Modeling Sensor Uncertainty in Cross-Modal Contrastive Learning for Pedestrian Re-Identification
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Pedestrian re-identification in autonomous driving is affected by heterogeneous sensor conditions and modality-dependent noise. Building on CLIP-based uncertainty-aware modeling, this paper presents a cross-modal contrastive learning framework that aligns visual and textual pedestrian representations while accounting for sensor uncertainty. Modality-specific confidence weights are introduced to reduce the influence of unreliable features during representation learning. The method is evaluated on three autonomous driving datasets, including nuScenes-ReID and two large-scale urban benchmarks, comprising more than 120,000 pedestrian samples and 45,000 identities. Comparisons are conducted against recent vision-only and vision–language baselines, including PCB, MGN, TransReID, and CLIP-based ReID models. Experimental results show improvements of 3.9%–5.0% in rank-1 accuracy and 4.2%–5.6% in mean average precision under low-visibility and adverse weather conditions.
Fields of Study & Subjects
Video Surveillance and Tracking Methods
Autonomous Vehicle Technology and Safety
Advanced Neural Network Applications
Publication details
- Journal / source
- American Journal of Machine Learning
- Volume
- 7
- Issue
- 1
- Pages
- 69–76
- Publication date
- 31 January 2026
- Language
- EN