行人重识别(REID)
文章平均质量分 57
行人重识别是博主的研究课题,所以会分享较多行人重识别经典论文和较新论文的阅读感悟。希望对于该课题的初学者有所帮助,也希望这里可以成为和同样对此课题感兴趣的大佬们交流的平台。感谢大家支持!
嘻嘻哈哈yjy
多多指教
展开
-
In-depth exploration of attribute information for person re-identification
行人重识别之属性信息In-depth exploration of attribute information for person re-identification这次分享一篇自己的文章,这篇文章针对了行人的属性信息。首先,我定量分析了属性信息的重要性。假设,我们已知某一类的属性标签,用该标签直接辅助baseline进行识别。如下图,性能都会有不小的提高。进一步,我使用所有类别的属性(图中的sum)对baseline进行辅助。可以看出,模型的精度大幅提升。这证明了属性信息有很大的价值。原创 2020-07-02 10:52:31 · 618 阅读 · 2 评论 -
Instance-Guided Context Rendering for Cross-Domain Person Re-Identification
行人重识别之cross domainInstance-Guided Context Rendering for Cross-Domain Person Re-Identification (ICCV2019)原文链接最近写了很多cross domain的文章,打算用这篇文章来收尾。没想到这篇文章给我带来了十足的惊喜。文章基于GAN转换图像风格,效果非常好。如下图所示(忽略baseline),...原创 2019-12-17 23:09:55 · 1186 阅读 · 2 评论 -
Cross-Dataset Person Re-Identification via Unsupervised Pose Disentanglement and Adaptation
行人重识别之cross domainCross-Dataset Person Re-Identification via Unsupervised Pose Disentanglement and Adaptation (ICCV2019)原文链接这篇文章从图像风格迁移的角度进行了cross domain,用到了生成对抗网络(GAN),对GAN不熟悉的朋友可以参考这篇文章,让你两分钟对GAN...原创 2019-12-16 18:27:43 · 1324 阅读 · 5 评论 -
Asymmetric Co-Teaching for Unsupervised Cross-Domain Person Re-Identification
行人重识别之cross domainAsymmetric Co-Teaching for Unsupervised Cross-Domain Person Re-Identification (AAAI 2020)原文链接代码连接这篇文章为cross domain问题提供了一个新思路:在使用聚类算法给target数据库打伪标签的时候,会存在一些离群点(outliers),即这些图像特征不是...原创 2019-12-15 15:12:33 · 1674 阅读 · 0 评论 -
EANet: Enhancing Alignment for Cross-Domain Person Re-identification
行人重识别之cross domainEANet: Enhancing Alignment for Cross-Domain Person Re-identification (2018arXiv)原文链接这篇文章从alignment(对准)的角度解决cross domain问题,最后的模型在source到source和source到target的效果均有了明显的提升。本文的算法思路比较简单清...原创 2019-12-14 22:32:20 · 917 阅读 · 1 评论 -
Transferable Joint Attribute-Identity Deep Learning for Unsupervised Person Re-Identification
行人重识别之cross domainTransferable Joint Attribute-Identity Deep Learning for Unsupervised Person Re-Identification (CVPR2018)原文链接一篇关于cross domain的文章,发现这类文章有一个相似的规律,即学习行人更具本质的特征。这样在cross domain的时候,这些特征...原创 2019-12-11 16:50:28 · 1130 阅读 · 0 评论 -
Self-similarity Grouping: A Simple Unsupervised Cross Domain Adaptation Approach for Person Re-id
行人重识别之cross domainSelf-similarity Grouping: A Simple Unsupervised Cross Domain Adaptation Approach for Person Re-identification (ICCV2019)原文链接代码链接核心思想:两个数据库之间,图像风格迥异。如果将图像进行分割,那么每一块之间的风格差异或许会小一点。这...原创 2019-12-10 14:27:39 · 2589 阅读 · 5 评论 -
Self-Training With Progressive Augmentation for Unsupervised Cross-Domain Person Re-Identification
行人重识别之cross domainSelf-Training With Progressive Augmentation for Unsupervised Cross-Domain Person Re-Identification (ICCV2019)原文链接代码连接cross domain:很多reid的算法都是针对用同一个数据库训练和测试,这在现实生活中意义不大。所以,如何使得在A...原创 2019-12-09 21:51:16 · 738 阅读 · 1 评论 -
Camera viewpoint learning for person re-identification
行人重识别之摄像头视角Camera viewpoint learning for person re-identification这次分享一篇自己的文章,这篇文章针对了摄像头视角的问题。先来说一个行人重识别中本质而又容易被忽略的问题:行人重识别是跨摄像头检索任务,也就是说,找到同一个摄像头中的同一个人是没有意义的。最终计算准确率的时候,会把rank序列中和query摄像头视角相同且是同一个...原创 2019-12-07 21:37:52 · 628 阅读 · 6 评论 -
Interaction-and-Aggregation Network for Person Re-identification
