Person Re-identification by Multi-Channel Parts-Based CNN with Improved Triplet Loss Function

Cheng, De; Gong, Yi; Zhou, Sanping; Wang, Jinjun; Zheng, Nanning

doi:10.1109/cvpr.2016.149

Cited by 1,111 publications

(838 citation statements)

References 34 publications

Supporting

Mentioning

836

Contrasting

Order By: Relevance

“…Only K-LFDA when trained with mom LE [24] feature attains comparable performance than DMN. However, motivated to resolve the challenges for reidentification in real world (i.e., multimodal image space, and diverse impostors) IRM3 + CVI ( = 15) has much better results than MCP-CNN [39], E2E-CAN [31], Quadruplet-Net [33], and JLML [34], while our IRM3 + CVI ( = 15) has 1.49% higher rank@1 than DLPA [32]. DLPA extracts deep features by semantically aligning body parts, as well as rectifying pose variations.…”

Section: Results On Cuhk01mentioning

confidence: 99%

“…Since the positive samples for each person are too scarce compared to the number of negative samples, therefore, following the protocol of data augmentation in [49] we augment each person pair five times. Similarly, following the protocol in [39] we generate 20 triplets for each positive pair. Now, the triplet samples imp and Ng for person using impostor and negative Gallery are given as…”

Section: Triplet Formationmentioning

confidence: 99%

See 1 more Smart Citation

Impostor Resilient Multimodal Metric Learning for Person Reidentification

Syed

Han

et al. 2018

Advances in Multimedia

View full text Add to dashboard Cite

In person reidentification distance metric learning suffers a great challenge from impostor persons. Mostly, distance metrics are learned by maximizing the similarity between positive pair against impostors that lie on different transform modals. In addition, these impostors are obtained from Gallery view for query sample only, while the Gallery sample is totally ignored. In real world, a given pair of query and Gallery experience different changes in pose, viewpoint, and lighting. Thus, impostors only from Gallery view can not optimally maximize their similarity. Therefore, to resolve these issues we have proposed an impostor resilient multimodal metric (IRM3). IRM3 is learned for each modal transform in the image space and uses impostors from both Probe and Gallery views to effectively restrict large number of impostors. Learned IRM3 is then evaluated on three benchmark datasets, VIPeR, CUHK01, and CUHK03, and shows significant improvement in performance compared to many previous approaches.

show abstract

Section: Results On Cuhk01mentioning

confidence: 99%

Section: Triplet Formationmentioning

confidence: 99%

Impostor Resilient Multimodal Metric Learning for Person Reidentification

Syed

Han

et al. 2018

Advances in Multimedia

View full text Add to dashboard Cite

show abstract

“…As showed in figure [8] , all the network share the same parameters W. we extracted the discriminative features from the raw images into a learned feature space. This is like .…”

Section: The Overall Structurementioning

confidence: 99%

A Study on Person Re-Identification System using Enhanced RNN

Choi¹,

Xu²

2017

JIIBC

View full text Add to dashboard Cite

The person Re-identification is the most challenging part of computer vision due to the significant changes in human pose and background clutter with occlusions. The picture from non-overlapping cameras enhance the difficulty to distinguish some person from the other. To reach a better performance match, most methods use feature selection and distance metrics separately to get discriminative representations and proper distance to describe the similarity between person and kind of ignoring some significant features. This situation has encouraged us to consider a novel method to deal with this problem. In this paper, we proposed an enhanced recurrent neural network with three-tier hierarchical network for person re-identification. Specifically, the proposed recurrent neural network (RNN) model contain an iterative expectation maximum (EM) algorithm and three-tier Hierarchical network to jointly learn both the discriminative features and metrics distance. The iterative EM algorithm can fully use of the feature extraction ability of convolutional neural network (CNN) which is in series before the RNN. By unsupervised learning, the EM framework can change the labels of the patches and train larger datasets. Through the three-tier hierarchical network, the convolutional neural network, recurrent network and pooling layer can jointly be a feature extractor to better train the network. The experimental result shows that comparing with other researchers' approaches in this field, this method also can get a competitive accuracy. The influence of different component of this method will be analyzed and evaluated in the future research.

