CVPR2020|行人检测与重识算法推荐论文源码大盘点

 

本文收集了CVPR 2020 一些行人检测与人员重识别优秀论文,我们知道在视频监控相关领域这些技术方向可以得到很好得广泛应用。

 

行人检测及人群计数从内容来看主要解决行人与行人、行人与物体间的遮挡透视,和尺度问题带来得挑战

 

人员重识别有基于静态和动态视图ReID,方向可细分为:跨分辨率、跨域、跨模态(可见光-红外)、遮挡、非监督、射频信号人员重识别等。

 

相关论文

1.Detection in Crowded Scenes: One Proposal, Multiple Predictions

旷视研究院提出密集场景检测新方法:一个候选框,多个预测结果

论文地址:

https://arxiv.org/pdf/2003.09163.pdf

源码地址:

https://github.com/megvii-model/CrowdDetection

 

2.Detection in Crowded Scenes: One Proposal, Multiple Predictions

按代表性Region划分的NMS-通过Proposal配对实现拥挤行人检测

论文地址:

https://arxiv.org/pdf/2003.12729.pdf

 

3.STINet: Spatio-Temporal-Interactive Network for Pedestrian Detection and Trajectory Prediction

STINET:用于行人检测和轨迹预测的时空交互网络

论文地址:

https://arxiv.org/pdf/2005.04255.pdf

 

4.STINet: Spatio-Temporal-Interactive Network for Pedestrian Detection and Trajectory Prediction

多模态学习满足行人检测

论文地址:

http://openaccess.thecvf.com/content_CVPR_2020/papers/Luo_Where_What_Whether_Multi-Modal_Learning_Meets_Pedestrian_Detection_CVPR_2020_paper.pdf

 

5.Temporal-Context Enhanced Detection of Heavily Occluded Pedestrians

地平线提出用时序信息提升行人检测准确度

论文地址:

https://cse.buffalo.edu/~jsyuan/papers/2020/TFAN.pdf

 

6.Attention Scaling for Crowd Counting

摘要

  人群计数的主要任务是学习图片与密度图之间的映射关系。由于人群密度在图片上的变化较大,以数据为驱动的网络很容易在人群数量估计时出现误差。为解决这一问题,我们提出了一种减轻不同区域计数性能差异的方法。主要包括Density Attention Network(DANet) 和Attention Scaling Network(ASNet)两个网络。DANet为ASNet提供了与不同密度水平区域相关的注意力掩膜。ASNet首先生成中间密度图和缩放因子,然后将它们与注意力掩膜相乘,以输出多张基于注意力的不同密度水平的密度图。这些密度图相加得到最终密度图。注意尺度因子有助于减弱不同区域的估计误差。此外,还提出了一种新的自适应金字塔损失(APLoss)来分层计算子区域的估计损失,从而减少训练偏差。并在多个数据集上取得了不错的效果

论文地址:

http://openaccess.thecvf.com/content_CVPR_2020/papers/Jiang_Attention_Scaling_for_Crowd_Counting_CVPR_2020_paper.pdf

源码地址:

https://github.com/gjy3035/Awesome-Crowd-Counting

 

7.Reverse Perspective Network for Perspective-Aware Object Counting

反向透视网络用于透视感知对象计数

 

 

8.Adaptive Dilated Network With Self-Correction Supervision for Counting

 

它由自适应膨胀卷积网络和自校正监督组成。在这一部分,我们首先会从高斯混合模型(GMM)的角度理解传统的目标密度图,然后我们将介绍如何利用一种期望最大化(EM)的方式进行自纠正更新标签,最后将介绍自适应膨胀率卷积的网络结构和实现细节

论文地址:

http://openaccess.thecvf.com/content_CVPR_2020/papers/Bai_Adaptive_Dilated_Network_With_Self-Correction_Supervision_for_Counting_CVPR_2020_paper.pdf

 

9.Camera On-Boarding for Person Re-Identification Using Hypothesis Transfer Learning

使用假设转移学习的机载人重新识别相机

论文地址:

https://vcg.engr.ucr.edu/sites/g/files/rcwecm2661/files/2020-04/09517.pdf

源码地址:

https://github.com/REID-HTL/reid_htl

 

