Research map / Label-efficient and robust learning / Recognition and long-tail
Vehicle Re-Identification in Aerial Imagery: Dataset and Approach
Peng Wang, Bingliang Jiao, Lu Yang, Yifei Yang, Shizhou Zhang, Wei Wei, Yanning Zhang
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 ID classification loss to triplet loss (ablation) | Triplet loss ResNet-50 · VRAI test set | mAP: 46.99 -> 77.28 | Improved |
| add triplet loss to ID classification loss (ablation) | ID classification ResNet-50 · VRAI test set | mAP: 72.96 -> 77.28 | Improved |
| use deeper backbone ResNet-101 instead of ResNet-50 (ablation) | Triplet + ID loss · VRAI test set | mAP: 77.28 -> 77.48 | No clear effect |
| use ResNet-152 instead of ResNet-50 (ablation) | Triplet + ID loss · VRAI test set | mAP: 77.28 -> 77.54 | No clear effect |
| add attribute classification losses (color, type, others) (ablation) | Triplet + ID loss · VRAI test set | mAP: 77.28 -> 78.09 | Improved |
| add discriminative part weighted-feature branch with extra triplet loss (ablation) | Ours Multi-task · VRAI test set | mAP: 78.09 -> 78.63 | Improved |
| use weighted feature instead of average feature for distance computation (ablation) | Ours Multi-task + DP · VRAI test set | mAP: 78.31 -> 78.63 | Improved |
| use weighted feature instead of average feature for distance computation (ablation) | Ours Multi-task + DP · VRAI test set | CMC-5: 89.01 -> 88.49 | Hurt |
| vary weighting parameter gamma from 1.1 to 1.9 (ablation) | Ours Multi-task + DP · VRAI test set | mAP: 78.34 / 78.63 / 78.48 / 78.36 / 78.41 | No clear effect |
| replace horizontal-split part methods (MGN, RAM) by discriminative part detection | Ours Multi-task + DP · VRAI test set | mAP: 69.49 -> 78.63 | Improved |
| use VGG-16 instead of ResNet-50 for RAM | RAM · VRAI test set | mAP: 69.37 -> 57.33 | Hurt |
| compare algorithm with human annotators | Ours · VRAI human evaluation (100 queries) | accuracy: 97% -> 80% | Hurt |
Closest papers by meaning
- Disentangled Feature Learning Network and a Comprehensive Benchmark for Vehicle Re-Identification (TPAMI 2021 · similarity 0.89)
- VERI-Wild: A Large Dataset and a New Method for Vehicle Re-Identification in the Wild (CVPR 2019 · similarity 0.87)
- UCM-VeID V2: A Richer Dataset and A Pre-training Method for UAV Cross-Modality Vehicle Re-Identification (CVPR 2025 · similarity 0.85)
- AG-VPReID: A Challenging Large-Scale Benchmark for Aerial-Ground Video-based Person Re-Identification (CVPR 2025 · similarity 0.85)
- GSAlign: Geometric and Semantic Alignment Network for Aerial-Ground Person Re-Identification (NeurIPS 2025 · similarity 0.84)
- RemDet: Rethinking Efficient Model Design for UAV Object Detection (AAAI 2025 · similarity 0.83)
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.