show abstract

“…use the same attributes as [83] when testing on PRID [65] dataset). Evaluated on four sepereate datasets (iLIDS, PRID, VIPeR, SAIVT-SoftBio), they are consistently able to outperform the techniques they evaluate against with particularly positive results achieved on the iLIDS dataset where they were able to achieve an accuracy rate of almost 100% at rank 50. person from an enrolled image including siamese convolutional neural networks (SCNN) [145,160] and multi channel CNN models such as triplets [24,93]. However in the majority of this research the computational expense exceeds that witnessed in the hand crafted counterparts making them currently difficult to run on real time systems.…”

Section: Person Re-identificationmentioning

confidence: 99%

“…Using this stored background model, the similarity of the template to the target location in the background is computed, 24) and use this similarity to determine if the particle is a better match to the background or the foreground,…”

Section: Particle Based Searchmentioning

confidence: 99%

Locating People in Video Surveillance from Semantic Descriptions

Halstead¹

View full text Add to dashboard Cite

AcknowledgmentsThe completion of my PhD research programme has been an enjoyable and at times difficult and complex journey that has involved a number of people that have assisted me in so many ways throughout the process. The people that I thank here is by no means a comprehensive list as there isn't enough paper for the names I have to mention.I would firstly like to thank my supervisory team, Clinton Fookes, Sridha Sridharan, and Simon Denman for providing valuable guidance throughout the course of my study. I would like to thank Clinton and Sridha for providing me with the opportunity to study my PhD, and also for providing the Vision and Signal Processing (VSP) laboratory environment that allowed me to research at a high level, and interact with other PhD students to assist with my PhD. I would like to make special mention of Simon for his day to day support and being consistently available to help me with my endless questions, and without his help, guidance, and encouragement I would not have made it to the end of my PhD. Another member of the of the VSP discipline that I would like to thank is David Dean, as a member of VSP he assisted me multiple times throughout the course of my PhD.I would also like to thank the other students and past and present staff within the VSP laboratory, thank you for the sharing of knowledge but mostly thank you all for the entertainment, and enjoyable moments that we shared throughout this time. AbstractIn surveillance and security situations it is often necessary to search and locate a subject of interest from a verbally or textually supplied subject description. At present, searches using these forms of queries are still predominantly performed manually and often ineffectively, via physical searching of a premises or personally watching countless hours of video footage, looking for the specific subject of interest. Surveillance networks are typically monitored by few people viewing several monitors displaying single and multiple camera feeds. In these situations, the ability to translate an externally supplied description into one of computer vision relevance would be of great assistance.To date, state-of-the-art soft biometric based search techniques primarily include some form of pre-enrollment component, where the target subject is enrolled from one camera and searched in other cameras for a match, or alternatively enrolled in an earlier instance of the same camera, and searched for in later footage. These re-identification techniques exclude however, instances where the subject has not previously been located and enrolled from the footage.Soft biometric based semantic search is one technique that provides a solution to this problem. The target query no longer requires scene interrogation and instead can be generated from a textually supplied query. While soft biometrics have obvious limitations, non-permanence and not individually identifiable, they are generally simplistic for a human operator to visually extract and relay to another human operator. They also have th...

show abstract

Person Re-identification by Multi-Channel Parts-Based CNN with Improved Triplet Loss Function

Cited by 1,111 publications

References 34 publications

Impostor Resilient Multimodal Metric Learning for Person Reidentification

Impostor Resilient Multimodal Metric Learning for Person Reidentification

A Study on Person Re-Identification System using Enhanced RNN

Locating People in Video Surveillance from Semantic Descriptions

Contact Info

Product

Resources

About