行人重识别之网络设计Interaction-and-Aggregation Network for Person Re-identification原文链接许久没有看过在网络结构上进行设计的行人重识别文章,这篇文章针对行人重识别图像空间位置不匹配的问题设计了SIA模块,为了提取图像中的细节特征设计了CIA模块,两者结合成为IA模块,在深度网络中即插即用,几乎不会增加计算量,实验结果较好。S...原创 2019-08-25 13:18:38 · 2030 阅读 · 4 评论 -
Pyramidal Person Re-IDentification via Multi-Loss Dynamic Training
行人重识别之动态分块训练Pyramidal Person Re-IDentification via Multi-Loss Dynamic Training原文链接这篇文章提出了两个思想:图像局部和整体的信息都很重要,要联合使用。在训练的过程中,损失函数的权重应该是动态变化的。这两个思想的关联性不大,我们分开来看。图像的分块和信息的利用:对于行人重识别中图像的分块重要性我们已经...原创 2019-08-24 14:22:08 · 2155 阅读 · 8 评论 -
Adaptive Transfer Network for Cross-Domain Person Re-Identification
行人重识别之多维度数据库风格迁移Adaptive Transfer Network for Cross-Domain Person Re-Identification原文链接之前研读的数据库风格迁移方面的文章,直接使用cyclegan对风格进行迁移,外加一些损失函数进行约束。这篇文章对数据库之间差异的原因进行分析并分为三类:光照差异、摄像头角度差异、分辨率差异。在此基础上,利用三者进行风格迁...原创 2019-08-23 16:27:38 · 2501 阅读 · 0 评论 -
AANet: Attribute Attention Network for Person Re-Identifications
行人重识别之多信息融合AANet: Attribute Attention Network for Person Re-Identifications原文链接:http://openaccess.thecvf.com/content_CVPR_2019/papers/Tay_AANet_Attribute_Attention_Network_for_Person_Re-Identificati...原创 2019-08-19 11:57:45 · 1451 阅读 · 0 评论 -
Patch-based Discriminative Feature Learning for Unsupervised Person Re-identification
行人重识别之分块无监督学习Patch-based Discriminative Feature Learning for UnsupervisedPerson Re-identification原文链接:http://openaccess.thecvf.com/content_CVPR_2019/papers/Yang_Patch-Based_Discriminative_Feature_L...原创 2019-08-09 11:57:07 · 2131 阅读 · 8 评论 -
Joint Discriminative and Generative Learning for Person Re-identification
行人重识别之端到端的生成与识别Joint Discriminative and Generative Learning for Person Re-identification原文链接:http://openaccess.thecvf.com/content_CVPR_2019/papers/Zheng_Joint_Discriminative_and_Generative_Learning_...原创 2019-08-08 12:47:58 · 1122 阅读 · 3 评论 -
Towards Rich Feature Discovery with Class Activation Maps Augmentation for Person Re-Identification
行人重识别之注意力机制Towards Rich Feature Discovery with Class Activation Maps Augmentation forPerson Re-Identification原文链接:http://openaccess.thecvf.com/content_CVPR_2019/papers/Yang_Towards_Rich_Feature_Dis...原创 2019-08-07 12:17:29 · 1441 阅读 · 5 评论 -
Distilled Person Re-identification: Towards a More Scalable System
行人重识别之泛化能力Distilled Person Re-identification: Towards a More Scalable System原文链接:http://openaccess.thecvf.com/content_CVPR_2019/papers/Wu_Distilled_Person_Re-Identification_Towards_a_More_Scalable_S...原创 2019-08-06 13:02:41 · 995 阅读 · 3 评论 -
Densely Semantically Aligned Person Re-Identification
行人重识别之语义分割网络Densely Semantically Aligned Person Re-Identification原文链接:http://openaccess.thecvf.com/content_CVPR_2019/papers/Zhang_Densely_Semantically_Aligned_Person_Re-Identification_CVPR_2019_pape...原创 2019-08-02 12:07:41 · 1117 阅读 · 2 评论 -
Learning to Reduce Dual-level Discrepancy for Infrared-Visible Person Re-identification
行人重识别之红外图像识别(CVPR2019)Learning to Reduce Dual-level Discrepancy for Infrared-Visible Person Re-identification原文链接:http://openaccess.thecvf.com/content_CVPR_2019/papers/Wang_Learning_to_Reduce_Dual-L...原创 2019-08-01 12:33:22 · 1349 阅读 · 3 评论 -
行人重识别中的domain adaptation
题外话:最近投了一篇PR,行人重识别方向,主要研究摄像头视角变化对Re-ID的影响。在改论文的过程中,编辑推荐了两篇相关的文章,也是研究类似的问题。不得不说的是,在修改参考文献格式的时候,用到了Mendeley。将期刊参考文献的模板数据下载并导入后,该软件可以自动修改参考文献的格式。一些知名期刊甚至可以使用Mendeley直接导入相应格式数据。很巧的是,在收到编辑推荐的两篇论文后,Mendele...原创 2019-06-19 15:33:07 · 2243 阅读 · 3 评论 -
