Dynamic Label Graph Matching for Unsupervised Video Re-identification
Dynamic Label Graph Matching for Unsupervised Video Re-identification
复制标题
DOI：
10.1109/iccv.2017.550
复制
发表时间：
2017-09
期刊：
影响因子：
--
通讯作者：
Mang Ye;A. J. Ma;Liang Zheng;Jiawei Li;P. Yuen
中科院分区：
文献类型：
--
作者：
Mang Ye;A. J. Ma;Liang Zheng;Jiawei Li;P. Yuen
标签估计是无监督行人重识别（re-ID）系统中的一个重要组成部分。本文重点关注跨摄像机标签估计，它随后可用于特征学习以学习鲁棒的重识别模型。具体而言，我们建议为每个摄像机中的样本构建一个图，然后引入图匹配方案用于跨摄像机的标签关联。由于跨摄像机的显著差异，现有图匹配方法直接输出的标签可能存在噪声且不准确，因此本文提出一种动态图匹配（DGM）方法。DGM通过使用中间估计的标签学习一个更好的特征空间，迭代地更新图像图和标签估计过程。DGM在两个方面具有优势：1）随着迭代，估计标签的准确性显著提高；2）DGM对有噪声的初始训练数据具有鲁棒性。在包括大规模MARS数据集在内的三个基准数据集上进行的大量实验表明，DGM与完全监督的基线相比具有竞争力的性能，并且优于竞争的无监督学习方法。
Label estimation is an important component in an unsupervised person re-identification (re-ID) system. This paper focuses on cross-camera label estimation, which can be subsequently used in feature learning to learn robust re-ID models. Specifically, we propose to construct a graph for samples in each camera, and then graph matching scheme is introduced for cross-camera labeling association. While labels directly output from existing graph matching methods may be noisy and inaccurate due to significant cross-camera variations, this paper propose a dynamic graph matching (DGM) method. DGM iteratively updates the image graph and the label estimation process by learning a better feature space with intermediate estimated labels. DGM is advantageous in two aspects: 1) the accuracy of estimated labels is improved significantly with the iterations; 2) DGM is robust to noisy initial training data. Extensive experiments conducted on three benchmarks including the large-scale MARS dataset show that DGM yields competitive performance to fully supervised baselines, and outperforms competing unsupervised learning methods.1