论文标题:UnrealPerson: An Adaptive Pipeline towards Costless Person Re-identification
论文地址:UnrealPerson: An Adaptive Pipeline towards Costless Person Re-identification
Unreal数据集下载:FlyHighest/UnrealPerson
本文对我们最新的工作UnrealPerson进行简单介绍,文章已被CVPR2021接收为oral paper。我们在实验中充分验证了在虚拟生成数据上预训练的有效性,所提方法能够大大降低再识别算法训练、部署时的标注成本。我们设计的数据生成流程生成的行人图像,无论是直接迁移、监督学习,还是无监督学习,都有出色的表现。
Motivation
^Dissecting Person Re-Identification From the Viewpoint of Viewpointhttps://www.semanticscholar.org/paper/Dissecting-Person-Re-Identification-From-the-of-Sun-Zheng/3ff74b685615f50736e10294811281c41de3d61e
^Surpassing Real-World Source Training Data: Random 3D Characters for Generalizable Person Re-Identificationhttps://www.semanticscholar.org/paper/Surpassing-Real-World-Source-Training-Data%3A-Random-Wang-Liao/9e99f02d153728a8bcad2dbe8f60dad79a457154
^UnrealCV: Virtual Worlds for Computer Visionhttps://www.semanticscholar.org/paper/UnrealCV%3A-Virtual-Worlds-for-Computer-Vision-Qiu-Zhong/cae56bb2657943bb07823fdf076625643e75095a
^Joint Visual and Temporal Consistency for Unsupervised Domain Adaptive Person Re-Identificationhttps://www.semanticscholar.org/paper/Joint-Visual-and-Temporal-Consistency-for-Domain-Li-Zhang/7dac9cc7e0b4ad6e63db59cdefd3a805bd1db279
^Rethinking the Distribution Gap of Person Re-identification with Camera-Based Batch Normalizationhttps://www.semanticscholar.org/paper/Rethinking-the-Distribution-Gap-of-Person-with-Zhuang-Wei/6dc2c37a62ad509649bf20487114f0c805deb794