Research map / Detection and segmentation / CNN features
Repulsion Loss: Detecting Pedestrians in a Crowd
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 |
|---|---|---|---|
| Add RepGT and RepBox losses | Faster R-CNN, ResNet-50 · CityPersons validation, Reasonable | MR^-2: 14.6 -> 13.2 | Helped |
| Add RepGT and RepBox losses | Faster R-CNN, ResNet-50 · CityPersons validation, Heavy | MR^-2: 60.6 -> 56.9 | Helped |
| Add RepGT and RepBox losses | Faster R-CNN, ResNet-50 · CityPersons validation, Bare | MR^-2: 7.9 -> 7.6 | Helped |
| Add RepGT and RepBox losses; pre-train on CityPersons | Faster R-CNN, ResNet-50 · Caltech-USA test, Reasonable, IoU=0.5 | MR^-2: 5.1 -> 4.0 | Helped |
| Add RepGT loss | Faster R-CNN, ResNet-101 · PASCAL VOC 2007 test, Crowd | mAP: 38.7 -> 40.8 | Helped |
| Use sigma=1.0 for RepGT (ablation) | Faster R-CNN, ResNet-50 with RepGT · CityPersons validation, Reasonable | MR^-2: 14.6 -> 13.7 | Helped |
| Use sigma=0 for RepBox (ablation) | Faster R-CNN, ResNet-50 with RepBox · CityPersons validation, Reasonable | MR^-2: 14.6 -> 13.7 | Helped |
| Set RepGT weight alpha=0.5 and RepBox weight beta=0.5 (ablation) | Faster R-CNN, ResNet-50 with RepGT and RepBox · CityPersons validation, Reasonable | MR^-2: 13.9 -> 13.2 | Helped |
| Add RepGT loss alone (ablation) | Faster R-CNN, ResNet-50 · CityPersons validation, Reasonable | MR^-2: 14.6 -> 13.7 | Helped |
| Add RepBox loss alone (ablation) | Faster R-CNN, ResNet-50 · CityPersons validation, Reasonable | MR^-2: 14.6 -> 13.7 | Helped |
| Add RepGT loss | Faster R-CNN, ResNet-101 · PASCAL VOC 2007 test, all objects | mAP: 79.5 -> 79.8 | No clear effect |
| Increase test scale to ×1.3 (ablation) | RepLoss (RepGT + RepBox) · CityPersons validation, Reasonable | MR^-2: 13.2 -> 11.6 | Helped |
Closest papers by meaning
- NMS by Representative Region: Towards Crowded Pedestrian Detection by Proposal Pairing (CVPR 2020 · similarity 0.86)
- Adaptive NMS: Refining Pedestrian Detection in a Crowd (CVPR 2019 · similarity 0.86)
- Improving Occlusion and Hard Negative Handling for Single-Stage Pedestrian Detectors (CVPR 2018 · similarity 0.86)
- PedHunter: Occlusion Robust Pedestrian Detector in Crowded Scenes (AAAI 2020 · similarity 0.85)
- Detection in Crowded Scenes: One Proposal, Multiple Predictions (CVPR 2020 · similarity 0.85)
- Point-to-Region Loss for Semi-Supervised Point-Based Crowd Counting (CVPR 2025 · similarity 0.83)
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.