10.Hierarchical Clustering With Hard-Batch Triplet Loss for Person Re-Identification

具有硬批量三重损失的层次聚类,用于人员重新识别

论文地址:

https://arxiv.org/pdf/1910.12278.pdf

源码地址:

https://github.com/zengkaiwei/HCT

 

11.Real-world Person Re-Identification via Degradation Invariance Learning

通过退化不变性学习对现实世界中的人进行重新识别

论文地址:

https://arxiv.org/pdf/2004.04933.pdf

 

12.Unity Style Transfer for Person Re-Identification

通过退化不变性学习对现实世界中的人进行重新识别

与传统意义上的风格迁移不同,用于Re-ID的风格迁移更像是对一组图库统一风格的描述。之前已经有如DiscoGAN和CycleGAN的工作,这篇论文在二者的基础上更进一步,结合了二者的优点,使得该模型能生成稳定的相机风格化图片,从而实现数据增强的目的

论文地址:

https://arxiv.org/pdf/2003.02068.pdf

 

13.Online Joint Multi-Metric Adaptation From Frequent Sharing-Subset Mining for Person Re-Identification

频繁共享子集挖掘的在线联合多指标适应,用于人员重新识别

 

论文地址:

http://openaccess.thecvf.com/content_CVPR_2020/papers/Zhou_Online_Joint_Multi-Metric_Adaptation_From_Frequent_Sharing-Subset_Mining_for_Person_CVPR_2020_paper.pdf

 

14.Style Normalization and Restitution for Generalizable Person Re-Identification

样式归一化和可归纳人重新识别的归类

论文地址:

https://arxiv.org/pdf/2005.11037v1.pdf

 

15.Relation-Aware Global Attention for Person Re-Identification

重新认识个人的关系感知全球关注

论文地址:

https://arxiv.org/pdf/1904.02998v2.pdf

源码地址:

https://github.com/microsoft/Relation-Aware-Global-Attention-Networks

 

16.Salience-Guided Cascaded Suppression Network for Person Re-Identification

显着指导的级联抑制网络用于人员重新识别

论文地址:

http://openaccess.thecvf.com/content_CVPR_2020/papers/Chen_Salience-Guided_Cascaded_Suppression_Network_for_Person_Re-Identification_CVPR_2020_paper.pdf

 

17.Spatial-Temporal Graph Convolutional Network for Video-Based Person Re-Identification

基于时空图卷积网络的视频人识别

论文地址:

http://openaccess.thecvf.com/content_CVPR_2020/papers/Yang_Spatial-Temporal_Graph_Convolutional_Network_for_Video-Based_Person_Re-Identification_CVPR_2020_paper.pdf

 

18.Learning Multi-Granular Hypergraphs for Video-Based Person Re-Identification

学习多粒度超图,用于基于视频的人员重新识别

 

论文地址:

http://openaccess.thecvf.com/content_CVPR_2020/papers/Yan_Learning_Multi-Granular_Hypergraphs_for_Video-Based_Person_Re-Identification_CVPR_2020_paper.pdf

代码地址:

https://github.com/daodaofr/hypergraph_reid

 

18.Multi-Granularity Reference-Aided Attentive Feature Aggregation for Video-Based Person Re-Identification

用于基于视频的人员重新识别的多粒度参考辅助注意特征聚合

论文地址:

https://arxiv.org/pdf/2003.12224.pdf

 

19.Transferable, Controllable, and Inconspicuous Adversarial Attacks on Person Re-identification With Deep Mis-Ranking

具有严重误行的人员重新识别的可传递,可控制和不起眼的对抗攻击

论文地址:

http://openaccess.thecvf.com/content_CVPR_2020/papers/Wang_Transferable_Controllable_and_Inconspicuous_Adversarial_Attacks_on_Person_Re-identification_With_CVPR_2020_paper.pdf

 

20.Inter-Task Association Critic for Cross-Resolution Person Re-Identification

跨任务人员重新识别的任务间协会批评

论文地址:

http://openaccess.thecvf.com/content_CVPR_2020/papers/Cheng_Inter-Task_Association_Critic_for_Cross-Resolution_Person_Re-Identification_CVPR_2020_paper.pdf

