Abstract:With the rapid development of intelligent transportation systems and the popularity of smart city infrastructure, Vehicle Re-ID technology has become an important research field. The vehicle Re-ID task faces an important challenge, which is the high similarity between different vehicles. Existing methods use additional detection or segmentation models to extract differentiated local features. However, these methods either rely on additional annotations or greatly increase the computational cost. Using attention mechanism to capture global and local features is crucial to solve the challenge of high similarity between classes in vehicle Re-ID tasks. In this paper, we propose LKA-ReID with large kernel attention. Specifically, the large kernel attention (LKA) utilizes the advantages of self-attention and also benefits from the advantages of convolution, which can extract the global and local features of the vehicle more comprehensively. We also introduce hybrid channel attention (HCA) combines channel attention with spatial information, so that the model can better focus on channels and feature regions, and ignore background and other disturbing information. Experiments on VeRi-776 dataset demonstrated the effectiveness of LKA-ReID, with mAP reaches 86.65% and Rank-1 reaches 98.03%.

What problem does this paper attempt to address?

This paper attempts to solve the high - similarity problem in the vehicle re - identification (Vehicle Re - ID) task. Specifically, there are highly similar appearance features (such as color, model, etc.) among different vehicles, which makes it a great challenge to track the same vehicle across cameras. Existing methods usually rely on additional detection or segmentation models to extract differential local features, but these methods either require additional annotation data or significantly increase the computational cost. To solve these problems, the authors proposed the LKA - ReID model and introduced the large - kernel attention mechanism (Large Kernel Attention, LKA) and the hybrid channel attention mechanism (Hybrid Channel Attention, HCA). The following are the main contributions of this paper: 1. **Proposing the LKA - ReID model**: By introducing the large - kernel attention mechanism (LKA), this model can capture long - distance dependencies, thus dealing with the challenges brought by vehicle appearance and view - point changes. 2. **Introducing the HCA mechanism**: Combining channel attention with spatial information enables the model to better focus on key feature areas and channels and ignore background and other interfering information. 3. **Experimental verification**: The experimental results on the VeRi - 776 dataset show that the LKA - ReID model has a competitive performance, with mAP reaching 86.65% and Rank - 1 accuracy reaching 98.03%. ### Formula summary - **Large - kernel attention mechanism (LKA)**: \[ F'=\text{Conv}_{1\times1}(\text{GELU}(F)) \] \[ \text{Attention}=\text{Conv}_{1\times1}(\text{DW - D - Conv}(\text{DW - Conv}(F'))) \] \[ \text{Output}=\text{Conv}_{1\times1}(\text{Attention}\otimes F') + F \] - **Hybrid channel attention mechanism (HCA)**: \[ k = \Phi(C)=\left\lceil\frac{\log_2 C}{\gamma}+b\right\rceil_{\text{odd}} \] where \(k\) is the size of the one - dimensional convolution kernel, \(C\) is the channel dimension, \(\gamma\) and \(b\) are hyper - parameters with default values of 2, and \(\left\lceil x\right\rceil_{\text{odd}}\) represents taking the odd number closest to \(x\). Through these innovations, the LKA - ReID model can significantly improve the performance of the vehicle re - identification task while maintaining a low computational complexity.

LKA-ReID:Vehicle Re-Identification with Large Kernel Attention

Efficient but Lightweight Network for Vehicle Re-Identification with Center-Constraint Loss

Multi-scale attention vehicle re-identification

Multi-View Spatial Attention Embedding for Vehicle Re-Identification

SCAN: Spatial and Channel Attention Network for Vehicle Re-Identification

A Dual Self-Attention Mechanism for Vehicle Re-Identification

Discriminative Feature Learning with Co-occurrence Attention Network for Vehicle ReID

Transformer-Based Attention Network for Vehicle Re-Identification

Attributes Guided Feature Learning for Vehicle Re-Identification

Identity-Guided Spatial Attention for Vehicle Re-Identification

Part-Regularized Near-Duplicate Vehicle Re-Identification.

Multi-attention deep reinforcement learning and re-ranking for vehicle re-identification

A Dual-Path Model With Adaptive Attention for Vehicle Re-Identification

Aggregating Global and Local Visual Representation for Vehicle Re-IDentification

A vehicle re-identification framework based on the improved multi-branch feature fusion network

Multi-axis interactive multidimensional attention network for vehicle re-identification

Vehicle re-identification based on dimensional decoupling strategy and non-local relations

VARID: Viewpoint-Aware Re-IDentification of Vehicle Based on Triplet Loss

V2ReID: Vision-Outlooker-Based Vehicle Re-Identification

Coarse-to-fine: A RNN-based hierarchical attention model for vehicle re-identification

TANet: Text region attention learning for vehicle re-identification