Research map / Detection and segmentation / CNN features
Occlusion-aware R-CNN: Detecting Pedestrians in a Crowd
Shifeng Zhang, Longyin Wen, Xiao Bian, Zhen Lei, Stan Z. Li
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 |
|---|---|---|---|
| add aggregation loss (AggLoss) to the baseline regression loss (ablation) | adaptive Faster R-CNN · CityPersons val Reasonable (x1.3) | MR-2: 12.5 -> 11.4 (reduces MR-2 by 1.1%) | Improved |
| add PORoI pooling in place of RoI pooling (ablation) | adaptive Faster R-CNN · CityPersons val Reasonable (x1.3) | MR-2: 12.5 -> 11.7 (reduces MR-2 by 0.8%) | Improved |
| add PORoI pooling on top of AggLoss (ablation) | OR-CNN-A · CityPersons val Reasonable (x1.3) | MR-2: 11.4 -> 11.0 | Improved |
| add AggLoss on top of PORoI pooling (ablation) | OR-CNN-P · CityPersons val Reasonable (x1.3) | MR-2: 11.7 -> 11.0 | Improved |
| add AggLoss and PORoI at x1 scale (ablation) | baseline · CityPersons val Reasonable (x1) | MR-2: 14.4 -> 12.8 | Improved |
| add AggLoss and PORoI (ablation) | baseline · CityPersons val Heavy (x1.3) | MR-2: 54.5 -> 51.3 | Improved |
| add AggLoss and PORoI (ablation) | baseline · CityPersons val Partial (x1.3) | MR-2: 16.8 -> 13.7 | Improved |
| add AggLoss and PORoI (ablation) | baseline · CityPersons val Bare (x1.3) | MR-2: 6.8 -> 5.9 | Improved |
| add AggLoss to reduce sensitivity to NMS threshold (ablation) | baseline · CityPersons val, FPPI=1e-2 NMS sweep | variance of miss rate: 0.230 -> 0.095 | Improved |
| fix part visibility scores to 1 instead of predicted visibility (ablation) | OR-CNN (PORoI) · CityPersons val | MR-2 | Hurt |
| use OR-CNN (VGG-16, x1.3) instead of Repulsion Loss (ResNet-50, x1.3) | detector · CityPersons val Reasonable | MR-2: 11.6 -> 11.0 | Improved |
| use OR-CNN instead of Repulsion Loss on CityPersons test | detector · CityPersons test Reasonable-Small | MR-2: 15.67 -> 14.19 | Improved |
Closest papers by meaning
- Mask-Guided Attention Network for Occluded Pedestrian Detection (ICCV 2019 · similarity 0.91)
- Occluded Pedestrian Detection Through Guided Attention in CNNs (CVPR 2018 · similarity 0.90)
- Count- and Similarity-aware R-CNN for Pedestrian Detection (ECCV 2020 · similarity 0.90)
- PedHunter: Occlusion Robust Pedestrian Detector in Crowded Scenes (AAAI 2020 · similarity 0.90)
- Bi-box Regression for Pedestrian Detection and Occlusion Estimation (ECCV 2018 · similarity 0.89)
- Multi-View Pedestrian Occupancy Prediction with a Novel Synthetic Dataset (AAAI 2025 · similarity 0.87)
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.
