Stepwise Metric Promotion for Unsupervised Video Person Re-identification

Reference[原文]: Joselynzhao.top & 夏木青 | Stepwise Metric Promotion for Unsupervised Video Person Re-identification

Abstract

two assumptions two assumptions

1. different video track-lets typically contain different persons。

2)within each tracklet, the frames are mostly of the same per-son.

Our method is built onreciprocal nearest neighbor search and can eliminate thehard negative label matches, i.e., the cross-camera nearestneighbors of the false matches in the initial rank list. Thetracklet that passes the reciprocal nearest neighbor checkis considered to have the same ID with the query.

Introduction

our workis motivated from three aspects.
First, videos contain muchricher information than single images,
Second, video tracklets, produced by pedestrian detec-tion and tracking (Fig. 1), are reliable data source for unsu-pervised learning methods.
Third, feature learning using tracklets from the sameview may result in low discriminative ability.

framework

In brief,two steps are involved:

1. classifier initialization using thetracklets in the same camera

2. iterations between cross-camera tracklet association and featurelearning.

Our Method.

The proposed metric promotion approachiterates between model update and label estimation. Forlabel estimation, under camera A, we use each tracklet asquery to search for its k nearest neighbors (NNs) in cameraB. Among these k candidates, the best match is selected asbeing associated with the query tracklet. We employ neg-ative mining to reduce the impact of false positive matchesin the k-NNs. This k-NN search process is then reverselyrepeated using the best match as query to see whether theinitial query is its best match, a confirmation protocol toensure that the initial query and the best match are truly as-sociated. The the associated pairs are adopted for modelupdating.

