适于行人重识别的二分支EfficientNet网络设计  

Design of A Two-Branch EfficientNet for Person Re-Identification

作　　者：尹梓睿 张索非[2] 张磊[1] 吴晓富[1] Yin Zirui;Zhang Suofei;Zhang Lei;Wu Xiaofu(College of Communication and Information Engineering,Nanjing University of Posts and Telecommunication,Nanjing,Jiangsu 210003,China;Internet of Things College,Nanjing University of Posts and Telecommunication,Nanjing,Jiangsu 210003,China)

机构地区：[1]南京邮电大学通信与信息工程学院,江苏南京210003 [2]南京邮电大学物联网学院,江苏南京210003

出　　处：《信号处理》2020年第9期1481-1488,共8页Journal of Signal Processing

基　　金：国家自然科学基金(61372123,61701252)。

摘　　要：鉴于ResNet的强大表达能力,其在行人重识别领域获得了广泛的应用。虽然基于ResNet50构建的行人重识别网络取得了优异的性能,但流行的ResNet50仍存在模型体积大、效率低等局限性。与之相比,EfficientNet作为一种新兴的深度模型,具有设计合理、运行高效等特点,并在ImageNet数据集上有着更出色的性能表现。为此,本文尝试将EfficientNet系列网络引入到行人重识别领域,替代比较流行的ResNet50主干网络,提供了一个全新的骨干网基线。本文重点根据EfficientNet系列网络给出一种二分支行人重识别网络构造。相比于ResNet50,基于EfficientNet构造的二分支行人重识别网络具有网络参数规模小、性能提升明显的特点。实验结果表明:所构造的网络在行人重识别流行数据集上均有良好的表现。Due to its strong expressive ability,ResNet is widely used in Person Re-Identification.Although the use of ResNet50 for person ReID has achieved excellent performance,the popular ResNet-50-based solution still has the limitations of large model size and low efficiency in implementations.Recently,EfficientNet was proposed as an emerging alternative for design of deep neural models.Compared with ResNet-50,EfficientNet,as an emerging depth model,has the characteristics of reasonable design,running efficiently and so on,and has more excellent performance in ImageNet data set.In this paper,we try to introduce the EfficientNet series network into the area of Person Re-Identification,replace the more popular ResNet50 trunk network,and provide an entirely new baseline network.Then,we propose to design a two-branch EfficientNet for Person Re-Identification.Compared with various ResNet-50-based solutions,the proposed EfficientNet has a significantly small size on network,but often achieves better performance.Experimental results show that the proposed network performs very well on the popular Person Re-Identification data set.

关 键 词：行人重识别 特征提取 效率网络 多分支结构 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]