Research map / Detection and segmentation / Weakly supervised detection and segmentation
Semi-Supervised Pedestrian Instance Synthesis and Detection With Mutual Reinforcement
Si Wu, Sihao Lin, Wenhao Wu, Mohamed Azzam, Hau-San Wong
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 class-specific discriminator Dspe (ablation) | Our Model w/o Dspe · Caltech1X (synthesis quality) | FID: 103.60 -> 44.18 | Improved |
| add class-specific discriminator Dspe (ablation) | Our Model w/o Dspe · Caltech1X (synthesis quality) | Inception score: 2.39 -> 2.74 | Improved |
| add class-conditional discriminator pair (Dcon+Dspe) instead of SN-GAN (ablation) | SN-GAN · Caltech1X (synthesis quality) | FID: 216.66 -> 44.18 | Improved |
| re-train base detector on pseudo-labeled data from the PRC | Faster R-CNN base detector · CUHK-Square (5% labeled) | log-average miss rate: 0.3467 -> 0.1924 | Improved |
| re-train base detector on pseudo-labeled data from the PRC | Faster R-CNN base detector · MIT-Traffic (5% labeled) | log-average miss rate: 0.3458 -> 0.1509 | Improved |
| re-train base detector on pseudo-labeled data from the PRC | Faster R-CNN base detector · Caltech1X (5% labeled) | log-average miss rate: 0.4565 -> 0.2379 | Improved |
| add synthesized instances when training the PRC (ablation) | PRC · Caltech1X (5% labeled) | log-average miss rate | Improved |
| add mean feature matching term (ablation) | PRC · Caltech1X (5% labeled) | log-average miss rate | Improved |
| compare PRC against fully supervised base detector | PRC trained with synthesized instances · CUHK-Square (5% labeled) | log-average miss rate | Improved |
| re-training the detector instead of using the PRC alone | Base detector re-trained vs PRC · Caltech1X (5% labeled) | log-average miss rate | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Faster R-CNN — Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks (TPAMI 2016) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Optimal Proposal Learning for Deployable End-to-End Pedestrian Detection (CVPR 2023 · similarity 0.85)
- What Can Help Pedestrian Detection? (CVPR 2017 · similarity 0.85)
- Progressive Refinement Network for Occluded Pedestrian Detection (ECCV 2020 · similarity 0.85)
- Where, What, Whether: Multi-Modal Learning Meets Pedestrian Detection (CVPR 2020 · similarity 0.85)
- Count- and Similarity-aware R-CNN for Pedestrian Detection (ECCV 2020 · similarity 0.85)
- Mask-Guided Attention Network for Occluded Pedestrian Detection (ICCV 2019 · similarity 0.85)
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.
