Weakly Supervised Person Re-ID: Differentiable Graphical Learning and a New Benchmark

Weakly Supervised Person Re-ID: Differentiable Graphical Learning and a New Benchmark
复制标题

弱监督人员重新识别:可微分图形学习和新基准

DOI:
10.1109/tnnls.2020.2999517
复制
发表时间:
2019-04
影响因子:
10.4
通讯作者:
Lin Liang
Lin Liang
中科院分区:
计算机科学1区
文献类型:
--
作者:
Wang Guangrun;Wang Guangcong;Zhang Xujie;Lai Jianhuang;Yu Zhengtao;Lin Liang

文献摘要

参考文献

被引文献

相似文献

对现有数据集(例如CUHK03和Market-1501)进行准确的注释,对人员再识别大有好处。由于这些数据集中的每个图像都必须分配适当的标签,因此注释的成本相当高。在这项工作中,我们通过用不准确的注释代替准确的注释来简化Re-ID的注释,即我们按照时间将图像分组到袋中,并为每个袋分配一个袋级标签。这大大减少了注释工作,并导致创建了一个名为SYSU- $30k$的大规模Re-ID基准。新基准包含3万个类别,比CUHK03(1.3万个类别)和Market-1501(1.5万个类别)大约20倍,比ImageNet(1万个类别)大30倍。总共有29606918张图片。学习具有袋级注释的Re-ID模型称为弱监督Re-ID问题。为了解决这个问题,我们引入了一个可微的图形模型来捕获包中所有图像的依赖关系,并为每个人的图像生成一个可靠的伪标签。伪标签进一步用于监督Re-ID模型的学习。与完全监督的Re-ID模型相比,我们的方法在SYSU- $30k$和其他数据集上达到了最先进的性能。代码、数据集和预训练模型可在https://github.com/wanggrun/SYSU-30k上获得。
Person reidentification (Re-ID) benefits greatly from the accurate annotations of existing data sets (e.g., CUHK03 and Market-1501), which are quite expensive because each image in these data sets has to be assigned with a proper label. In this work, we ease the annotation of Re-ID by replacing the accurate annotation with inaccurate annotation, i.e., we group the images into bags in terms of time and assign a bag-level label for each bag. This greatly reduces the annotation effort and leads to the creation of a large-scale Re-ID benchmark called SYSU- $30k$ . The new benchmark contains 30k individuals, which is about 20 times larger than CUHK03 (1.3k individuals) and Market-1501 (1.5k individuals), and 30 times larger than ImageNet (1k categories). It sums up to 29606918 images. Learning a Re-ID model with bag-level annotation is called the weakly supervised Re-ID problem. To solve this problem, we introduce a differentiable graphical model to capture the dependencies from all images in a bag and generate a reliable pseudolabel for each person’s image. The pseudolabel is further used to supervise the learning of the Re-ID model. Compared with the fully supervised Re-ID models, our method achieves state-of-the-art performance on SYSU- $30k$ and other data sets. The code, data set, and pretrained model will be available at https://github.com/wanggrun/SYSU-30k.
DOI: 10.5244/c.28.48
发表时间: 2014
期刊: --
影响因子: --
作者:
Hanxiao Wang;S. Gong;T. Xiang
通讯作者: Hanxiao Wang;S. Gong;T. Xiang
DOI: 10.5244/c.31.135
发表时间: 2017-08
期刊: ArXiv
影响因子: --
作者:
Rui Yu;Zhichao Zhou;S. Bai;X. Bai
通讯作者: Rui Yu;Zhichao Zhou;S. Bai;X. Bai
DOI: 10.5244/c.24.21
发表时间: 2010
影响因子: 4.9
作者:
B. Prosser;Weishi Zheng;S. Gong;T. Xiang
通讯作者: B. Prosser;Weishi Zheng;S. Gong;T. Xiang
DOI: 10.1109/cvpr.2019.00375
发表时间: 2019-06
期刊: 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
影响因子: --
作者:
Q. Yang;Hong-Xing Yu;Ancong Wu;Weishi Zheng
通讯作者: Q. Yang;Hong-Xing Yu;Ancong Wu;Weishi Zheng
DOI: --
发表时间: 2017-03
期刊: ArXiv
影响因子: --
作者:
Alexander Hermans;Lucas Beyer;B. Leibe
通讯作者: Alexander Hermans;Lucas Beyer;B. Leibe