用于行人重识别的细粒度网络

doi:10.15966/j.cnki.dnydx.2024.03.019

全文: PDF( KB)
输出: BibTeX | EndNote (RIS)

摘要针对不同类型的注意机制堆叠在行人重识别任务中识别效率过低，且过于侧重提取单一粒度的信息导致很难扩展到真实世界的应用场景等问题，提出了一种基于局部注意力的细粒度模块，将获取的粗粒度特征表示分割成多个部分，引导网络去提取更细粒度的局部线索。此外，考虑到单一分支在提取多粒度线索方面的不足，设计了一个多分支的细粒度网络，以进一步提取更丰富的线索。在主流数据集上的实验结果验证了文中所提网络的有效性。

	服务
	把本文推荐给朋友
	加入引用管理器
	E-mail Alert
	RSS
	（）
	作者相关文章

关键词 ：行人重识别, 注意力机制, 细粒度线索

Abstract：To solve the problem that the recognition efficiency of different types of attention mechanisms is too low in the person reidentification task, and the focus is too much on extracting single-granularity information, which makes it difficult to extend to realworld application scenarios, a fine-grained module based on local attention is proposed. The obtained coarse-grained feature repre‐ sentation is divided into multiple parts, guiding the network to extract finer-grained local clues. In addition, considering the short‐ comings of a single branch in extracting multi-granularity clues, a multi-branch fine-grained network is designed to further extract richer clues. Experimental results on mainstream datasets verify the effectiveness of the network proposed in this article

年卷期日期: 2024-05-28 出版日期: 2024-07-02

引用本文:

吴数立. 用于行人重识别的细粒度网络[J]. 电脑与电信, 2024, 1(3): 45-.
WU Shu-li. A Fine-grained Network for Person Re-identification. Computer & Telecommunication, 2024, 1(3): 45-.

链接本文:

https://www.computertelecom.com.cn/CN/10.15966/j.cnki.dnydx.2024.03.019 或 https://www.computertelecom.com.cn/CN/Y2024/V1/I3/45