A5100614868- Video Surveillance and Tracking Methods
- Advanced Neural Network Applications
- Advanced Image and Video Retrieval Techniques
- Human Pose and Action Recognition
- Visual Attention and Saliency Detection
- Advanced Graph Neural Networks
- Infrared Target Detection Methodologies
- Face and Expression Recognition
- Face recognition and analysis
- Robotics and Sensor-Based Localization
- Image Enhancement Techniques
- Anomaly Detection Techniques and Applications
- Remote-Sensing Image Classification
- Gait Recognition and Analysis
- Domain Adaptation and Few-Shot Learning
- Image Retrieval and Classification Techniques
- Advanced Image Fusion Techniques
- Optical Network Technologies
- Advanced Image Processing Techniques
- Photonic and Optical Devices
- Gaze Tracking and Assistive Technology
- Advanced Vision and Imaging
- Image and Signal Denoising Methods
- Neural Networks and Reservoir Computing
- Data Management and Algorithms
Shaanxi Normal University
2023-2024
Anhui University
2019-2024
Nanjing Normal University
2022-2024
Jiangsu Center for Collaborative Innovation in Geographical Information Resource Development and Application
2022-2024
Université de technologie de belfort-montbéliard
2024
Zhengzhou University of Science and Technology
2018-2023
Shanghai Jiao Tong University
2022-2023
Taiyuan University of Technology
2021-2023
Xi'an Jiaotong University
2023
Wuhan National Laboratory for Optoelectronics
2022
Photonic neural networks perform brain-inspired computations using photons instead of electrons to achieve substantially improved computing performance. However, existing architectures can only handle data with regular structures but fail generalize graph-structured beyond Euclidean space. Here, we propose the diffractive graph network (DGNN), an all-optical representation learning architecture based on photonic units (DPUs) and on-chip optical devices address this limitation. Specifically,...
RGB-Thermal object tracking attempts to locate target using complementary visual and thermal infrared data. Existing RGB-T trackers fuse different modalities by robust feature representation learning or adaptive modal weighting. However, how integrate dual attention mechanism for is still a subject that has not been studied yet. In this paper, we propose two mechanisms tracking. Specifically, the local implemented exploiting common of RGB data train deep classifiers. We also introduce global...
Recently, hashing techniques have witnessed an increase in popularity due to their low storage cost and high query speed for large scale data retrieval task, e.g., image retrieval. Many methods been proposed; however, most existing focus on single view data. In many scenarios, there are multiple views samples. Thus, those working can not make full use of rich information contained multi-view Although some proposed data; they usually relax binary constraints or separate the process learning...
Recognizing pedestrian attributes is an important task in the computer vision community due to it plays role video surveillance. Many algorithms have been proposed handle this task. The goal of paper review existing works using traditional methods or based on deep learning networks. Firstly, we introduce background attribute recognition (PAR, for short), including fundamental concepts and corresponding challenges. Secondly, benchmarks, popular datasets evaluation criteria. Thirdly, analyze...
Semantic segmentation is a crucial approach for remote sensing interpretation. High-precision semantic results are obtained at the cost of manually collecting massive pixelwise annotations. Remote imagery contains complex and variable ground objects obtaining abundant manual annotations expensive arduous. The semi-supervised learning (SSL) strategy can enhance generalization capability model with small number labeled samples. In this study, novel adversarial network developed information...
Over-smoothing has emerged as a severe obstacle to node classification with message passing based graph convolutional networks (GCNs). Classification performance dramatically deteriorates for deep GCNs, over the observed noisy topology cannot adequately propagate intra-class information and over-mix features of nodes from different communities (classes). Existing optimization methods GCNs sufficiently exploit underlying ground-truth community structure distinguish communities. In this paper,...
Developments of video processing technology make it much easier to tamper with video. In some situation, such as in a lawsuit, is necessary prove videos are not tampered. This contradiction poses challenges ascertain integrity digital videos. Most tamperings occur pixel domain. However, nowadays usually stored compressed format, H.264/AVC. For attackers decompress original bitstreams and recompress into As result, by detecting double compression, we can authenticate this paper, propose an...
Message passing has evolved as an effective tool for designing graph neural networks (GNNs). However, most existing methods message simply sum or average all the neighboring features to update node representations. They are restricted by two problems: 1) lack of interpretability identify significant prediction GNNs and 2) feature overmixing that leads oversmoothing issue in capturing long-range dependencies inability handle graphs under heterophily low homophily. In this article, we propose...
The tracking-by-detection framework requires a set of positive and negative training samples to learn robust tracking models for precise localization target objects. However, existing mostly treat different independently while ignores the relationship information among them. In this paper, we propose novel structure-aware deep neural network overcome such limitations. particular, construct graph represent pairwise relationships samples, additionally take natural language as supervised both...
Recently, hashing based approximate nearest neighbor search has attracted much attention in large scale data task. Moreover, some cross-modal methods have also been proposed to perform efficient of different modalities. However, there are still problems be further considered. For example, them cannot make use label information, which contains helpful information generate hash codes; firstly relax binary constraints during optimization, then threshold continuous outputs binary, could...
Small scale face detection is a very difficult problem. In order to achieve higher accuracy, we propose novel method, termed SE-IYOLOV3, for small in this work. improve the YOLOV3 first, which anchorage box with average intersection ratio obtained by combining niche technology on basis of k-means algorithm. An upsampling added form network structure that suitable detecting dense faces. The number prediction boxes five times more than network. To further performance, adopt SENet enhance...