A5088135786- Video Surveillance and Tracking Methods
- Human Pose and Action Recognition
- Gait Recognition and Analysis
- Face recognition and analysis
- Computer Graphics and Visualization Techniques
- Advanced Vision and Imaging
- Human Motion and Animation
- Interactive and Immersive Displays
- 3D Surveying and Cultural Heritage
- Gastrointestinal motility and disorders
- Gut microbiota and health
- Video Analysis and Summarization
- Data Visualization and Analytics
- Advanced Neural Network Applications
- Eating Disorders and Behaviors
- Memory and Neural Mechanisms
- Neuroscience and Neuropharmacology Research
- Receptor Mechanisms and Signaling
- Image Enhancement Techniques
Shanghai Jiao Tong University
2022-2025
Hefei National Center for Physical Sciences at Nanoscale
2023
University of Science and Technology of China
2022-2023
Beijing Institute of Technology
2013-2014
Singapore Management University
2013
Person re-identification (ReID) plays an important role in applications such as public security and video surveillance. Recently, learning from synthetic data [ 9 ], which benefits the popularity of engine, has attracted great attention public. However, existing datasets are limited quantity, diversity, realisticity, cannot be efficiently used for ReID problem. To address this challenge, we manually construct a large-scale person dataset named FineGPR with fine-grained attribute annotations....
Generalizable person re-identification (Re-ID) is a very hot research topic in machine learning and computer vision, which plays significant role realistic scenarios due to its various applications public security video surveillance. However, previous methods mainly focus on the visual representation learning, while neglect explore potential of semantic features during training, easily leads poor generalization capability when adapted new domain. In this paper, we present unified perspective...
As a fundamental problem in video surveillance, person re-identification (re-ID) contributes lot to the development of modern metro city. Recently, learning from synthetic data on re-ID task, which benefits popularity engine, has achieved remarkable performance both supervised and unsupervised manner. However, previous researches mainly lay emphasis employing achieve state-of-the-art with strong backbone, while neglects perform quantitative studies how visual factors affect system. To...
Generalizable person re-identification (Re-ID) is a very hot research topic in machine learning and computer vision, which plays significant role realistic scenarios due to its various applications public security video surveillance. However, previous methods mainly focus on the visual representation learning, while neglect explore potential of semantic features during training, easily leads poor generalization capability when adapted new domain. In this paper, we propose Multi-Modal...
Large-scale vector data produce the field clustering in flow visualization. To emphasize essential features, a new method for 2D fields is proposed this paper. With method, firstly initialized as cluster, which then iteratively divided into hierarchy of clusters. During iteration, clusters are segmented with streamlines instead straight lines. This change enables it to since consistent behaviors, and shaped by aligned underlying flow. It easy capture patterns features from resulting...
Person re-identification (re-ID) plays an important role in applications such as public security and video surveillance. Recently, learning from synthetic data, which benefits the popularity of data engine, has attracted great attention eyes. However, existing datasets are limited quantity, diversity realisticity, cannot be efficiently used for re-ID problem. To address this challenge, we manually construct a large-scale person dataset named FineGPR with fine-grained attribute annotations....
Illumination is an important way to provide cues for spatial perception, and special lighting models are proposed illuminate lines in 3D space. To achieve better perception of streamlines, a new streamline illumination method this paper, which based on the fact that sample stream surface. In our method, view-dependent tiny surface passing through chosen compute line's Phong/Blinn model. Compared with existing line models, more simple realistic. Test results show it can enhance streamlines...
Unsupervised Domain Adaptation (UDA) Person reidentification (ReID) strives towards fine-tuning the model trained on a labelled source-domain dataset to target-domain dataset, which has grown by leaps and bounds due advancement of deep convolution neural network (CNN). However, traditional CNN-based methods mainly focus learning small discriminative features in local pedestrian region, fails exploit potential rich structural patterns suffers from information loss details caused operators. To...
Pretraining is a dominant paradigm in computer vision. Generally, supervised ImageNet pretraining commonly used to initialize the backbones of person re-identification (Re-ID) models. However, recent works show surprising result that CNN-based on has limited impacts Re-ID system due large domain gap between and data. To seek an alternative traditional pretraining, here we investigate semantic-based as another method utilize additional textual data against pretraining. Specifically, manually...
Employing clustering strategy to assign unlabeled target images with pseudo labels has become a trend for person re-identification (re-ID) algorithms in domain adaptation. A potential limitation of these clustering-based methods is that they always tend introduce noisy labels, which will undoubtedly hamper the performance our re-ID system. To handle this limitation, an intuitive solution utilize collaborative training purify label quality. However, there exists challenge complementarity two...
Animation sketching tools have been shown to make creating animations fast and easy, but editing an animation can it more complex harder modify. To preserve simplicity flexibility while animations, we explored the use of flexible grouping structure multiple centers for rotation scaling. We built a modified version K-Sketch system, which allows center scaling change over time. The preserves when by automatically converting existing motions about old new center. evaluated our method examining...