Article
Engineering, Electrical & Electronic
Ziqi Jin, Jinheng Xie, Bizhu Wu, Linlin Shen
Summary: In this paper, a weakly supervised pedestrian segmentation framework is proposed to directly generate the foreground mask from person re-identification datasets with only image-level subject ID labels. The Image Synthesis Augmentation (ISA) technique is also introduced to further enhance the dataset. Experimental results demonstrate that the proposed framework learns robust and discriminative features, achieving significant improvement in mAP compared to the baseline on widely used datasets including Market-1501, CUHK03, and MSMT17. The code will be made available soon.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Computer Science, Artificial Intelligence
Anusha Aswath, Ahmad Alsahaf, Ben N. G. Giepmans, George Azzopardi
Summary: This review summarizes the progress of deep learning-based segmentation techniques in large-scale cellular electron microscopy (EM) over the past six years. It discusses the application of deep learning in EM segmentation, including supervised, unsupervised, and self-supervised learning methods, and examines their adaptability in segmenting cellular and sub-cellular structures. Evaluation measures for benchmarking EM datasets in various segmentation tasks are also provided. Finally, the current trends and future prospects of EM segmentation with large-scale models and unlabeled images are discussed.
MEDICAL IMAGE ANALYSIS
(2023)
Article
Engineering, Electrical & Electronic
Xiaokang Zhang, Yan Yan, Jing-Hao Xue, Yang Hua, Hanzi Wang
Summary: In this article, a novel semantic-aware occlusion-robust network (SORN) is proposed to effectively address the challenges of occlusion in person re-identification tasks. Experimental results demonstrate the superiority of the proposed method in occluded and partial person re-identification compared to other state-of-the-art methods.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2021)
Article
Computer Science, Information Systems
Guoqing Zhang, Weisi Lin, Arun Kumar Chandran, Xuan Jing
Summary: Recently, person re-identification (re-ID) has gained attention for its applications in public security. Some studies have tried to extract local representations from important regions through human semantic parsing, but valuable cues beyond the human body may be misclassified as noise. Additionally, low-quality images or occlusions can affect the accuracy of semantic regions generated by parsing models. In this paper, a complementary network is proposed to extract discriminative and robust local representations with additional clues, achieving competitive performance.
INFORMATION SCIENCES
(2023)
Article
Engineering, Electrical & Electronic
Xiujun Shu, Xiao Wang, Xianghao Zang, Shiliang Zhang, Yuanqi Chen, Ge Li, Qi Tian
Summary: This article introduces a large-scale spatio-temporal person re-identification dataset called LaST, which has larger spatial and temporal ranges and more challenging re-ID environments compared to existing datasets. The authors evaluate 14 re-ID algorithms on LaST and propose an easy-to-implement baseline algorithm. The article also demonstrates that models pre-trained on LaST can generalize well to existing datasets with short-term and cloth-changing scenarios.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2022)
Article
Computer Science, Artificial Intelligence
Jing Liu, Prayag Tiwari, Tri Gia Nguyen, Deepak Gupta, Shahab S. Band
Summary: This paper proposes an automated framework named multi-scale local-global for person re-identification. The framework includes two components, a high-order attention module to model the subtle differences among pedestrians and generate informative attention features, and a novel architecture called spectral feature transformation to optimize group wise similarities. Experimental results demonstrate the superiority of the proposed method on three benchmark datasets.
Article
Computer Science, Artificial Intelligence
Jing Zhao, Long Lan, Da Huang, Jing Ren, Wenjing Yang
Summary: This study proposes a heterogeneous learning framework based on pseudo-supervised learning, achieving improved feature expression and label prediction in the case of few-shot labeled samples. By constructing a novel isomer as the feature extractor and training it with pseudo-supervised data, the quality of pseudo-labels is enhanced. Furthermore, the introduction of a cross-level asynchronous match mechanism and a knowledge fusion strategy further optimizes the performance.