References

[1] Y. Cho and K. Yoon. Improving person re-identification via pose-aware multi-shot matching. In CVPR, 2016.
[2] J. Dai, Y. Zhang, and H. Lu. Cross-view semantic projection learningfor person re-identification. In PR, 2017.
[3] A. Dehghan, S. M. Assari, and M. Shah. GMMCP tracker: Globallyoptimal generalized maximum multi clique problem for multiple ob-ject tracking. In CVPR, 2015.
[4] W. F and Z. C. Label propagation through linear neighborhoods. InTKDE, 2008.
[5] H. Fan, L. Zheng, and Y. Yang. Unsupervised person re-identification: Clustering and fine-tuning. arXiv, 2017
[6] M. Farenzena, L. Bazzani, A. Perina, V. Murino, and M. Cristani.Person re-identification by symmetry-driven accumulation of localfeatures. In CVPR, 2010.
[7] P. F. Felzenszwalb, R. B. Girshick, D. A. McAllester, and D. Ra-manan. Object detection with discriminatively trained part-basedmodels. TPAMI, 2010.
[8] D. Gray and H. Tao. Viewpoint invariant pedestrian recognition withan ensemble of localized features. In ECCV, 2008.
[9] O. Hamdoun, F. Moutarde, B. Stanciulescu, and B. Steux. Personre-identification in multi-camera system by signature based on inter-est point descriptors collected on short video sequences. In ICDSC,2008.
[10] J. F. Henriques, J. Carreira, R. Caseiro, and J. Batista. Beyond hardnegative mining: Efficient detector learning via block-circulant de-composition. In ICCV, 2013.
[11] M. Hirzer, C. Beleznai, P. M. Roth, and H. Bischof. Person re-identification by descriptive and discriminative classification. InSCIA, 2011.
[12] S. Karanam, Y. Li, and R. J. Radke. Person re-identification withdiscriminatively trained viewpoint invariant dictionaries. In ICCV,2015.
[13] S. Karanam, Y. Li, and R. J. Radke. Sparse re-id: Block sparsity forperson re-identification. In CVPR, 2015.
[14] M. K ¨ostinger, M. Hirzer, P. Wohlhart, P. M. Roth, and H. Bischof.Large scale metric learning from equivalence constraints. In CVPR,2012.
[15] N. Li, R. Jin, and Z. Zhou. Top rank optimization in linear time. InNIPS, 2014.
[16] W. Li, R. Zhao, T. Xiao, and X. Wang. Deepreid: Deep filter pairingneural network for person re-identification. In CVPR, 2014.
[17] Y. Li, Z. Wu, S. Karanam, and R. J. Radke. Multi-shot human re-identification using adaptive fisher discriminant analysis. In BMVC,2015.
[18] Z. Li, S. Chang, F. Liang, T. S. Huang, L. Cao, and J. R. Smith.Learning locally-adaptive decision functions for person verification.In CVPR, 2013
.[19] S. Liao, Y. Hu, X. Zhu, and S. Z. Li. Person re-identification by localmaximal occurrence representation and metric learning. In CVPR,2015.
[20] S. Liao, G. Zhao, V. Kellokumpu, M. Pietik¨ainen, and S. Z. Li. Mod-eling pixel process with scale invariant local patterns for backgroundsubtraction in complex scenes. In CVPR, 2010
[21] C. Liu, C. C. Loy, S. Gong, and G. Wang. POP: person re-identification post-rank optimisation. In ICCV, 2013.
[22] K. Liu, B. Ma, W. Zhang, and R. Huang. A spatio-temporal appear-ance representation for viceo-based pedestrian re-identification. InICCV, 2015.[23] W. Liu and T. Zhang. Bidirectional label propagation over graphs.IJSI, 2013.
[24] X. Liu, M. Song, D. Tao, X. Zhou, C. Chen, and J. Bu. Semi-supervised coupled dictionary learning for person re-identification.In CVPR, 2014.[25] B. M, H. S, and J. F. Hard negative mining for metric learning basedzero-shot classification. In ECCV Workshops, 2016.[26] B. Ma, Y. Su, and F. Jurie. Bicov: a novel image representation forperson re-identification and face verification. In BMVC, 2012.
[27] B. Ma, Y. Su, and F. Jurie. Covariance descriptor based on bio-inspired features for person re-identification and face verification.IVC, 2014.[28] X. Ma, X. Zhu, S. Gong, X. Xie, J. Hu, K.-M. Lam, and Y. Zhong.Person re-identification by unsupervised video matching. PR, 2017.[29] N. McLaughlin, J. Martinez del Rincon, and P. Miller. Recurrentconvolutional network for video-based person re-identification. InCVPR, 2016.
[30] M. Niall, M. del Rincon Jesus, and M. Paul. Recurrent convolutionalnetwork for video-based person re-identification. In CVPR, 2016.[31] S. Pedagadi, J. Orwell, S. A. Velastin, and B. A. Boghossian. Localfisher discriminant analysis for pedestrian re-identification. In CVPR,2013.
[32] B. J. Prosser, W. Zheng, S. Gong, and T. Xiang. Person re-identification by support vector ranking. In BMVC, 2010.
[33] P. Siva, C. Russell, and T. Xiang. In defence of negative mining forannotating weakly labeled data. In ECCV, 2012.
[34] C. Sun, D. Wang, and H. Lu. Person re-identification via distancemetric learning with latent variables. 2017.
[35] M. Tetsu, O. Takahiro, S. Einoshin, and S. Yoichi. Hierarchical gaus-sian descriptor for person re-identification. In CVPR, 2016
.[36] R. Wang, S. Shan, X. Chen, and W. Gao. Manifold-manifold distancewith application to face recognition based on image set. In CVPR,2008.
[37] T. Wang, S. Gong, X. Zhu, and S. Wang. Person re-identification byvideo ranking. In ECCV, 2014.
[38] T. Wang, S. Gong, X. Zhu, and S. Wang. Person re-identification bydiscriminative selection in video ranking. TPAMI, 2016.
[39] Y. Yan, B. Ni, Z. Song, C. Ma, Y. Yan, and X. Yang. Person re-identification via recurrent feature aggregation. In ECCV, 2016
.[40] M. Ye, C. Liang, Y. yu, and etal. Person re-identification via rankingaggregation of similarity pulling and dissimilarity pushing. In TMM,2016.
[41] M. Ye, J. Ma, J. Li, L. Zheng, and P. Yuen. Label graph matching forunsupervised video re-identification. In ICCV, 2017.
[42] J. You, A. Wu, X. Li, and W.-S. Zheng. Top-push video-based personre-identification. In CVPR, 2016.
[43] L. Zhang, T. Xiang, and S. Gong. Learning a discriminative nullspace for person re-identification. In CVPR, 2016.
[44] Y. Zhang, B. Li, and H. L. A. I. X. Ruan. Sample-specific svm learn-ing for person re-identification. In CVPR, 2016
.[45] Z. Zhang, X. Jing, and T. Wang. Label propagation based semi-supervised learning for software defect prediction. ASE, 2017.
[46] Z. Zhang, M. Zhao, and T. W. S. Chow. Label propagation andsoft-similarity measure for graph based constrained semi-supervisedlearning. In IJCNN, 2014
.[47] R. Zhao, W. Ouyang, and X. Wang. Unsupervised salience learningfor person re-identification. In CVPR, 2013.
[48] L. Zheng, Z. Bie, Y. Sun, J. Wang, C. Su, S. Wang, and Q. Tian.MARS: A video benchmark for large-scale person re-identification.In ECCV, 2016.
[49] L. Zheng, L. Shen, L. Tian, S. Wang, J. Wang, and Q. Tian. Scalableperson re-identification: A benchmark. In ICCV, 2015.
[50] L. Zheng, Y. Yang, and A. G. Hauptmann. Person re-identification:Past, present and future. arXiv, 2016.
[51] W. Zheng, S. Gong, and T. Xiang. Person re-identification by proba-bilistic relative distance comparison. In CVPR, 2011.
[52] Z. Zheng, L. Zheng, and Y. Yang. Unlabeled samples generated bygan improve the person re-identification baseline in vitro. In ICCV,2017.
[53] Z. Zhong, L. Zheng, D. Cao, and S. Li. Re-ranking person re-identification with k-reciprocal encoding. In CVPR, 2017.
[54] X. Zhu, X. Jing, F. Wu, and H. Feng. Video-based person re-identification by simultaneously learning intra-video and inter-videodistance metrics. In IJCAI, 2016.