MARS: A Video Benchmark for Large-Scale Person Re-Identification

Zheng, Liang; Bie, Zhi; Sun, Yifan; Wang, Jingdong; Su, Chi; Wang, Shengjin; Tian, Qi

doi:10.1007/978-3-319-46466-4_52

Cited by 727 publications

(714 citation statements)

References 41 publications

Supporting

Mentioning

703

Contrasting

Order By: Relevance

“…CNN can learn discriminative embeddings by itself without part-matching. Zheng et al [1], [8], [16] directly use a conventional finetuning approach on Market1501 [3], PRW [8] and MARS [16] and outperform many recent results. Wu et al [17] combine CNN embeddings with the hand-crafted features in the FC layer.…”

Section: B Identification Modelsmentioning

confidence: 99%

See 1 more Smart Citation

A Discriminatively Learned CNN Embedding for Person Reidentification

Zheng

Yang

2017

ACM Trans. Multimedia Comput. Commun. Appl.

Self Cite

611

375

View full text Add to dashboard Cite

Abstract-In this paper, we revisit two popular convolutional neural networks (CNN) in person re-identification (re-ID), i.e.,verification and identification models. The two models have their respective advantages and limitations due to different loss functions. In this paper, we shed light on how to combine the two models to learn more discriminative pedestrian descriptors. Specifically, we propose a siamese network that simultaneously computes the identification loss and verification loss. Given a pair of training images, the network predicts the identities of the two input images and whether they belong to the same identity. Our network learns a discriminative embedding and a similarity measurement at the same time, thus making full usage of the re-ID annotations.Our method can be easily applied on different pre-trained networks. Albeit simple, the learned embedding improves the state-of-the-art performance on two public person re-ID benchmarks. Further, we show our architecture can also be applied in image retrieval.

show abstract

Section: B Identification Modelsmentioning

confidence: 99%

“…The high-level feature from the fine-tuned CNN has shown a discriminative ability [8], [16] and it is more compact than the activations in the intermediate layers. So in our model, …”

Section: Verification Lossmentioning

confidence: 99%

A Discriminatively Learned CNN Embedding for Person Reidentification

Zheng

Yang

2017

ACM Trans. Multimedia Comput. Commun. Appl.

Self Cite

611

375

View full text Add to dashboard Cite

show abstract

“…Ahmed et al [15] presented a deep convolutional architecture that captured local relationships between person images based on mid-level features. Generally, deep learning is usually utilized to learn feature representations by using deep convolutional features [14][15][16][17] or from the fully connected features [18][19][20] in person re-identification works.…”

Section: Related Workmentioning

confidence: 99%

“…(i) Feature construction and learning aim at designing or studying discriminative appearance descriptions [8][9][10][11][12][13][14][15][16][17][18][19][20] that are robust for distinguishing different pedestrians across arbitrary cameras. However, handcrafted feature construction is extremely challenging due to miscellaneous and complicated variations.…”

Section: Introductionmentioning

confidence: 99%

Co-Metric Learning for Person Re-Identification

Leng

2018

Advances in Multimedia

View full text Add to dashboard Cite

Person re-identification, aiming to identify the same pedestrian images across disjoint camera views, is a key technique of intelligent video surveillance. Although existing methods have developed both theories and experimental results, most of effective ones pertain to fully supervised training styles, which suffer the small sample size (SSS) problem a lot, especially in label-insufficient practical applications. To bridge SSS problem and learning model with small labels, a novel semisupervised co-metric learning framework is proposed to learn a discriminative Mahalanobis-like distance matrix for label-insufficient person re-identification. Different from typical co-training task that contains multiview data originally, single-view person images are firstly decomposed into pseudo two views, and then metric learning models are produced and jointly updated based on both pseudo-labels and references iteratively. Experiments carried out on three representative person re-identification datasets show that the proposed method performs better than state of the art and possesses low label sensitivity.

show abstract

“…On the other hand, assuming the availability of multiple shots of a target person available, multi-shot re-ID also attracted the interests of many researchers [50,51,37]. Furthermore, by extending multiple shots to a short video clip, Wang et al [52] started the work on video-based re-ID and drew a lot attentions from other researchers [53,54,55,56]. Around the same time, several unsupervised re-ID methods [57,58,59,60] were proposed to tackle the challenge of ground truth labeling for person re-ID.…”

mentioning

confidence: 99%

The empirical moment matrix and its application in computer vision

GOU¹

View full text Add to dashboard Cite

(GOG). The images on the first row are from the "probe" view and the second row are from the "gallery" view. In these examples, with the help of higher order moments, moM has better invariance property when the person has a fine-detailed non-Gaussian appearance, e.g., the dot-pattern in column 1, white strip on the backpack in column 2, black strip on a white shirt in column 3 and the stripe-pattern in column 4. . The best result for each metric learning is marked in red and the second best is in blue. The best combination across the whole dataset is mark in bold red. I would also like to thank Professor Mario Sznaier for his guidance and help on all kinds of different fundamental theoretical problems. His endless support makes me feel worried less to pursue the highest goal. I also want to thank Professor Richard Radke for showing me how to collaborate with other team players and the careful counsel on my manuscripts.I would like to thank Professor Jennifer Dy for serving as my committee member and for her insightful comments on my work.I would like to thank Dr. Milind Rajadhyaksha and Dr. Kivanc Kose for their help and wise advice on medical image processing. I'm thrilled to see my knowledge can help cancer patients.I also want to thank Professor Dana Brooks, Professor Mark Niedre and Professor Ningfang Mi for their patience and thoughtful guidance on the organization of NEPSSS seminars. I enjoyed the time with NEPSSS! I would like to thank the supportive faculties and staff in ALERT, especially Professor Michael Silevitch for his insightful advice and the "so what who cares" and Deanna Beirne for her help to maintain the machines and equipment of the project.I want to thank my collaborators from RPI, Dr. Ziyan Wu, Dr. Yang Li and Dr. Srikrishna Karanam for their thoughtful discussions and help on the CLE Airport project.I spent my six year Ph.D. life in one of the best labs, and I would thank all my lab mates, collaborators, mentors and best friends in the Robust System Lab. I would like to thank Binlong, Fei, Walter, Xuefeng, Xiao, Burak, Tom, Oliver, Yongfang, Yin, Ichiro, Zulqarnain, Jing, Weijia, Caglayan, Rasmus, Jose, Sadjad, Tianyu, Wenqian, Bengisu, Angels, Qian, Begum, Xikang, Dong and Yuexi for their sharing and support.Last but the most important, I would like to thank three unique women in my life, my wife Dr. Wenjuan Qin and my daughters Shuhan and Shuxin. They are my source of happiness and power to help and support me through the tough but exciting six years.x Notations R, N set of real number, set of nonnegative integers x, x, X scalar, a vector in R n , a matrix in R m×n x(i) the i-th entry of x X(i, j) the (i, j)-th entry of X X F Frobenius norm of the matrix X ∈ R m×n Embedding local properties of an image, for instance its color intensities or the magnitude and orientation of its gradients, to create a representative feature is a critical component in many computer vision tasks, such as detection, classification, segmentation and tracking. A feature that is representative yet invariant t...

show abstract

MARS: A Video Benchmark for Large-Scale Person Re-Identification

Cited by 727 publications

References 41 publications

A Discriminatively Learned CNN Embedding for Person Reidentification

A Discriminatively Learned CNN Embedding for Person Reidentification

Co-Metric Learning for Person Re-Identification

The empirical moment matrix and its application in computer vision

Contact Info

Product

Resources

About