Partition and Reunion: A Viewpoint-Aware Loss for Vehicle Re-Identification

Chen, Haobo; Liu, Yang; Huang, Yang; Ke, Wei; Sheng, Hao

doi:10.1109/icip46576.2022.9897541

Cited by 8 publications

(10 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Meantime, the extraction of local information representing different features from different parts of the vehicle using a rigid division of predefined parts has been shown in many works [8][9][10][11][12][13]23] to be an effective method to improve the Re-ID performance. MGN [13] divides the feature map into multiple stripes along the horizontal direction.…”

Section: Vehicle Re-idmentioning

confidence: 99%

“…This is not only labor-intensive and time-consuming for annotation, but also increases the computational complexity of the vehicle Re-ID task. To address the problems with the above methods, some other methods [8][9][10][11][12][13] focus on mining fine-grained cues in local regions using rigid divisions of predefined parts. HSKT [8] uses a rigid segmentation strategy to segment the vehicle image into multiple parts, and then directly uses these parts to extract local features.…”

Section: Introductionmentioning

confidence: 99%

“…HSKT [8] uses a rigid segmentation strategy to segment the vehicle image into multiple parts, and then directly uses these parts to extract local features. Chen et al [9] divides the feature maps of vehicle images along different directions to extract rich fine-grained local features. Although these methods are simple and effective, the information captured between different parts can interfere with each other, so we propose a dimensional decoupling strategy that allows different parts to focus only on themselves, thus improving the characterization of features.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

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

Li,

Pang,

Meng

2024

PLoS ONE

View full text Add to dashboard Cite

Vehicle re-identification (Re-ID) is a challenging task that aims to recognize the same vehicle across different non-overlapping cameras. Existing attention mechanism-based methods for vehicle Re-ID often suffer from significant intra-class variation and inter-class variation due to various factors such as illumination, occlusion, viewpoint, etc. In this paper, we propose a novel network architecture for vehicle Re-ID, named Dimensional Decoupling Strategy and Non-local Relationship Network (DMNR-Net), which uses three modules to extract complementary features: global feature extraction module, non-local relationship capture module(NRCM), and dimensional decoupling module (DDS). The global feature extraction module captures complete and coarse-grained features from the whole image; the NRCM module extracts saliency information from feature maps in both spatial and channel dimensions; and the DDS decouples spatial and channel features into two branches to extract fine-grained features and focus on specific subspaces. We conduct extensive experiments on two popular publicly datasets, VeRi-776 and VehicleID, to evaluate the effectiveness of our method. The experimental results show that our DMNR-Net outperforms state-of-the-art methods by a large margin on both datasets.

show abstract

Section: Vehicle Re-idmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

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

Li,

Pang,

Meng

2024

PLoS ONE

View full text Add to dashboard Cite

show abstract

“…(2) We reconstruct the graph by introducing a learnable token node to learn spatial structure information and global context information simultaneously, while avoiding over-smoothing. (3) In each branch, we jointly learn the triplet loss and crossentropy loss.…”

Section: Introductionmentioning

confidence: 99%

SSR-Net: A Spatial Structural Relation Network for Vehicle Re-identification

Wei

Lang

et al. 2023

ACM Trans. Multimedia Comput. Commun. Appl.

View full text Add to dashboard Cite

Vehicle re-identification (Re-ID) represents the task aiming to identify the same vehicle from images captured by different cameras. Recent years have seen various feature learning based approaches merely focusing on feature representations including global features or local features to obtain more subtle details to identify highly similar vehicles. However, few such methods consider the spatial geometrical structure relationship among local regions or between the global and local regions. By contrast, in this study, we propose a Spatial Structural Relation Network (SSR-Net) which explores the above-mentioned two kinds of relations simultaneously to learn more discriminative features by modeling the spatial structure information and global context information. In this paper we propose to adopt a Graph Convolution Network (GCN), for modeling spatial structural relationships among characteristic features. The GCN model aggregating the local and global features is shown to be more discriminative and robust to several car image transformations. To improve the performance of our proposed network, we jointly combine the classification loss with metric learning loss. Extensive experiments conducted on the public VehicleID and VeRi-776 datasets validate the effectiveness of our approach in comparison with recent works.

show abstract

“…The original intention of vehicle re-identification is to better solve traffic supervision and criminal investigation problems. It has developed from the early sensor-based method [2][3][4], the artificial feature-based method [5][6], to the current method mainly based on deep learning [7][8][9][10]. The focus of vehicle re-identification lies in feature extraction and comparison.…”

Section: Introductionmentioning

confidence: 99%

Research on Vehicle Re-identification for Vehicle Road Collaboration

Sun

Wang²,

Deng³

et al. 2023

J. Phys.: Conf. Ser.

View full text Add to dashboard Cite

Vehicles and roads cooperate to perceive traffic targets, which can reduce the perception blind spots of vehicles and improve driving safety. In this paper, we proposes a vehicle re-identification method oriented to vehicle-road coordination. This method first designs a lightweight vehicle re-identification network based on ShufflenetV2 to solve the computational efficiency problem of vehicle-road coordination scenarios, which can efficiently complete vehicle feature extraction; then, due to the real-time requirements of scenario communication, an adaptive feature conversion mechanism is designed in combination with the LSH algorithm, which can make the re-identification module to dynamically perform binary bit feature conversion and adjust the dimension according to the communication channel state; finally, a loss function for the conversion of vehicle re-identification features is designed, which can greatly reduce the accuracy loss rate of converting floating-point features to bit features. Experiments show that our method can efficiently complete the information extraction and comparison of vehicle re-identification features in the vehicle-road coordination scenario, and can improve the perception efficiency of vehicle-road coordination while taking into account performance and bandwidth.

show abstract

Partition and Reunion: A Viewpoint-Aware Loss for Vehicle Re-Identification

Cited by 8 publications

References 22 publications

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

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

SSR-Net: A Spatial Structural Relation Network for Vehicle Re-identification

Research on Vehicle Re-identification for Vehicle Road Collaboration

Contact Info

Product

Resources

About