Article
Computer Science, Artificial Intelligence
Jiao Liu, Mingquan Lin, Mingbo Zhao, Choujun Zhan, Bing Li, John Kwok Tai Chui
Summary: Video surveillance is crucial for public safety in smart cities, and Person Re-Identification (Re-ID) is an essential task for video surveillance. This paper proposes a novel scalable manifold embedding approach for Person Re-ID, which incorporates graph weight construction and manifold regularized term in the same framework. Experimental results on benchmark datasets demonstrate the superior performance of the proposed method compared to other state-of-the-art graph-based methods.
APPLIED INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Feiping Nie, Zheng Wang, Rong Wang, Xuelong Li
Summary: This paper proposes a novel locality preserved dimensionality reduction framework named SALE, which can explore local discriminative embedding and global structure to improve the discriminative power of embedded data. Experimental results demonstrate the superiorities of this method in local structure exploration and classification task.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2022)
Article
Computer Science, Information Systems
Jingliu Lai, Hongmei Chen, Tianrui Li, Xiaoling Yang
Summary: In this study, a novel semi-supervised sparse feature selection framework is proposed, which improves the quality of the similarity matrix through adaptive graph learning and alleviates the negative influence of redundant features through redundancy minimization regularization.
INFORMATION SCIENCES
(2022)
Article
Computer Science, Artificial Intelligence
Yunze Chen, Junjie Huang, Zheng Zhu, Xianlei Long, Qingyi Gu
Summary: Deep facial recognition greatly benefits from large-scale training data, but the high labeling costs remain a bottleneck. To address this, we propose a semi-supervised learning method that utilizes limited labeled data and abundant unlabeled data. Our approach tackles challenges such as identity overlaps and over-decomposition, resulting in improved performance and reduced labeling costs.
IMAGE AND VISION COMPUTING
(2022)
Article
Computer Science, Information Systems
Moxian Song, Hongyan Li, Chenxi Sun, Derun Cai, Shenda Hong
Summary: Semi-supervised partial label learning is a learning method that deals with partially labeled and unlabeled data. This paper proposes a novel approach by label set assignment and dependence-maximized dimensionality reduction to obtain reliable label confidences. Extensive experiments validate the effectiveness and superiority of the proposed method.
INFORMATION SCIENCES
(2022)
Article
Engineering, Electrical & Electronic
Lin Li, Hongchun Qu, Zhaoni Li, Jian Zheng, Fei Guo
Summary: In this paper, a new supervised dimensionality reduction method called DP-ARGE is proposed to overcome the challenges in graph embedding based supervised DR methods. DP-ARGE dynamically learns the graph in a low-dimensional subspace and enhances the discrimination of projected samples through a novel discriminative regularization term. The automatic parameter estimation strategy and the extension to a semi-supervised method further improve the performance.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2022)
Article
Computer Science, Artificial Intelligence
Yuan Yuan, Xin Li, Qi Wang, Feiping Nie
Summary: Many semi-supervised learning methods, especially graph-based approaches, have been developed and achieved satisfactory performance. Graph quality directly impacts classification accuracy, but algorithms often use k-Nearest Neighbor, leading to inaccuracies due to outliers. Limited labeled data and potential errors in labels can be addressed by ALGSSL, which adapts and regularizes the graph for improved performance.
Article
Engineering, Electrical & Electronic
Yuchi Liu, Hailin Shi, Hang Du, Rui Zhu, Jun Wang, Liang Zheng, Tao Mei
Summary: In this paper, we propose an effective solution for semi-supervised face recognition that is robust to label noise. Our solution can identify wrongly-labelled samples and preserve clean samples, and it further enhances training by conducting high-confidence labelling on unlabelled data. Our method compares favorably against the state-of-the-art methods in a wide range of benchmarks.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2022)
Article
Mathematics
Runxin Li, Jiaxing Du, Jiaman Ding, Lianyin Jia, Yinong Chen, Zhenhong Shang
Summary: In this paper, a new approach called SMDR-IC is proposed for semi-supervised multi-label dimensionality reduction learning. The proposed method incorporates label propagation mechanism and investigates instance correlations using the k-nearest neighbor technique. Experimental results show that SMDR-IC outperforms other related methods on public multi-label datasets.
Article
Computer Science, Information Systems
Fengzhe Jin, Yong Peng, Feiwei Qin, Junhua Li, Wanzeng Kong
Summary: In this paper, a Graph Adaptive Semi-supervised Discriminative Subspace Learning (GASDSL) model is proposed for EEG-based emotion recognition. GASDSL aims to explore a discriminative subspace that improves emotion recognition accuracy. Comparative studies show that GASDSL achieves satisfactory results compared to other semi-supervised learning models.
