A5114910951- Video Surveillance and Tracking Methods
- Advanced Neural Network Applications
- Particle physics theoretical and experimental studies
- Quantum Chromodynamics and Particle Interactions
- UAV Applications and Optimization
- Anomaly Detection Techniques and Applications
- Advanced Vision and Imaging
- Optical measurement and interference techniques
- Human Mobility and Location-Based Analysis
- Image Processing and 3D Reconstruction
- Image Enhancement Techniques
- Image Processing Techniques and Applications
- 3D Shape Modeling and Analysis
- Robotics and Sensor-Based Localization
- High-Energy Particle Collisions Research
- Multimodal Machine Learning Applications
- Impact of Light on Environment and Health
- Video Analysis and Summarization
- Black Holes and Theoretical Physics
- Domain Adaptation and Few-Shot Learning
- Myasthenia Gravis and Thymoma
- Traffic Prediction and Management Techniques
- Groundwater flow and contamination studies
- Human Pose and Action Recognition
- Brain Tumor Detection and Classification
University of Chinese Academy of Sciences
2018-2024
Nankai University
2024
Institute of High Energy Physics
2020-2024
Beijing University of Technology
2024
Shanghai University
2024
Affiliated Hospital of Nantong University
2023
Nantong University
2023
Jilin University
2023
Xi'an Institute of Optics and Precision Mechanics
2022
Dalian University of Technology
2022
One of the critical challenges object counting is dramatic scale variations, which introduced by arbitrary perspectives. We propose a reverse perspective network to solve variations input images, instead generating maps smooth final outputs. The explicitly evaluates distortions, and efficiently corrects distortions uniformly warping images. Then proposed delivers images with similar instance scales regressor. Thus regression doesn't need multi-scale receptive fields match various scales....
By detecting the defect location in high-resolution insulator images collected by unmanned aerial vehicle (UAV) various environments, occurrence of power failure can be timely detected and caused economic loss reduced. However, accuracies existing detection methods are greatly limited complex background interference small target detection. To solve this problem, two deep learning based on Faster R-CNN (faster region-based convolutional neural network) proposed paper, namely Exact (exact...
The crowd counting is challenging for deep networks due to several factors. For instance, the can not efficiently analyze perspective information of arbitrary scenes, and they are naturally inefficient handle scale variations. In this work, we deliver a simple yet efficient multi-column network, which integrates analysis method with network. proposed explicitly excavates drives network scenes. More concretely, explore from estimated density maps quantify space into separate We then embed...
Crowd counting is a difficult task because of the diversity scenes. Most existing crowd methods adopt complex structures with massive backbones to enhance generalization ability. Unfortunately, performance on large-scale data sets not satisfactory. In order handle various scenarios less network, we explored how efficiently use multi-expert model for tasks. We mainly focus train more efficient expert networks and choose most suitable expert. Specifically, propose task-driven similarity metric...
Biochar is a widely used material for the remediation of contaminated soils. However, recovery biochar after its application in soil remains significant challenge. To regulate migration particles and identify possible strategy from soil, this study investigated transport behavior microparticles (MPs) nanoparticles (NPs) real under different potential gradients (0, 0.5 1.0·V cm-1) humic acid (HA) concentrations 5 10 mg·L-1). It was demonstrated that electric field presence HA could enhance...
Convolutional neural networks (CNNs) have been widely adopted in the visual tracking community, significantly improving state-of-the-art. However, most of them ignore important cues lying distribution training data and high-level features that are tightly coupled with target/background classification. In this paper, we propose to improve accuracy via online training. On one hand, squeeze redundant by analyzing dataset low-level feature space. other design statistic-based losses increase...
Position and orientation (PO) estimation with microscopic vision is essential for various micromanipulation tasks. Herein, to improve the accuracy flexibility of PO estimation, a generic algorithm proposed based on Discriminative Correlation Filter (DCF), in which position-estimator an orientation-estimator are combined into one framework developed mutual correction mechanism. The extraction spectral features utilized decouple rotation translation transformations target. And DCF employed...
With the advantage of high mobility, Unmanned Aerial Vehicles (UAVs) are used to fuel numerous important applications in computer vision, delivering more efficiency and convenience than surveillance cameras with fixed camera angle, scale view. However, very limited UAV datasets proposed, they focus only on a specific task such as visual tracking or object detection relatively constrained scenarios. Consequently, it is great importance develop an unconstrained benchmark boost related...
6D object pose estimation holds essential roles in various fields, particularly the grasping of industrial workpieces. Given challenges like rust, high reflectivity, and absent textures, this paper introduces a point cloud based framework (PS6D). PS6D centers on slender multi-symmetric objects. It extracts multi-scale features through an attention-guided feature extraction module, designs symmetry-aware rotation loss center distance sensitive translation to regress each centroid instance,...
Typical dynamic ST data includes trajectory (representing individual-level mobility) and traffic state population-level mobility). Traditional studies often treat as distinct, independent modalities, each tailored to specific tasks within a single modality. However, real-world applications, such navigation apps, require joint analysis of data. Treating these types two separate domains can lead suboptimal model performance. Although recent advances in pre-training foundation models aim...
Early diagnosis of melanoma can substantially increase patient survival rate. Currently, dermoscopy is the dominant approach for clinical detection, but this method requires interaction with a trained professional resulting in financial burden which major limiting factor many patients, especially those remote and rural locations. It has been proposed that deep convolutional neural networks (CNNs) could allow an automated approaches melanoma. However, there limited work regarding use CNNs to...
With the optimization and upgrading of ARM architecture popularization mobile smart devices, application requirements deep learning technology in real life are increasing. It is a big problem to run integrated neural network model on platform. In this paper, aiming at problems existing deployment target detection lightweight terminal, improved convolution parallel computing optimized, an effective inference acceleration scheme proposed. YOLOv3 algorithm based one-stage used for multi-scale...
The task of video captioning is to generate comprehensible and grammatically correct sentences which describe the main visual content videos. Existing neural modules based methods improve model interpretability by separately predicting words different part-of-speech. However, separation may lead confusing semantics. In this work, a method referred as Differentiate Visual Features with Guidance Signals (DVFGS) proposed, enhances semantic consistency through guidance signals. This process...
Various micromanipulations require the assessment of target's position and orientation (PO) under microscopic vision. Here a universal method based on multi-feature fusion discrimination correlation filtering (DCF) is suggested to improve accuracy PO estimation. A continuous convolution operator first introduced achieve sub-pixel localization for high-precision location Then, by decoupling translation rotation with Fourier-Mellin transformation, information expressed using features...