Unifying Person and Vehicle Re-Identification

Organisciak, Daniel; Sakkos, Dimitrios; Ho, Edmond S. L.; Aslam, Nauman; Shum, Hubert P. H.

doi:10.1109/access.2020.3004092

Cited by 10 publications

(9 citation statements)

References 59 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…[4] proposed a weighted regularization triplet (WRT) loss function by using softmax weight distributions to weight triplet samples, which makes triplet samples are more robust to outliers. The similar weighted regularization is also presented in [7, 17]. Chu et al.…”

Section: Related Workmentioning

confidence: 95%

“…Benefit from the development of deep learning, the methods of improving object re‐identification methods can be classified in two categories: (1) loss function designs [4, 7, 16–19, 26], and (2) feature learning architectures [2, 3, 5, 8–15, 27–29].…”

Section: Related Workmentioning

confidence: 99%

“…With the explosive development of deep learning, many object re-identification methods are dedicated to design feature learning architectures [2,3,5,[8][9][10][11][12][13][14][15] and loss functions [4,7,[16][17][18][19]. Regarding feature learning architectures, there are many impressive works.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

A sample‐proxy dual triplet loss function for object re‐identification

Shen

Zhu

et al. 2022

IET Image Processing

View full text Add to dashboard Cite

Object re-identification, such as vehicle re-identification or pedestrian re-identification, plays a significant role in intelligent video surveillance systems for public security. Due to viewpoint variations and appearance changes, both pedestrians and vehicles usually have complex intra-class variations. However, most existing object re-identification methods often use a sample-level triplet loss function cooperating with a single-proxy softmax loss function, which could not handle complex intra-class variations well. In this paper, a sample-proxy dual triplet (SPDT) loss function is proposed, which works with a multiproxy softmax (MPS) loss function. The MPS loss function is in charge of learning multiple proxies to represent a class. The SPDT loss function is responsible for enlarging inter-class distances as well as shrinking intra-class distances on both sample and proxy levels. Therefore, the method not only handles multi-proxy intra-class variations but also fully learns discrimination on samples and proxies. Experiments on two large datasets, that is, VeRi776 and DukeMTMC-reID, demonstrate that the method is superior to state-of-the-art object re-identification approaches.

show abstract

Section: Related Workmentioning

confidence: 95%

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

A sample‐proxy dual triplet loss function for object re‐identification

Shen

Zhu

et al. 2022

IET Image Processing

View full text Add to dashboard Cite

show abstract

“…To handle this problem, it is often beneficial to consider metric learning, usually in the form of the triplet loss [21] or centre loss [22]. The triplet loss in particular has seen extensive use for person [19], [23] and vehicle [24] reID, and can even handle both task simultaneously [25]. It is therefore natural to consider the triplet loss for UAV reID.…”

Section: A Re-identificationmentioning

confidence: 99%

UAV-ReID: A Benchmark on Unmanned Aerial Vehicle Re-identification in Video Imagery

Organisciak¹,

Poyser²,

Alsehaim³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

As unmanned aerial vehicles (UAVs) become more accessible with a growing range of applications, the potential risk of UAV disruption increases. Recent development in deep learning allows vision-based counter-UAV systems to detect and track UAVs with a single camera. However, the coverage of a single camera is limited, necessitating the need for multicamera configurations to match UAVs across cameras -a problem known as re-identification (reID). While there has been extensive research on person and vehicle reID to match objects across time and viewpoints, to the best of our knowledge, there has been no research in UAV reID. UAVs are challenging to re-identify: they are much smaller than pedestrians and vehicles and they are often detected in the air so appear at a greater range of angles. Because no UAV data sets currently use multiple cameras, we propose the first new UAV re-identification data set, UAV-reID, that facilitates the development of machine learning solutions in this emerging area. UAV-reID has two settings: Temporally-Near to evaluate performance across views to assist tracking frameworks, and Big-to-Small to evaluate reID performance across scale and to allow early reID when UAVs are detected from a long distance. We conduct a benchmark study by extensively evaluating different reID backbones and loss functions. We demonstrate that with the right setup, deep networks are powerful enough to learn good representations for UAVs, achieving 81.9% mAP on the Temporally-Near setting and 46.5% on the challenging Big-to-Small setting. Furthermore, we find that vision transformers are the most robust to extreme variance of scale.

show abstract

“…Content may change prior to final publication. [41] 62.1 77.8 --HA-CNN [42] 63.8 80.5 --PCB+RPP [43] 69.2 83.3 90.5 92.5 Mancs [44] 71.8 84.9 --CAMA [45] 72.9 85.8 --BFE [46] 75.8 88.7 --BAT-net w/ AiA [47] [16] 67.4 68.0 66.0 68.0 MGTS [41] 76.7 81.6 --MidTriNet [48] 88.5 91.0 --Ours 70.9 72.4 67.3 69.9…”

Section: Volume 4 2016mentioning

confidence: 99%

Deep Learning Research With an Expectation-Maximization Model for Person Re-Identification

2020

View full text Add to dashboard Cite

In existing person re-identification methods based on deep learning, the extraction of good features is still a key step. Some efforts divide the image of a person into multiple parts to extract more detailed information from semantically coherent parts but ignore their correlation with each other. Others adopt self attention to reallocate weights of pixels for learning the association between different regions. This association can improve the accuracy of the person re-identification task, but the features obtained by this type of algorithm have high redundancy, which is not conducive to the expression of feature information. In order to address the above challenges, we propose a feature extraction method based on a novel attention mechanism which combines the expectation maximization (EM) algorithm and non-local operation. We embed the attention module into the ResNet50 backbone network. The attention module captures the correlation between different regional features through non-local operation and then reconstructs these features through the EM algorithm. In addition, we divide the network into a global branch and a local branch, where the global branch extracts the complete features, and the local branch uses the Batch DropBlock method to erase a portion of the features to achieve feature diversity. Finally, extensive experiments validate the superiority of the proposed model for person re-ID over a wide variety of state-of-the-art methods on three large-scale benchmarks, including DukeMTMC-ReID, Market-1501 and CUHK03.

show abstract

Unifying Person and Vehicle Re-Identification

Cited by 10 publications

References 59 publications

A sample‐proxy dual triplet loss function for object re‐identification

A sample‐proxy dual triplet loss function for object re‐identification

UAV-ReID: A Benchmark on Unmanned Aerial Vehicle Re-identification in Video Imagery

Deep Learning Research With an Expectation-Maximization Model for Person Re-Identification

Contact Info

Product

Resources

About