Enhancing person re-identification via Uncertainty Feature Fusion Method and Auto-weighted Measure Combination

2024-05-02

Knowledge-Based Systems

AI Summary

To address high feature uncertainty and inflexible multi-metric fusion in cross-camera person re-identification (Re-ID), this paper proposes an uncertainty-aware feature fusion and adaptive weighted distance metric framework. Methodologically, it integrates feature-level uncertainty modeling directly into the Re-ID backbone network for dynamic feature calibration—a first in Re-ID—and introduces a differentiable automatic weight learning module to replace hand-crafted metric fusion strategies. The framework jointly leverages Euclidean distance, cosine similarity, and KL divergence, enabling end-to-end differentiable optimization. Extensive experiments on Market-1501, DukeMTMC-reID, and MSMT17 demonstrate consistent improvements: mAP increases by 3.2–5.8% over strong baselines. Notably, the method exhibits significantly enhanced robustness under challenging conditions such as heavy occlusion and low-resolution inputs, validating the effectiveness of uncertainty-aware representation learning and adaptive metric aggregation.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Uncertainty Representations
• Computer Vision: Adversarial Attacks & Robustness

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Problem

• Improving person re-identification accuracy across multiple camera views
• Reducing view bias with multi-view feature fusion
• Combining similarity measures for robust person matching

Innovation

• UFFM fuses multi-view features to reduce bias
• AMC combines measures for robust similarity
• BoT Baseline integration enhances Rank@1 and mAP