depth prediction and OD-aware NMS to CCP + CL (ablation) | RetinaNet · CrowdHuman val | MR^-2: 47.93 -> 47.35 | Helped |
| Remove the pairwise score-alignment loss while keeping CL re-pasting (ablation) | Faster R-CNN with CCP + CL · CrowdHuman val | MR^-2: 40.80 -> 41.15 | Hurt |
| Remove the pairwise score-alignment loss while keeping CL re-pasting (ablation) | RetinaNet with CCP + CL · CrowdHuman val | MR^-2: 47.93 -> 48.88 | Hurt |
Closest papers by meaning
- Detection in Crowded Scenes: One Proposal, Multiple Predictions (CVPR 2020 · similarity 0.88)
- Simple Copy-Paste Is a Strong Data Augmentation Method for Instance Segmentation (CVPR 2021 · similarity 0.87)
- Learning to Generate Multiple Objects from Dense and Occluded Layouts (NeurIPS 2026 · similarity 0.83)
- HINTED: Hard Instance Enhanced Detector with Mixed-Density Feature Fusion for Sparsely-Supervised 3D Object Detection (CVPR 2024 · similarity 0.82)
- Image Copy Detection for Diffusion Models (NeurIPS 2024 · similarity 0.82)
- CaMuViD: Calibration-Free Multi-View Detection (CVPR 2025 · similarity 0.82)
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.