Article
Computer Science, Artificial Intelligence
Xuanzhao Wang, Zhengping Che, Bo Jiang, Ning Xiao, Ke Yang, Jian Tang, Jieping Ye, Jingyu Wang, Qi Qi
Summary: This article proposes a novel video anomaly detection method based on frame prediction, with better performance and noise tolerance loss, which outperforms existing state-of-the-art methods as confirmed by experiments.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Computer Science, Information Systems
Shuning Chang, Yanchao Li, Shengmei Shen, Jiashi Feng, Zhiying Zhou
Summary: In this work, we propose a novel lightweight anomaly detection model for weakly-supervised video anomaly detection. The model fully utilizes normal videos to train a classifier with discriminative ability for normal videos, and employs a contrastive attention module to improve the selection of anomalous segments. Experimental results demonstrate that our model significantly improves the frame-level AUC compared to state-of-the-art methods.
IEEE TRANSACTIONS ON MULTIMEDIA
(2022)
Article
Computer Science, Artificial Intelligence
Ning Huyan, Dou Quan, Xiangrong Zhang, Xuefeng Liang, Jocelyn Chanussot, Licheng Jiao
Summary: This paper investigates the issue of outlier detection in deep learning methods and proposes an unsupervised outlier detection method using a memory module and a contrastive learning module. Extensive experiments demonstrate that the proposed method performs well on four benchmark datasets and outperforms eleven state-of-the-art methods.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2022)
Article
Computer Science, Artificial Intelligence
Jing Ren, Mingliang Hou, Zhixuan Liu, Xiaomei Bai
Summary: Graph anomaly detection is a popular and vital task in various real-world scenarios. Recent studies have extended deep learning-based methods and achieved preferable performance. However, existing methods lack efficiency for embedded devices. In this study, we propose EAGLE, an efficient anomaly detection model for heterogeneous graphs, which contrasts abnormal nodes with normal ones based on their distances to the local context. Experimental results demonstrate that EAGLE outperforms state-of-the-art methods on three different heterogeneous network datasets.
IEEE INTELLIGENT SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
MyeongAh Cho, Taeoh Kim, Woo Jin Kim, Suhwan Cho, Sangyoun Lee
Summary: In contemporary society, surveillance anomaly detection is a critical task, which is challenging due to the lack of labeled videos with anomalous events. Most existing methods use autoencoders to detect anomalies based on their failure to reconstruct abnormal scenes. However, the explicit separation of appearance and motion information limits their reciprocal representation capabilities. In contrast, we propose an implicit two-path autoencoder that implicitly models appearance and motion features and combines them for anomaly detection. We also suggest normal density estimation using generative models to enhance the performance of the autoencoder.
PATTERN RECOGNITION
(2022)
Article
Computer Science, Artificial Intelligence
Dongxiao He, Chundong Liang, Cuiying Huo, Zhiyong Feng, Di Jin, Liang Yang, Weixiong Zhang
Summary: Heterogeneous Information Networks (HINs) are powerful models for complex systems. However, the presence of missing attributes in HINs can significantly degrade the performance of supervised and unsupervised learning. In this study, an unsupervised heterogeneous graph contrastive learning approach was developed to address this issue by unifying attribute completion and representation learning. The results showed that this approach outperformed several state-of-the-art methods and improved the performance of existing HIN models.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Wei Zhu, Weijian Li, E. Ray Dorsey, Jiebo Luo
Summary: In this paper, a novel anomaly detection method for time series data is proposed. By contrasting the whole time series with its sub-sequences at different timestamps in a latent space, the model effectively captures local and global features using convolutional neural networks and attention mechanism. The proposed approach shows promising potential in tackling real-world anomaly detection tasks.
Article
Engineering, Civil
Qian Zhang, Mingxin Zhang, Jinghe Liu, Xuanyu He, Ran Song, Wei Zhang
Summary: This paper proposes a simple but effective method to solve the hard-positive problem in vessel re-identification. It constructs a multi-level contrastive learning framework trained with a specifically designed intra-batch cluster-level contrastive loss and an instance-level contrastive loss. Experimental results on a newly proposed dataset show that this method achieves state-of-the-art performance compared to other unsupervised methods.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2023)
Article
Computer Science, Information Systems
Kian Yu Gan, Yu Tong Cheng, Hung-Khoon Tan, Hui-Fuang Ng, Maylor Karhang Leung, Joon Huang Chuah
Summary: Video anomaly detection aims to identify anomalous segments in a video by training with weakly supervised video-level labels. This paper focuses on two crucial factors that affect the performance of video anomaly detection models. Firstly, the paper proposes a U-Net like structure to capture both local and global temporal dependencies effectively. Secondly, it introduces weakly supervised contrastive regularization to address overfitting and improve feature generalizability. Experimental results on the UCF-Crime dataset demonstrate that the proposed approach outperforms several state-of-the-art methods.
