A Dual Self-Attention mechanism for vehicle re-Identification. (May 2023)

• Record Type: Journal Article
• Authors: Zhu, Wenqian; Wang, Zhongyuan; Wang, Xiaochen; Hu, Ruimin; Liu, Huikai; Liu, Cheng; Wang, Chao; Li, Dengshi
• Is Part Of: Pattern recognition. Volume 137(2023)
• Publication Date: 2023-05
• Subjects: Cross-region attention -- Dual self-attention -- Multi-attention network -- Vehicle re-identification -- Feature embedding; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2022.109258
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: A novel multi-attention network simulating the visual attention of humans is designed to realize highly efficient alignment and feature embedding globally and locally for vehicle re-ID, improving the performance of re-ID both contextually and spatially. We propose a dual self-attention mechanism where the static self-attention and the dynamicself-attention (named cross-region attention) are designed to capture long-range dependenciesglobally and the position-related-range dependencies locally respectively, so as to enhance thespatial/positional awareness of our model. We conduct extensive experiments to validate the effectiveness of our model and show that the proposed method achieves the best experimental results on most metrics compared withmany other state-of-the-art methods, e.g., Rank-1 accuracy of 80.6%, 78.1%, 75.0% on three testing subsets of VehicleID and 92.9%, 91.8%, 90.4% on those of Vehicle-1M. Our model also achieves 76.3% mAP and 94.8% on Veri776. Abstract: Vehicle re-identification has attracted tremendous attention from computer vision communities for its extensive applications in intelligent transportation and public security, while the high inter-class similarity and the large intra-class difference between vehicles bring out great challenges for re-identification (re-ID). To tackle these challenges, we learn from the self-attention mechanism in Natural Language Processing and propose a dual self-attention module to learn different regional dependencies: static self-attention for selectively refining semantic features and dynamic self-attention (called cross-region attention) for enhancing the spatial awareness of local feature. The static self-attention refines attended pixels within the entire image and salient regions, while the cross-region attention creatively captures the position-related regional dependencies for pixels within the windscreen area. These attention modules capture long-range dependencies and relative position information between different pixels or regions for vehicle feature learning globally and locally, realizing an efficient vehicle feature embedding by concatenating these augmented features for vehicle re-ID. Extensive experiments demonstrate the effectiveness and promising performance of our approach against the state-of-the-arts.