Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

QA-ReID: Quality-Aware Query-Adaptive Convolution Leveraging Fused Global and Structural Cues for Clothes-Changing ReID

Yuxiang Wang, Kunming Jiang, Tianxiang Zhang, Ke Tian, Gaozhe Jiang

TL;DR

The paper tackles clothes-changing person re-identification by introducing QA-ReID, a dual-branch framework that fuses RGB-based global appearance with parsing-based structural cues through a multi-modal attention module. It advances matching with QAConv-QA, which applies pixel-level quality weighting and bidirectional constraints to focus on informative body regions while mitigating clothing variations. The method is trained with a joint loss combining identity, triplet, and pairwise matching objectives, and it achieves state-of-the-art results on CC-ReID benchmarks PRCC, LTCC, and VC-Clothes, demonstrating robustness to clothing changes. The approach offers practical impact for surveillance and related applications by improving cross-clothing re-identification performance while providing interpretable attention patterns that emphasize identity-related regions.

Abstract

Unlike conventional person re-identification (ReID), clothes-changing ReID (CC-ReID) presents severe challenges due to substantial appearance variations introduced by clothing changes. In this work, we propose the Quality-Aware Dual-Branch Matching (QA-ReID), which jointly leverages RGB-based features and parsing-based representations to model both global appearance and clothing-invariant structural cues. These heterogeneous features are adaptively fused through a multi-modal attention module. At the matching stage, we further design the Quality-Aware Query Adaptive Convolution (QAConv-QA), which incorporates pixel-level importance weighting and bidirectional consistency constraints to enhance robustness against clothing variations. Extensive experiments demonstrate that QA-ReID achieves state-of-the-art performance on multiple benchmarks, including PRCC, LTCC, and VC-Clothes, and significantly outperforms existing approaches under cross-clothing scenarios.

QA-ReID: Quality-Aware Query-Adaptive Convolution Leveraging Fused Global and Structural Cues for Clothes-Changing ReID

TL;DR

The paper tackles clothes-changing person re-identification by introducing QA-ReID, a dual-branch framework that fuses RGB-based global appearance with parsing-based structural cues through a multi-modal attention module. It advances matching with QAConv-QA, which applies pixel-level quality weighting and bidirectional constraints to focus on informative body regions while mitigating clothing variations. The method is trained with a joint loss combining identity, triplet, and pairwise matching objectives, and it achieves state-of-the-art results on CC-ReID benchmarks PRCC, LTCC, and VC-Clothes, demonstrating robustness to clothing changes. The approach offers practical impact for surveillance and related applications by improving cross-clothing re-identification performance while providing interpretable attention patterns that emphasize identity-related regions.

Abstract

Unlike conventional person re-identification (ReID), clothes-changing ReID (CC-ReID) presents severe challenges due to substantial appearance variations introduced by clothing changes. In this work, we propose the Quality-Aware Dual-Branch Matching (QA-ReID), which jointly leverages RGB-based features and parsing-based representations to model both global appearance and clothing-invariant structural cues. These heterogeneous features are adaptively fused through a multi-modal attention module. At the matching stage, we further design the Quality-Aware Query Adaptive Convolution (QAConv-QA), which incorporates pixel-level importance weighting and bidirectional consistency constraints to enhance robustness against clothing variations. Extensive experiments demonstrate that QA-ReID achieves state-of-the-art performance on multiple benchmarks, including PRCC, LTCC, and VC-Clothes, and significantly outperforms existing approaches under cross-clothing scenarios.
Paper Structure (11 sections, 7 equations, 3 figures, 3 tables)

This paper contains 11 sections, 7 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Feature Extraction and Fusion
  4. Quality-Aware Pixel-Level Matching
  5. Joint Loss Function
  6. Experiments
  7. Experimental Settings
  8. Comparison with State-of-the-Art Methods
  9. Ablation Study
  10. Visualization
  11. Conclusion

Figures (3)

  • Figure 1: (a) The conventional global matching. (b) Local matching using QAConv. (c) Our proposed QAConv-QA.
  • Figure 2: Illustration of our QA-ReID, which takes both RGB images and clothing-invariant representations obtained from human parsing as inputs. Then, a multi-modal attention fusion module is employed to combine the complementary features, and a quality-aware query adaptive convolution module (QAConv-QA) is introduced to compute similarity scores between images.
  • Figure 3: Visualization of attention maps on LTCC dataset, comparing our method with CAL.