Article
Computer Science, Information Systems
Jia-Jen Wu, Keng-Hao Chang, I-Chen Lin
Summary: In this work, a generalizable person Re-ID framework named PMN is proposed, which can be directly used in target domains with stable performance after training in source domain(s) once. A part-based architecture and a Scale Adjusting Module (SAM) are employed to improve the framework's discriminative power and handle style differences, respectively. Extensive experiments demonstrate the superiority of PMN over state-of-the-art generalizable methods on multiple popular Re-ID benchmarks with cross-domain setting. Furthermore, the advantage of using PMN as a backbone for domain adaptation methods is also demonstrated.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Computer Science, Information Systems
Huanhuan Cai, Lei Huang, Wenfeng Zhang, Zhiqiang Wei
Summary: The paper proposes an end-to-end multi-task training network for semi-supervised person re-identification, which improves performance by imposing part segmentation constraint on feature maps and designing a multiple branch network structure. Fusion of loss functions aids in learning discriminative features effectively.
MULTIMEDIA TOOLS AND APPLICATIONS
(2022)
Article
Automation & Control Systems
Nayan Kumar Subhashis Behera, Pankaj Kumar Sa, Khan Muhammad, Sambit Bakshi
Summary: Person Re-identification (PRId) is essential for associating photographs/videos of individuals obtained from various occasions or across cameras, especially in emergencies. Part-level features play a crucial role in person retrieval, and using convolutional partition of body parts to learn discriminative features is highlighted in this research. The proposed method of Convolutional Part Refine (CPR) shows competitive performance and addresses the within-part inconsistency issue in partition strategies.
IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING
(2023)
Article
Engineering, Electrical & Electronic
Xiujun Shu, Ge Li, Xiao Wang, Weijian Ruan, Qi Tian
Summary: The paper proposes a semantic-guided pixel sampling approach for cloth-changing person re-identification task, achieving promising results by forcing the model to learn cues irrelevant to upper clothes and pants.
IEEE SIGNAL PROCESSING LETTERS
(2021)
Article
Computer Science, Information Systems
Yaoyu Li, Hantao Yao, Changsheng Xu
Summary: This paper investigates the application of unsupervised domain adaptation in person re-identification, introducing two consistency constraints to handle intra-domain image variations and fully mining the underlying consistency constraints. The proposed iterative Intra-domain Consistency Enhancement (ICE) approach based on the Mean Teacher framework achieves significant improvement compared with the state-of-the-art.
IEEE TRANSACTIONS ON MULTIMEDIA
(2022)
Article
Computer Science, Information Systems
Chuan Zhu, Wenjun Zhou, Yingjun Zhu, Jianmin Ma
Summary: This paper proposes a neighboring-part dependency mining and feature fusion network (NDMF-Net) to address the issue of existing methods ignoring less obvious features and spatial interdependencies. Experimental results demonstrate that our method is effective and achieves state-of-the-art performance.
Article
Computer Science, Artificial Intelligence
Dandan Zhu, Qiangqiang Zhou, Tian Han, Yongqing Chen, Defang Zhao, Xiaokang Yang
Summary: This paper proposes a novel Multi-scale Deep Feature Learning with correlation metric (MDFLCM) model to address the scale problem in Re-ID, by extracting multi-scale high-level features using a specially designed end-to-end multi-scale deep convolutional network (MS-DCN) and adding an extra correlation layer. Extensive comparative evaluations on four public datasets demonstrate the effectiveness of the proposed MDFLCM model in Re-ID.
KNOWLEDGE-BASED SYSTEMS
(2021)
Article
Computer Science, Information Systems
Kaiyang Liao, Keer Wang, Yuanlin Zheng, Guangfeng Lin, Congjun Cao
Summary: Person re-identification utilizes computer vision technology to identify specific pedestrians in images or videos. Cross-device retrieval images present greater challenges due to variations in device styles. Current algorithms often ignore the impact of different perspectives, postures, and backgrounds on features, despite using multi-feature fusion methods. This paper proposes a multi-scale feature method based on a saliency model, which effectively filters out complex background interference and achieves more robust features by fusing global and local features using a feature weighting method. Experimental results on three datasets demonstrate the superiority of the proposed method.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Lihua Fu, Yu Zhao, Xiaowei Sun, Jialiang Huang, Dan Wang, Yu Ding
Summary: This paper proposes a VOS method based on motion-aware ROI prediction and adaptive reference updating, which effectively addresses the difficulties in traditional video object segmentation methods. Experiments show that the proposed method outperforms existing approaches on public benchmark datasets.
EXPERT SYSTEMS WITH APPLICATIONS
(2021)