JOURNAL OF KING SAUD UNIVERSITY-COMPUTER AND INFORMATION SCIENCES
(2023)
Article
Computer Science, Artificial Intelligence
Yue Fan, Anna Kukleva, Dengxin Dai, Bernt Schiele
Summary: This paper proposes an improved consistency regularization framework that achieves invariance by decreasing distances between features, leading to improved performance. Experimental results demonstrate that this method outperforms previous work in imbalanced semi-supervised learning tasks and achieves state-of-the-art results on various standard benchmarks.
INTERNATIONAL JOURNAL OF COMPUTER VISION
(2023)
Article
Computer Science, Artificial Intelligence
Prashant Shukla, Abhishek, Shekhar Verma, Manish Kumar
Summary: This paper proposes a rotation-based affinity metric for accurate graph Laplacian approximation, aiming to improve the accuracy of low-dimensional representation in manifold learning. Extensive experiments on both synthetic and real world datasets demonstrate that the proposed method outperforms existing nonlinear dimensionality reduction techniques in low-dimensional representation for synthetic datasets.
PATTERN ANALYSIS AND APPLICATIONS
(2021)
Article
Automation & Control Systems
Jianhui Yuan, Rongzhen Zhao, Tianjing He, Pengfei Chen, Kongyuan Wei, Ziyang Xing
Summary: Traditional graph embedding methods have limitations in handling high-dimensional fault data, and a novel dimensionality reduction method called SMGJE is proposed to address this issue and applied to rotor fault diagnosis. SMGJE characterizes the structure of high-dimensional data by constructing simple graphs and hypergraphs, overcoming the "averaging effect" in hypergraphs.
Article
Computer Science, Artificial Intelligence
Xinmin Tao, Yixuan Bao, Xiaohan Zhang, Tian Liang, Lin Qi, Zhiting Fan, Shan Huang
Summary: A semi-supervised kernel local Fisher discriminant analysis algorithm based on density peak clustering pseudo-labels (SDPCKLFDA) is proposed to effectively use both labeled and unlabeled data for learning. The algorithm generates pseudo cluster labels using the density peak clustering algorithm and constructs regularization strategies based on these pseudo-labels. The optimal projection vector is obtained by solving the objective function of the local Fisher discriminant analysis. The algorithm improves the discriminant performance of extracted features and is suitable for multimodal and noisy data.
NEURAL COMPUTING & APPLICATIONS
(2022)
Article
Computer Science, Artificial Intelligence
Wenrao Pang, Gang Wu
Summary: This article introduces the challenges of incremental and decremental problems in semi-supervised learning and proposes methods for incremental and decremental semi-supervised discriminant analysis. By updating scatter matrices and proposing decremental algorithms, the proposed methods demonstrate superior performance in experiments.
PATTERN RECOGNITION
(2022)
Article
Engineering, Electrical & Electronic
Lichen Zhao, Jinyang Guo, Dong Xu, Lu Sheng
Summary: In this work, a simple but effective 3D object detection method called Transformer3D-Det (T3D) is proposed, which introduces a transformer based vote refinement module to refine the voting results and significantly improve performance.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2021)
Article
Engineering, Electrical & Electronic
Jiaheng Liu, Dong Xu
Summary: In this work, a strong two-stream baseline method called GeometryMotion-Net is proposed for 3D action recognition. By utilizing new modules and virtual point clouds in both geometry and motion streams, discriminant geometry and bidirectional motion features are extracted, demonstrating effectiveness and efficiency on mainstream datasets.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2021)
Article
Engineering, Electrical & Electronic
Kaisiyuan Wang, Lu Sheng, Shuhang Gu, Dong Xu
Summary: The proposed SPU method effectively upsamples sparse, non-uniform, and orderless point cloud sequences by exploiting rich temporal dependencies from multiple inputs. The use of a new temporal alignment module and gating mechanism allows for effective aggregation of transformed features into a fused feature, which can be fed into existing frame-based point cloud upsampling methods to generate dense point clouds. Comprehensive experiments demonstrate the effectiveness of the method on benchmark datasets.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2021)
Article
Computer Science, Artificial Intelligence
Zhenghao Chen, Shuhang Gu, Guo Lu, Dong Xu
Summary: This work proposes an end-to-end optimized learning framework for losslessly compressing 3D volumetric data. The framework utilizes hierarchical compression scheme, intra-slice auxiliary features, and estimation of entropy model based on intra-slice and inter-slice information. Experimental results show that the framework outperforms existing lossless volumetric image codecs in compressing 3D Medical Images and Hyper-Spectral Images.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2022)
Article
Computer Science, Artificial Intelligence
Rui Su, Dong Xu, Luping Zhou, Wanli Ouyang
Summary: The proposed progressive cross-stream cooperation (PCSC) framework improves spatial and temporal action localization by utilizing spatial region and features from one stream to help another stream iteratively generate better bounding boxes. This approach not only enhances the accuracy of action localization, but also improves the prediction of action classes.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2021)
