Research map / Detection and segmentation / Open-set and few-shot detection
Multi-Label Learning of Part Detectors for Heavily Occluded Pedestrian Detection
Chunluan Zhou, Junsong Yuan
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 |
|---|---|---|---|
| learn part detectors jointly with shared trees instead of separately (P1, channel features) (ablation) | full-body detector · Caltech Reasonable, Partial, Heavy (channel features) | log-average miss rate (%): 18.2 -> 17.0; 36.1 -> 34.2; 72.1 -> 70.5 | Improved |
| joint learning of part detectors, averaged over P1, P4, P6, P11 (ablation) | part detectors · Caltech Reasonable, Partial, Heavy (channel features) | log-average miss rate (%): +1.6; +4.2; +2.0 | Improved |
| joint instead of separate learning for part P11 (ablation) | P11 detector · Caltech Partial (channel features) | log-average miss rate (%): 42.4 -> 35.2 | Improved |
| joint learning for full-body detector with CNN features (ablation) | full-body detector · Caltech Heavy (CNN features) | log-average miss rate (%): 56.6 -> 50.5 | Improved |
| joint learning for full-body detector with CNN features (ablation) | full-body detector · Caltech Reasonable (CNN features) | log-average miss rate (%): 10.3 -> 9.9 | Improved |
| joint learning for full-body detector with CNN features (ablation) | full-body detector · Caltech Partial (CNN features) | log-average miss rate (%): 18.0 -> 17.2 | Improved |
| integrate part detectors by averaging top-S scores (JL-TopS) over JL-P1, channel features (ablation) | JL-P1 · Caltech Reasonable, Partial, Heavy (channel features) | log-average miss rate (%): 17.0 -> 16.6; 34.2 -> 32.7; 70.5 -> 69.6 | Improved |
| integrate part detectors with top-S averaging over JL-P1, CNN features (ablation) | JL-P1 · Caltech Partial and Heavy (CNN features) | log-average miss rate (%): 17.2 -> 16.6; 50.5 -> 49.2 | Improved |
| integrate part detectors with top-S averaging over JL-P1, CNN features (ablation) | JL-P1 · Caltech Reasonable (CNN features) | log-average miss rate (%): 9.9 -> 10.0 | No clear effect |
| use top-S average instead of max to integrate part scores, channel features (ablation) | JL-Max · Caltech Reasonable, Partial, Heavy (channel features) | log-average miss rate (%): 17.5 -> 16.6; 34.8 -> 32.7; 68.8 -> 69.6 | Mixed |
| use top-S average instead of max to integrate part scores, CNN features (ablation) | JL-Max · Caltech Reasonable, Partial, Heavy (CNN features) | log-average miss rate (%): 10.3 -> 10.0; 17.2 -> 16.6; 48.4 -> 49.2 | Mixed |
| train with partially occluded positive examples (SL-P1 vs LDCF-P1), channel features (ablation) | LDCF-P1 · Caltech Partial (channel features) | log-average miss rate (%): 38.8 -> 36.1 | Improved |
Closest papers by meaning
- Improving Occlusion and Hard Negative Handling for Single-Stage Pedestrian Detectors (CVPR 2018 · similarity 0.88)
- Bi-box Regression for Pedestrian Detection and Occlusion Estimation (ECCV 2018 · similarity 0.88)
- Jointly Learning Deep Features, Deformable Parts, Occlusion and Classification for Pedestrian Detection (TPAMI 2017 · similarity 0.88)
- Mask-Guided Attention Network for Occluded Pedestrian Detection (ICCV 2019 · similarity 0.87)
- Occluded Pedestrian Detection Through Guided Attention in CNNs (CVPR 2018 · similarity 0.87)
- Multi-Part Object Representations via Graph Structures and Co-Part Discovery (NeurIPS 2026 · 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.