Article
Computer Science, Artificial Intelligence
Romany F. Mansour, Jose Escorcia-Gutierrez, Margarita Gamarra, Jair A. Villanueva, Nallig Leal
Summary: Intelligent video surveillance applications are increasingly important, with anomaly detection and classification being key elements. The IVADC-FDRL model presented in this paper combines Faster RCNN and deep reinforcement learning technologies for efficient anomaly detection and classification.
IMAGE AND VISION COMPUTING
(2021)
Article
Computer Science, Artificial Intelligence
Peng Hu, Hongyuan Zhu, Jie Lin, Dezhong Peng, Yin-Ping Zhao, Xi Peng
Summary: In this paper, we propose a novel approach to make unsupervised cross-modal hashing benefit from contrastive learning by addressing the performance degradation issue caused by binary optimization for hashing and alleviating the influence brought by false-negative pairs (FNPs). To achieve this, we introduce a momentum optimizer and a Cross-modal Ranking Learning loss (CRL) that utilizes the discrimination from all negative pairs. Our method shows better retrieval performance and could be one of the first successful contrastive hashing methods.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Computer Science, Information Systems
Lin Wang, Haishu Tan, Fuqiang Zhou, Wangxia Zuo, Pengfei Sun
Summary: With the rapid increase of video surveillance points, video anomaly detection has gained attention in the security field. However, the distribution of normal and anomalous data is unbalanced in unlabeled video data. This research presents a double-flow convolutional long short-term memory variational autoencoder (DF-ConvLSTM-VAE) to model the probabilistic distribution of normal video in an unsupervised learning scheme, and reconstruct videos without anomaly objects for anomaly video detection. Experiments verify the effectiveness and competitiveness of the proposed model.
Article
Acoustics
Kun Zhou, Yuanhang Zhou, Wayne Xin Zhao, Ji-Rong Wen
Summary: In this paper, we propose a new framework L2P-CSR that improves the contrastive learning of sentence representations by adopting a learnable perturbation strategy. Our framework includes a safer perturbation mechanism that weakens the influence of tokens and features on the sentence representation, as well as a gradient-based algorithm to generate adaptive perturbations for dynamically updated sentence representations. Extensive experiments demonstrate that our approach outperforms competitive baselines on diverse sentence-related tasks.
IEEE-ACM TRANSACTIONS ON AUDIO SPEECH AND LANGUAGE PROCESSING
(2023)
Article
Engineering, Electrical & Electronic
Yunzuo Zhang, Yameng Liu, Pengfei Zhu, Weili Kang
Summary: This letter presents a joint Reinforcement and Contrastive Learning framework (RCL) for unsupervised video summarization. The proposed framework addresses the shortcomings of poor feature representation and inefficient context modeling. Extensive experiments demonstrate the superior performance of the proposed method.
IEEE SIGNAL PROCESSING LETTERS
(2022)
Article
Engineering, Electrical & Electronic
K. Deepak, S. Chandrakala, C. Krishna Mohan
Summary: Modeling abnormal spatiotemporal events using a normality modeling approach and a residual autoencoder is effective in detecting anomalies in surveillance videos and shows better results compared to deeper layer approaches. The reconstruction loss method is utilized for identifying irregularities, and good practices for training autoencoders for video anomaly detection are presented using benchmark datasets.
SIGNAL IMAGE AND VIDEO PROCESSING
(2021)
Article
Computer Science, Information Systems
Xiaoyu Zhang, Wei Gao, Ge Li, Qiuping Jiang, Runmin Cong
Summary: Due to the diverse nature of degradation process, recovery of mixed distorted images remains challenging. Training deep learning models for one degradation type leads to significant decline in performance for other degradation types. In this article, we propose a hierarchical toolkit to address the limitations of a single deep network and explore the use of reinforcement learning for efficient restoration of distorted images. Our method accurately captures distortion preferences and achieves quality improvements through exploration using various evaluation methods. Experimental results show significant improvement in peak signal-to-noise ratio compared to state-of-the-art RL-based methods.
ACM TRANSACTIONS ON MULTIMEDIA COMPUTING COMMUNICATIONS AND APPLICATIONS
(2023)
Article
Engineering, Electrical & Electronic
Yiming Wang, Dongxia Chang, Zhiqiang Fu, Jie Wen, Yao Zhao
Summary: In this paper, the problem of multi-view clustering on incomplete views is studied. A novel incomplete multi-view clustering framework is proposed, which incorporates cross-view relation transfer and multi-view fusion learning. Extensive experiments on real datasets demonstrate the effectiveness of the proposed method.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Engineering, Electrical & Electronic
Zhengxuan Xie, Feng Shao, Gang Chen, Hangwei Chen, Qiuping Jiang, Xiangchao Meng, Yo-Sung Ho
Summary: RGB-T salient object detection aims to detect and segment saliency regions on RGB images and thermal maps. The proposed CMDBIF-Net effectively reduces the modality difference and achieves outstanding performance.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Computer Science, Artificial Intelligence
Zhihao Wu, Chengliang Liu, Jie Wen, Yong Xu, Jian Yang, Xuelong Li
Summary: Weakly supervised object detection (WSOD) has gained attention due to its use of image-category annotations for training. We propose selecting high-quality proposals and addressing the issues of part domination and untight boxes to improve WSOD performance. Our approach significantly improves the performance of WSOD according to experiments on PASCAL VOC and MS COCO datasets.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2023)
Article
Computer Science, Information Systems
Zhihao Wu, Jie Wen, Yong Xu, Jian Yang, David Zhang
Summary: Weakly supervised object detection (WSOD) aims to train object detectors using only image-level annotations. Many recent works use multiple instance detection networks (MIDN), but these methods tend to detect salient objects and parts due to lack of instance-level annotations. In this paper, they propose an adaptive instance refinement (AIR) framework to address this issue.