 

21.Unsupervised Person Re-Identification via Softened Similarity Learning

通过软化相似学习进行无人监督的重新识别

论文地址:

http://openaccess.thecvf.com/content_CVPR_2020/papers/Lin_Unsupervised_Person_Re-Identification_via_Softened_Similarity_Learning_CVPR_2020_paper.pdf

 

22.Unsupervised Person Re-Identification via Multi-Label Classification

通过多标签分类对无人监督的人员进行重新识别

论文地址:

http://openaccess.thecvf.com/content_CVPR_2020/papers/Wang_Unsupervised_Person_Re-Identification_via_Multi-Label_Classification_CVPR_2020_paper.pdf

源码地址:

https://github.com/wangguanan/HOReID

 

23.High-Order Information Matters: Learning Relation and Topology for Occluded Person Re-Identification

高阶信息问题:重新关联的人的学习关系和拓扑识别。

论文地址:

http://openaccess.thecvf.com/content_CVPR_2020/papers/Wang_High-Order_Information_Matters_Learning_Relation_and_Topology_for_Occluded_Person_CVPR_2020_paper.pdf

源码地址:

https://github.com/hh23333/PVPM

 

24.Pose-Guided Visible Part Matching for Occluded Person ReID

闭塞者ReID的姿势指导可见部分匹配

 

论文地址:

http://openaccess.thecvf.com/content_CVPR_2020/papers/Gao_Pose-Guided_Visible_Part_Matching_for_Occluded_Person_ReID_CVPR_2020_paper.pdf

 

 

 

25.AD-Cluster: Augmented Discriminative Clustering for Domain Adaptive Person Re-Identification

AD群集:增强的区分性聚类,用于域自适应人员重新识别

论坛地址:
http://openaccess.thecvf.com/content_CVPR_2020/papers/Zhai_AD-Cluster_Augmented_Discriminative_Clustering_for_Domain_Adaptive_Person_Re-Identification_CVPR_2020_paper.pdf

 

26.Smoothing Adversarial Domain Attack and P-Memory Reconsolidation for Cross-Domain Person Re-Identification

平滑对抗域攻击和P记忆整合,以实现跨域人员重新识别

 

论文地址:

http://openaccess.thecvf.com/content_CVPR_2020/papers/Wang_Smoothing_Adversarial_Domain_Attack_and_P-Memory_Reconsolidation_for_Cross-Domain_Person_CVPR_2020_paper.pdf

 

27.Hi-CMD: Hierarchical Cross-Modality Disentanglement for Visible-Infrared Person Re-Identification

Hi-CMD:可视化红外人员重新识别的分层跨模态解缠

论坛地址:

http://openaccess.thecvf.com/content_CVPR_2020/papers/Choi_Hi-CMD_Hierarchical_Cross-Modality_Disentanglement_for_Visible-Infrared_Person_Re-Identification_CVPR_2020_paper.pdf

源码地址:

https://github.com/bismex/HiCMD

 

28.Cross-Modality Person Re-Identification With Shared-Specific Feature Transfer

具有共享特定特征转移的跨模式人员重新识别

论文地址:

http://openaccess.thecvf.com/content_CVPR_2020/papers/Lu_Cross-Modality_Person_Re-Identification_With_Shared-Specific_Feature_Transfer_CVPR_2020_paper.pdf

 

29.Learning Longterm Representations for Person Re-Identification Using Radio Signals

学习长期表示以使用无线电信号进行人员重新识别

 

论文地址:

http://rf-reid.csail.mit.edu/papers/rfreid_cvpr.pdf

 

30.COCAS: A Large-Scale Clothes Changing Person Dataset for Re-Identification

COCAS:用于重新识别的大规模换衣服人数据集

 论文地址:

https://arxiv.org/pdf/2005.07862.pdf

 

更多论文地址源码地址:关注“图像算法”微信公众号

  • 0
    点赞
  • 41
    收藏
    觉得还不错? 一键收藏
  • 0
    评论

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值