Invariance Matters: Exemplar Memory for Domain Adaptive Person Re-identification
行人重识别之监督学习与无监督学习结合(CVPR2019)Invariance Matters: Exemplar Memory for Domain Adaptive Person Re-identification原文链接:https://arxiv.org/abs/1904.01990这篇文章有两个贡献:将监督学习和无监督学习结合引入了存储器的概念,比较新颖上图是网络结构,有...原创 2019-04-09 16:15:05 · 3451 阅读 · 0 评论 -
Perceive Where to Focus: Learning Visibility-aware Part-level Features for Partial Person Re-id
行人重识别之局部识别(CVPR2019)Perceive Where to Focus: Learning Visibility-aware Part-level Featuresfor Partial Person Re-identification原文链接:https://arxiv.org/abs/1904.00537这篇文章的核心思想是:如下图(a),两幅图的空间位置失配。如图(...原创 2019-04-08 16:02:46 · 4048 阅读 · 2 评论 -
Re-ranking Person Re-identification with k-reciprocal Encoding
行人重识别之rerankRe-ranking Person Re-identification with k-reciprocal Encoding原文链接:http://arxiv.org/pdf/1701.08398这篇文章介绍了一种rerank方法,引用量较高,说明方法不错,一起来看一下算法的思想。如上图所示,绿色的是识别正确的。作者阐述了如下规律:如果被正确识别出的图片作为pro...原创 2019-04-02 12:10:11 · 1844 阅读 · 2 评论 -
Beyond Part Models: Person Retrieval with Refined Part Pooling (and A Strong Convolutional Baseline)
行人重识别之图像软分割Beyond Part Models: Person Retrieval with Refined Part Pooling(and A Strong Convolutional Baseline)原文链接:http://arxiv.org/pdf/1711.09349一篇从分割图像出发的行人重识别文章,主要贡献如下:将图像的软分割(利用神经网络的attention...原创 2019-04-01 15:13:42 · 1640 阅读 · 4 评论 -
AlignedReID: Surpassing Human-Level Performance in Person Re-Identification
行人重识别之局部特征AlignedReID: Surpassing Human-Level Performance in Person Re-Identification原文链接:https://arxiv.org/pdf/1711.08184.pdf本文的标题很唬人,其实叠加了很多的tricks。除了本文最重要的创新点局部特征联合之外,还利用了Mutual Learning、rerank、...原创 2019-03-28 14:34:31 · 1905 阅读 · 2 评论 -
SVDNet for Pedestrian Retrieval
行人重识别之特征向量正交化SVDNet for Pedestrian Retrieval原文链接:https://arxiv.org/pdf/1703.05693.pdf这篇文章的核心思想如下:虽然特征向量的维度不小(一般为2048),但是网络经过训练后,特征向量往往是高度相关的,降低了特征向量的表达能力,进而降低了REID的性能。所以,如果能够对特征向量进行去相关(正交化),提升特征向量...原创 2019-03-27 12:58:26 · 1094 阅读 · 1 评论 -
Improving Person Re-identification by Attribute and Identity Learning
行人重识别之行人属性Improving Person Re-identification by Attribute and Identity Learning原文链接:https://arxiv.org/pdf/1703.07220.pdf这篇文章的主要有两个贡献:对market1501数据库和Duke数据库的行人属性进行了标记,方便了后人的工作。数据库与代码链接:https://van...原创 2019-03-26 13:46:50 · 1850 阅读 · 10 评论 -
行人重识别之Market1501数据库介绍
Scalable Person Re-identification: A Benchmark原文链接:https://www.researchgate.net/profile/Liang_Zheng17/publication/300412402_Scalable_Person_Re-identification_A_Benchmark/links/5a127308458515cc5aa9d34...原创 2019-03-22 15:26:06 · 10175 阅读 · 1 评论 -
Unsupervised Person Re-identification by Soft Multilabel Learning
CVPR2019 行人重识别Unsupervised Person Re-identification by Soft Multilabel Learning原文链接:https://arxiv.org/abs/1903.06325本文介绍了一种无监督学习的REID新算法,主要有两个贡献:1. 提出了soft multilabel,一种应用于无监督学习的REID新思想。2. 基于so...原创 2019-03-19 14:21:15 · 5497 阅读 · 7 评论 -
Dissecting Person Re-identification from the Viewpoint of Viewpoint
CVPR2019 行人重识别Dissecting Person Re-identification from the Viewpoint of Viewpoint原文链接:https://arxiv.org/abs/1812.02162这篇文章主要有两个贡献:1. 提出了生成行人图片的数据生成器2. 基于该数据生成器,研究了视角对于REID的影响分别聊一聊一、数据生成器该数据生成器...原创 2019-03-14 14:46:57 · 1818 阅读 · 0 评论