IEEE TRANSACTIONS ON MULTIMEDIA
(2023)
Article
Engineering, Electrical & Electronic
Yaozu Kang, Qiuping Jiang, Chongyi Li, Wenqi Ren, Hantao Liu, Pengjun Wang
Summary: This paper proposes a perception-aware decomposition and fusion framework for underwater image enhancement (UIE). Two complementary pre-processed inputs are fused in a perception-aware and conceptually independent image space through a structural patch decomposition and fusion (SPDF) approach. The main advantage of SPDF is that it can fuse different components separately without any interactions and information loss. Comprehensive comparisons demonstrate that SPDF outperforms state-of-the-art UIE algorithms.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Engineering, Electrical & Electronic
Gang Chen, Feng Shao, Xiongli Chai, Hangwei Chen, Qiuping Jiang, Xiangchao Meng, Yo-Sung Ho
Summary: This paper proposes a novel Modality-Induced Transfer-Fusion Network (MITF-Net) for RGB-D and RGB-T salient object detection. The network is designed to capture the complementary information of multi-modality data and improve performance through modality transfer fusion and cycle-separated attention modules. The proposed MITF-Net achieves competitive and excellent performance on RGB-D and RGB-T salient object detection datasets.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Biochemical Research Methods
Ke Yan, Hongwu Lv, Jie Wen, Yichen Guo, Yong Xu, Bin Liu
Summary: In this study, a novel stacking framework PreTP-Stack is proposed for predicting different types of therapeutic peptides. The proposed method achieved better or highly comparable performance with the state-of-the-art methods for predicting eight types of therapeutic peptides. A user-friendly web-server predictor is also available for use.
IEEE-ACM TRANSACTIONS ON COMPUTATIONAL BIOLOGY AND BIOINFORMATICS
(2023)
Review
Mathematics
Shuping Zhao, Lunke Fei, Jie Wen
Summary: This paper summarizes six single-view palmprint representation methods published from 2004 to 2022 and discusses multiview-based palmprint recognition methods. There is currently no work summarizing the multiview fusion for different types of palmprint features.
Article
Computer Science, Artificial Intelligence
Guanghui Yue, Siying Li, Tianwei Zhou, Miaohui Wang, Jingfeng Du, Qiuping Jiang, Wei Gao, Tianfu Wang, Jun Lv
Summary: Automatic and accurate polyp segmentation is a challenging issue. We construct a benchmark dataset and propose a novel adaptive context exploration network (ACENet) which achieves superior performance on multiple evaluation metrics over state-of-the-art methods.
IEEE TRANSACTIONS ON EMERGING TOPICS IN COMPUTATIONAL INTELLIGENCE
(2023)
Article
Engineering, Electrical & Electronic
Jinguang Cheng, Zongwei Wu, Shuo Wang, Cedric Demonceaux, Qiuping Jiang
Summary: This article introduces a novel organism detection method called BCMNet, which fully utilizes texture and context clues during the encoding and decoding stages, improving the accuracy of locating the target organism object in marine scenes.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Engineering, Electrical & Electronic
Hangwei Chen, Feng Shao, Xiongli Chai, Yuese Gu, Qiuping Jiang, Xiangchao Meng, Yo-Sung Ho
Summary: This paper discusses the important topic of arbitrary neural style transfer, which has significant research value and wide industrial application. The focus is on improving the quality of arbitrary style transfer (AST) and the lack of exploration in quality evaluation of AST images. The paper proposes a new AST images quality assessment database (AST-IQAD) and a sparse representation-based method for measuring the image quality.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Computer Science, Artificial Intelligence
Chao Huang, Jie Wen, Yong Xu, Qiuping Jiang, Jian Yang, Yaowei Wang, David Zhang
Summary: In this article, a novel self-supervised framework for unsupervised video anomaly detection is proposed. The framework combines a self-attentive predictor, a vanilla discriminator, and a self-supervised discriminator to capture the regular patterns in videos and effectively detect abnormal frames.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Automation & Control Systems
Zhiqiang Fu, Yao Zhao, Dongxia Chang, Yiming Wang, Jie Wen
Summary: In this article, we propose a latent low-rank representation method with weighted distance penalty (LLRRWD) for clustering. By considering the local geometric information and using a weighted distance penalty, our method improves the discrimination of the learned affinity matrix, while reducing the effect of noise and redundancy.
IEEE TRANSACTIONS ON CYBERNETICS
(2023)