Research map / Architectures and efficiency / RNNs and normalization
Decoupled and Memory-Reinforced Networks: Towards Effective Feature Learning for One-Step Person Search
Chuchu Han, Zhedong Zheng, Changxin Gao, Nong Sang, Yi Yang
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 |
|---|---|---|---|
| Separate detection and re-ID RoI heads (ablation) | Faster R-CNN with FPN on CUHK-SYSU · CUHK-SYSU | Re-ID mAP: 84.3 -> 87.5 | Improved |
| Use RoI features only for identification with a one-stage detector (ablation) | Faster R-CNN with FPN, then RetinaNet with FPN on CUHK-SYSU · CUHK-SYSU | Re-ID mAP: 87.5 -> 90.0 | Improved |
| Use ground-truth boxes only, excluding selected proposals (ablation) | RetinaNet with FPN on CUHK-SYSU · CUHK-SYSU | Re-ID mAP: 90.0 -> 90.3 | Improved |
| Add memory-reinforced feature learning (ablation) | Decoupled RetinaNet on CUHK-SYSU and PRW · CUHK-SYSU and PRW | mAP: 90.3 -> 91.2; 36.1 -> 44.6 | Improved |
| Add memory-reinforced feature learning (ablation) | Decoupled Reppoint on CUHK-SYSU and PRW · CUHK-SYSU and PRW | mAP: 92.4 -> 92.9; 39.1 -> 46.0 | Improved |
| Increase input resolution from 1333*800 to 1500*900 (ablation) | DMRNet with RetinaNet · CUHK-SYSU and PRW | mAP: 91.2 -> 91.6; 44.6 -> 46.1 | Improved |
| Increase input resolution from 1333*800 to 1500*900 (ablation) | DMRNet with Reppoint · CUHK-SYSU and PRW | mAP: 92.9 -> 93.2; 46.0 -> 46.9 | Improved |
| Set the slow-moving encoder momentum to 0.999 rather than 0 (ablation) | DMRNet on CUHK-SYSU · CUHK-SYSU | mAP: 90.6 -> 91.2 | Improved |
| Use a queue rather than a look-up table for feature memory (ablation) | Memory-reinforced feature learning on CUHK-SYSU · CUHK-SYSU | mAP: +0.3% mAP | Improved |
| Add more unlabeled negative instances to the queue (ablation) | Memory-reinforced pairwise learning with L=2048 or L=5532 · CUHK-SYSU | mAP | Improved |
| Add more unlabeled instances to OIM (ablation) | OIM loss · CUHK-SYSU | mAP | No clear effect |
| Increase labeled queue length when the unlabeled queue is empty (ablation) | Memory-reinforced feature learning · CUHK-SYSU | mAP | Improved |
Closest papers by meaning
- DMRNet++: Learning Discriminative Features With Decoupled Networks and Enriched Pairs for One-Step Person Search (TPAMI 2022 · similarity 0.94)
- Hierarchical Online Instance Matching for Person Search (AAAI 2020 · similarity 0.88)
- Joint Detection and Identification Feature Learning for Person Search (CVPR 2017 · similarity 0.88)
- Person Search via A Mask-guided Two-stream CNN Model (ECCV 2018 · similarity 0.87)
- Dual Attention Matching Network for Context-Aware Feature Sequence Based Person Re-Identification (CVPR 2018 · similarity 0.87)
- Divide and Conquer: Hybrid Pre-training for Person Search (AAAI 2024 · 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.
