Abstract:Domain generalizable person re-identification (DG re-ID) aims to learn discriminative representations that are robust to distributional shifts. While data augmentation is a straightforward solution to improve generalization, certain augmentations exhibit a polarized effect in this task, enhancing in-distribution performance while deteriorating out-of-distribution performance. In this paper, we investigate this phenomenon and reveal that it leads to sparse representation spaces with reduced uniformity. To address this issue, we propose a novel framework, Balancing Alignment and Uniformity (BAU), which effectively mitigates this effect by maintaining a balance between alignment and uniformity. Specifically, BAU incorporates alignment and uniformity losses applied to both original and augmented images and integrates a weighting strategy to assess the reliability of augmented samples, further improving the alignment loss. Additionally, we introduce a domain-specific uniformity loss that promotes uniformity within each source domain, thereby enhancing the learning of domain-invariant features. Extensive experimental results demonstrate that BAU effectively exploits the advantages of data augmentation, which previous studies could not fully utilize, and achieves state-of-the-art performance without requiring complex training procedures. The code is available at \url{<a class="link-external link-https" href="https://github.com/yoonkicho/BAU" rel="external noopener nofollow">this https URL</a>}.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve an important problem in **Domain Generalizable Person Re - Identification (DG re - ID)**, that is, how to achieve robust feature representation between datasets with different distributions. Specifically, the paper focuses on the **polarization effect** exhibited by data augmentation techniques in the DG re - ID task: although some data augmentation methods can improve the performance of the in - distribution (source domain), they will damage the performance of the out - of - distribution (target domain). #### Reasons for the Polarization Effect The paper points out that this polarization effect will cause the representation space to become sparse and non - uniform, thus affecting the generalization ability of the model to unseen domains (OOD). The specific manifestations are as follows: - **Alignment**: The similarity between enhanced samples increases, making samples of the same category closer. - **Uniformity**: The distribution of enhanced samples becomes more concentrated, resulting in a reduced distinction between different categories. #### Solutions To solve this problem, the authors propose a new framework - **Balancing Alignment and Uniformity (BAU)**. This framework alleviates the polarization effect of data augmentation in the following ways: 1. **Alignment Loss and Uniformity Loss**: Applied to both the original image and the enhanced image simultaneously to maintain the alignment and uniformity of the feature representation. 2. **Weighting Strategy**: Introduce a reliability - based weighting strategy to improve the alignment loss and ensure the quality of the enhanced samples. 3. **Domain - specific Uniformity Loss**: Promote the uniformity within each source domain, thereby enhancing the learning of domain - invariant features. Through these methods, BAU effectively utilizes the advantages of data augmentation, improves the generalization ability of the model in different domains, and achieves state - of - the - art performance in multiple benchmark tests. ### Formula Summary The key formulas involved in the paper are as follows: - **Alignment Loss**: \[ L_{\text{align}}=\frac{1}{|I_{\text{pos}}|} \sum_{(i, j) \in I_{\text{pos}}}\|\tilde{f}_i - f_j\|^2_2 \] where \(I_{\text{pos}}=\{(i, j)|y_i = y_j\}\) is the index set of positive sample pairs. - **Uniformity Loss**: \[ L_{\text{uniform}}=\log \left(\frac{1}{|I_{\text{data}}|} \sum_{(i, j) \in I_{\text{data}}} e^{-2\|f_i - f_j\|^2_2}\right)+\log \left(\frac{1}{|I_{\text{data}}|} \sum_{(i, j) \in I_{\text{data}}} e^{-2\|\tilde{f}_i-\tilde{f}_j\|^2_2}\right) \] where \(I_{\text{data}}=\{(i, j)|i\neq j\}\) is the index set of all different sample pairs. - **Domain - specific Uniformity Loss**: \[ L_{\text{domain}}=\log \left(\frac{\sum_i \sum_{j \in N(f_i)} e^{-2\|f_i - c_j\|^2_2}}{\sum_i N!}\right)+

Generalizable Person Re-identification via Balancing Alignment and Uniformity

Adaptive Domain-Specific Normalization for Generalizable Person Re-Identification

Debiased Batch Normalization Via Gaussian Process for Generalizable Person Re-identification

Meta Distribution Alignment for Generalizable Person Re-Identification

Calibrated Feature Decomposition for Generalizable Person Re-Identification

Generalizable Person Re-identification Without Demographics

Group-aware Label Transfer for Domain Adaptive Person Re-identification

Exert Diversity and Mitigate Bias: Domain Generalizable Person Re-identification with a Comprehensive Benchmark

Semi-Supervised Domain Generalizable Person Re-Identification

Invariance Learning Under Uncertainty for Single Domain Generalization Person Re-Identification

A Unified Deep Semantic Expansion Framework for Domain-Generalized Person Re-identification

Rethinking Normalization Layers for Domain Generalizable Person Re-identification

Learning Disentangled Representation for Robust Person Re-identification

Domain generalized federated learning for Person Re-identification

TAL: Two-stream Adaptive Learning for Generalizable Person Re-identification

A New Date-Balanced Method Based on Adaptive Asymmetric and Diversity Regularization in Person Re-Identification.

Person30K: A Dual-Meta Generalization Network for Person Re-Identification

A domain generalized person re-identification algorithm based on meta-bond domain alignment ☆

Towards Discriminative Representation Learning for Unsupervised Person Re-identification

Learning to Adapt Across Dual Discrepancy for Cross-Domain Person Re-Identification

Domain-adaptive person re-identification via domain alignment and mutual pseudo-label refinement