Research map / Detection and segmentation / CNN features
Count- and Similarity-aware R-CNN for Pedestrian Detection
Jin Xie, Hisham Cholakkal, Rao Muhammad Anwer, Fahad Shahbaz Khan, Yanwei Pang, Ling Shao, Mubarak Shah
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 count-weighted loss (CW-loss) to baseline (ablation) | Baseline detector · CityPersons val (vis>=65%, x1.3) | MR-2 Heavy Occlusion (%): 53.5 -> 51.5 | Improved |
| add count-and-similarity branch (CSB) with CW-loss (ablation) | CaSe · CityPersons val (vis>=65%, x1.3) | MR-2 Heavy Occlusion (%): 51.5 -> 49.3 | Improved |
| add CAS-NMS using CSB predictions (ablation) | CaSe · CityPersons val (vis>=65%, x1.3) | MR-2 Heavy Occlusion (%): 49.3 -> 48.2 | Improved |
| add all CaSe components to baseline (x1.0) (ablation) | CaSe · CityPersons val (vis>=65%, x1.0) | MR-2 Heavy Occlusion (%): 57.0 -> 50.3 | Improved |
| add similarity embedding to CAS-NMS (vs count only) (ablation) | CaSe · CityPersons val (vis>=65%, x1.0) | MR-2 Heavy Occlusion (%): 53.1 -> 50.3 | Improved |
| replace Repulsion Loss with CW-loss | CaSe · CityPersons val (vis>=65%, x1.3) | MR-2 Heavy Occlusion (%): 55.3 -> 51.5 | Improved |
| replace Aggregation Loss with CW-loss | CaSe · CityPersons val (vis>=50%, x1.3) | MR-2 Heavy Occlusion (%): 52.6 -> 47.1 | Improved |
| use CAS-NMS instead of Adaptive-NMS (x1.3) | CaSe · CityPersons val (vis>=65%, x1.3) | MR-2 Heavy Occlusion (%): 54.0 -> 48.2 | Improved |
| use CaSe instead of MGAN without VBB supervision | CaSe · CityPersons val (best HO setting, x1.3) | MR-2 Heavy Occlusion (%): 39.4 -> 37.4 | Improved |
| use CaSe instead of MGAN on test set | CaSe · CityPersons test | MR-2 Heavy Occlusion (%): 41.0 -> 38.6 | Improved |
| add PedHunter training strategies to CaSe | CaSe · CityPersons val | MR-2 (%): 8.0% and 41.2% | Improved |
| use CaSe instead of MGAN on CrowdHuman | CaSe · CrowdHuman | MR-2 (%): 49.3 -> 47.9 | Improved |
Closest papers by meaning
- Occlusion-aware R-CNN: Detecting Pedestrians in a Crowd (ECCV 2018 · similarity 0.90)
- Mask-Guided Attention Network for Occluded Pedestrian Detection (ICCV 2019 · similarity 0.89)
- Bi-box Regression for Pedestrian Detection and Occlusion Estimation (ECCV 2018 · similarity 0.88)
- Is Faster R-CNN Doing Well for Pedestrian Detection? (ECCV 2016 · similarity 0.88)
- NMS by Representative Region: Towards Crowded Pedestrian Detection by Proposal Pairing (CVPR 2020 · similarity 0.88)
- Point-to-Region Loss for Semi-Supervised Point-Based Crowd Counting (CVPR 2025 · similarity 0.86)
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.
