中国科学院机构知识库网格
Chinese Academy of Sciences Institutional Repositories Grid
Deep Spatial Feature Reconstruction for Partial Person Reidentification: Alignment-free Approach

文献类型:会议论文

作者He LX(何凌霄); Liang J(梁坚); Li HQ(李海青); Sun ZN(孙哲南)
出版日期2018
会议日期6.18-6.21
会议地点美国盐湖城
英文摘要

Partial person re-identification (re-id) is a challenging problem, where only several partial observations (images) of people are available for matching. However, few studies have provided flexible solutions to identifying a person in an image containing arbitrary part of the body. In this paper, we propose a fast and accurate matching method to address this problem. The proposed method leverages Fully
Convolutional Network (FCN) to generate fix-sized spatial feature maps such that pixel-level features are consistent. To match a pair of person images of different sizes, a novel method called Deep Spatial feature Reconstruction (DSR) is further developed to avoid explicit alignment. Specifically, DSR exploits the reconstructing error from popular dictionary learning models to calculate the similarity between different spatial feature maps. In that way, we expect that the proposed FCN can decrease the similarity of coupled images from different persons and increase that from the same person. Experimental results on two partial person datasets demonstrate the efficiency and effectiveness of the proposed method in comparison with several state-ofthe-art partial person re-id approaches. Additionally, DSR achieves competitive results on a benchmark person dataset
Market1501 with 83.58% Rank-1 accuracy.

源URL[http://ir.ia.ac.cn/handle/173211/23698]  
专题自动化研究所_智能感知与计算研究中心
作者单位中科院自动化研究所
推荐引用方式
GB/T 7714
He LX,Liang J,Li HQ,et al. Deep Spatial Feature Reconstruction for Partial Person Reidentification: Alignment-free Approach[C]. 见:. 美国盐湖城. 6.18-6.21.

入库方式: OAI收割

来源:自动化研究所

浏览0
下载0
收藏0
其他版本

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。