Article
Computer Science, Artificial Intelligence
Junwei Han, Xiwen Yao, Gong Cheng, Xiaoxu Feng, Dong Xu
Summary: This paper proposes a new end-to-end fine-grained visual categorization system called P-CNN, which consists of three modules: SE block for recalibrating feature responses, PLN for locating object parts, and PCN for part classification. The paper also introduces new metric learning and part classification techniques. Experimental results demonstrate the effectiveness of this approach.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Engineering, Electrical & Electronic
Jiayi Tian, Jing Zhang, Wen Li, Dong Xu
Summary: In this work, we propose a novel approach referred to as Virtual Domain Modeling for Domain Adaptation (VDM-DA) to tackle the source data-free unsupervised domain adaptation problem. By generating virtual domain samples and designing an effective distribution alignment method, we successfully achieve the goal of distribution alignment between the source and target domains when training deep networks.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2022)
Article
Engineering, Electrical & Electronic
Jinyang Guo, Jiaheng Liu, Dong Xu
Summary: This paper proposes a fully automatic model compression framework called 3D-Pruning (3DP) for efficient 3D action recognition. The compressed model can process different frames and points based on their importance values, allowing for different computational complexities. Experimental results demonstrate the effectiveness of the 3DP framework for model compression.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2022)
Article
Engineering, Electrical & Electronic
Zongheng Tang, Yue Liao, Si Liu, Guanbin Li, Xiaojie Jin, Hongxu Jiang, Qian Yu, Dong Xu
Summary: In this work, a novel task called Human-centric Spatio-Temporal Video Grounding (HC-STVG) is introduced. HC-STVG aims to localize a spatio-temporal tube of the target person from an untrimmed video based on a given textual description, focusing on humans. This task is useful for healthcare and security applications.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Zhihao Hu, Dong Xu
Summary: In this paper, we propose a combination of complexity-guided slimmable decoder (cgSlimDecoder) and skip-adaptive entropy coding (SaEC) for efficient deep video compression. The cgSlimDecoder automatically determines the optimal channel width for each slimmable convolution layer and allocates the optimal number of parameters for different modules, supporting multiple complexity levels. The SaEC further speeds up the decoding process by skipping the entropy coding for well-predicted elements. Experimental results demonstrate that the proposed methods significantly improve coding efficiency with minimal performance drop.
2023 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)
(2023)
Proceedings Paper
Computer Science, Artificial Intelligence
Chenjian Gao, Qian Yu, Lu Sheng, Yi-Zhe Song, Dong Xu
Summary: This paper introduces a method for reconstructing a 3D shape based on a single sketch image. By analyzing the 3D-to-2D projection process, the density map is used as a proxy to facilitate the reconstruction process. Experimental results show that this method outperforms other baseline methods in both quantitative and qualitative aspects.
COMPUTER VISION - ECCV 2022, PT I
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Ziming Wang, Xiaoliang Huo, Zhenghao Chen, Jing Zhang, Lu Sheng, Dong Xu
Summary: In this work, a new Geometry-Aware Visual Feature Extractor (GAVE) is proposed for point cloud registration, which effectively fuses geometric and visual information for more reliable correspondence estimation, outperforming existing methods.
COMPUTER VISION - ECCV 2022, PT XXXII
(2022)
Article
Computer Science, Artificial Intelligence
Jiaheng Liu, Jinyang Guo, Dong Xu
Summary: This paper investigates the trade-off between accuracy and efficiency in 3D action recognition. A simple and efficient backbone network structure and an adaptive point sampling network are proposed to achieve this trade-off. Experimental results demonstrate the effectiveness and efficiency of the proposed approach for 3D action recognition.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2022)
Article
Computer Science, Artificial Intelligence
Kaisiyuan Wang, Lu Sheng, Shuhang Gu, Dong Xu
Summary: In this work, we propose a new patch-based framework called VPU for video-based point cloud upsampling. By effectively exploiting temporal dependency among multiple consecutive point cloud frames, our method extracts and aggregates rich local geometric clues to infer the local geometry distributions at the current frame. Our framework achieves substantial performance improvement over its single frame-based counterparts on multiple point cloud sequence datasets.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2022)
Article
Computer Science, Artificial Intelligence
Changsheng Li, Handong Ma, Ye Yuan, Guoren Wang, Dong Xu
Summary: The study presents a novel deep unsupervised active learning framework, which can learn nonlinear embedding and select representative samples using a self-supervised learning strategy, preserving both the global and cluster structure, addressing data imbalance issues, and improving model performance.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2022)