Domain Adaptive Attention Learning for Unsupervised Person Re-Identification

TL;DR

Cross-dataset person Re-ID suffers from large domain shifts and unlabeled target data. The authors propose a Domain Adaptive Attention Module (DAAM) that splits feature maps into domain-shared and domain-specific components with Fx^sh = A(Fx) ⊗ Fx and Fx^sp = (1 − A(Fx)) ⊗ Fx, enabling transfer-friendly representations. A two-branch architecture learns DS shared and DSP features, guided by a soft-label cross-entropy loss that leverages clustering-based pseudo labels without overfitting. Empirical results on Market-1501, DukeMTMC-reID and MSMT17 demonstrate state-of-the-art performance without target annotations, highlighting robust cross-domain transfer for scalable surveillance applications.

Abstract

Person re-identification (Re-ID) across multiple datasets is a challenging task due to two main reasons: the presence of large cross-dataset distinctions and the absence of annotated target instances. To address these two issues, this paper proposes a domain adaptive attention learning approach to reliably transfer discriminative representation from the labeled source domain to the unlabeled target domain. In this approach, a domain adaptive attention model is learned to separate the feature map into domain-shared part and domain-specific part. In this manner, the domain-shared part is used to capture transferable cues that can compensate cross-dataset distinctions and give positive contributions to the target task, while the domain-specific part aims to model the noisy information to avoid the negative transfer caused by domain diversity. A soft label loss is further employed to take full use of unlabeled target data by estimating pseudo labels. Extensive experiments on the Market-1501, DukeMTMC-reID and MSMT17 benchmarks demonstrate the proposed approach outperforms the state-of-the-arts.

Figures (6)

• Figure 1: The visualizations of the proposed DAAM. The DAAM (the warmer color means the greater weight) separates the original image (a) into two parts: the domain-shared part (b) and the domain-specific part (c). The former is discriminative and useful to the person Re-ID task, and the latter is used to model the domain-specific information (such as background) caused by the domain divergence.
• Figure 2: The framework of the proposed method. The domain adaptive attention module separates the feature map into domain-shared (DSH) and domain-specific (DSP) feature maps simultaneously. Then, a DSH branch and a DSP branch are introduced to learn these two feature maps respectively. The soft-label based cross entropy and hard-label based losses are adopted to the DSH branch to make the DSH feature discriminative to different persons and transferable to different domains. In contrary, the domain-specific loss is employed to make the DSP branch capture the domain distinguishable information to avoid to distract the DSH feature and separate the untransferable to different domains.
• Figure 3: Diagram of domain adaptive attention module.
• Figure 4: Evaluation with different $Iteration$.
• Figure 5: Evaluation with different percentage $p$.