Research map / Detection and segmentation / CNN features
Joint Detection and Identification Feature Learning for Person Search
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| jointly optimize pedestrian detection and identification in one CNN rather than using separate stages | Person search · CUHK-SYSU test (gallery size 100), CNN detections | CMC top-1 (%): 74.8 -> 78.7 | Helped |
| jointly optimize pedestrian detection and identification in one CNN rather than using separate stages | Person search · CUHK-SYSU test (gallery size 100), CNN detections | mAP (%): 68.6 -> 75.5 | Helped |
| add unlabeled-identity features as negative examples in OIM (ablation) | Ours · CUHK-SYSU test (gallery size 100), CNN detections | CMC top-1 (%): 76.1 -> 78.7 | Helped |
| add unlabeled-identity features as negative examples in OIM (ablation) | Ours · CUHK-SYSU test (gallery size 100), CNN detections | mAP (%): 72.7 -> 75.5 | Helped |
| replace Softmax loss with OIM loss | Inception re-identification model · CUHK03 | CMC top-1 accuracy (%): 73.2 -> 77.7 | Helped |
| replace Softmax loss with OIM loss | Inception re-identification model · Market1501 | CMC top-1 accuracy (%): 75.8 -> 77.9 | Helped |
| replace Softmax loss with OIM loss | Inception re-identification model · Duke | CMC top-1 accuracy (%): 54.4 -> 61.7 | Helped |
| replace Softmax loss with OIM loss | ResNet-50 re-identification model · CUHK03 | CMC top-1 accuracy (%): 70.8 -> 77.5 | Helped |
| replace Softmax loss with OIM loss | ResNet-50 re-identification model · Market1501 | CMC top-1 accuracy (%): 81.4 -> 82.1 | Helped |
| replace Softmax loss with OIM loss | ResNet-50 re-identification model · Duke | CMC top-1 accuracy (%): 62.5 -> 68.1 | Helped |
| project normalized pooled features into a 256-dimensional subspace rather than use the 2048-dimensional vector directly (ablation) | Person search model · CUHK-SYSU test (gallery size 100), CNN detections | CMC top-1 (%): 59.3 -> 78.7 | Helped |
| reduce feature subspace dimension from 256 to 128 (ablation) | Person search model · CUHK-SYSU test (gallery size 100), CNN detections | CMC top-1 (%): 78.7 -> 65.9 | Hurt |
Closest papers by meaning
- Hierarchical Online Instance Matching for Person Search (AAAI 2020 · similarity 0.92)
- Exploring Visual Context for Weakly Supervised Person Search (AAAI 2022 · similarity 0.89)
- Deep Neural Networks with Inexact Matching for Person Re-Identification (NeurIPS 2016 · similarity 0.89)
- Sequential End-to-end Network for Efficient Person Search (AAAI 2021 · similarity 0.88)
- Diverse Knowledge Distillation for End-to-End Person Search (AAAI 2021 · similarity 0.88)
- Divide and Conquer: Hybrid Pre-training for Person Search (AAAI 2024 · 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.