Differentiable Channel Selection in Self-Attention For Person Re-Identification

2025-05-13

AI Summary

To address channel redundancy and insufficient discriminability of self-attention mechanisms in person re-identification (Re-ID), this paper proposes the Differentiable Channel Selection Attention (DCS-Attention) module, which dynamically selects the most informative feature channels during attention weight computation. Grounded in the Information Bottleneck (IB) principle, we design a differentiable variational upper-bound loss, enabling the first end-to-end differentiable channel selection under IB constraints. We further introduce two complementary paradigms: DCS-FB, which adapts DCS-Attention to fixed backbone networks, and DCS-DNAS, which jointly optimizes the attention module and backbone via differentiable neural architecture search. To our knowledge, this work establishes the first IB-principled differentiable variational upper-bound loss function for Re-ID. Extensive experiments on Market-1501 and DukeMTMC-reID demonstrate state-of-the-art performance. The code is publicly available, validating the effectiveness and generalizability of the proposed method in learning identity-discriminative features.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Dimensionality Reduction/Feature Selection
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

In this paper, we propose a novel attention module termed the Differentiable Channel Selection Attention module, or the DCS-Attention module. In contrast with conventional self-attention, the DCS-Attention module features selection of informative channels in the computation of the attention weights. The selection of the feature channels is performed in a differentiable manner, enabling seamless integration with DNN training. Our DCS-Attention is compatible with either fixed neural network backbones or learnable backbones with Differentiable Neural Architecture Search (DNAS), leading to DCS with Fixed Backbone (DCS-FB) and DCS-DNAS, respectively. Importantly, our DCS-Attention is motivated by the principle of Information Bottleneck (IB), and a novel variational upper bound for the IB loss, which can be optimized by SGD, is derived and incorporated into the training loss of the networks with the DCS-Attention modules. In this manner, a neural network with DCS-Attention modules is capable of selecting the most informative channels for feature extraction so that it enjoys state-of-the-art performance for the Re-ID task. Extensive experiments on multiple person Re-ID benchmarks using both DCS-FB and DCS-DNAS show that DCS-Attention significantly enhances the prediction accuracy of DNNs for person Re-ID, which demonstrates the effectiveness of DCS-Attention in learning discriminative features critical to identifying person identities. The code of our work is available at https://github.com/Statistical-Deep-Learning/DCS-Attention.

Problem

• Selects informative channels in self-attention for Re-ID
• Integrates differentiable channel selection with DNN training
• Improves feature extraction for person Re-ID accuracy

Innovation

• Differentiable Channel Selection in self-attention
• Integration with DNN training via differentiable selection
• Optimized with Information Bottleneck variational bound