A5082405818- Advanced Neural Network Applications
- Video Surveillance and Tracking Methods
- Advanced Image and Video Retrieval Techniques
- Advanced Mathematical Modeling in Engineering
- Advanced Numerical Methods in Computational Mathematics
- Face recognition and analysis
- Composite Material Mechanics
- Face and Expression Recognition
- Visual Attention and Saliency Detection
- Advanced Thermodynamic Systems and Engines
- Software Reliability and Analysis Research
- Refrigeration and Air Conditioning Technologies
- Force Microscopy Techniques and Applications
- Advanced Text Analysis Techniques
- Sentiment Analysis and Opinion Mining
- Forensic Anthropology and Bioarchaeology Studies
- Spacecraft and Cryogenic Technologies
- Multimodal Machine Learning Applications
- Human Pose and Action Recognition
- Advanced Graph Neural Networks
- Time Series Analysis and Forecasting
- Thermal properties of materials
- Digital Media Forensic Detection
- Infrared Target Detection Methodologies
- Anomaly Detection Techniques and Applications
China United Network Communications Group (China)
2022-2024
George Institute for Global Health
2023
Public Health Clinical Center of Chengdu
2023
Xi'an Honghui Hospital
2021-2022
Children's Hospital of Fudan University
2022
Southern Medical University
2021
GFZ Helmholtz Centre for Geosciences
2019
Inha University
2018
Harbin Institute of Technology
2017
Group Sense (China)
2016-2017
Deep learning methods are powerful tools but often suffer from expensive computation and limited flexibility. An alternative is to combine light-weight models with deep representations. As successful cases exist in several visual problems, a unified framework absent. In this paper, we revisit two widely used approaches computer vision, namely filtered channel features Convolutional Neural Networks (CNN), absorb merits both by proposing an integrated method called Channel Features (CCF). CCF...
Face detection has drawn much attention in recent decades since the seminal work by Viola and Jones. While many subsequences have improved with more powerful learning algorithms, feature representation used for face still can't meet demand effectively efficiently handling faces large appearance variance wild. To solve this bottleneck, we borrow concept of channel features to domain, which extends image diverse types like gradient magnitude oriented histograms therefore encodes rich...
In recent years, collaborative classification of multimodal data, e.g., hyperspectral image (HSI) and light detection ranging (LiDAR), has been widely used to improve remote sensing accuracy. However, existing fusion approaches for HSI LiDAR suffer from limitations. Fusing the heterogeneous features proved be challenging, leading incomplete utilization information category representation. Additionally, during extraction spatial HSI, spectral are often disjointed. It leads difficulty fully...
The visual cues from multiple support regions of different sizes and resolutions are complementary in classifying a candidate box object detection. Effective integration local contextual these has become fundamental problem In this paper, we propose gated bi-directional CNN (GBD-Net) to pass messages among features during both feature learning extraction. Such message passing can be implemented through convolution between neighboring two directions conducted various layers. Therefore,...
Object detection is a fundamental problem in image understanding. One popular solution the R-CNN framework [15] and its fast versions [14, 27]. They decompose object into two cascaded easier tasks: 1) generating proposals from images, 2) classifying various categories. Despite that we are handling with relatively tasks, they not solved perfectly there's still room for improvement. In this paper, push "divide conquer" even further by dividing each task sub-tasks. We call proposed method...
In this paper, we propose a multi-level attention model to solve the weakly labelled audio classification problem. The objective of is predict presence or absence events in an clip. Recently, Google published large scale dataset called Audio Set, where each clip contains only events, without onset and offset time events. Our extension previously proposed single-level model. It consists several modules applied on intermediate neural network layers. output these are concatenated vector...
Current evaluation datasets for face detection, which is of great value in real-world applications, are still somewhat out-of-date. We propose a new detection dataset MALF (short Multi-Attribute Labelled Faces), contains 5,250 images collected from the Internet and ∼12,000 labelled faces. The highlights two main features: 1) It largest wild, annotation multiple facial attributes makes it possible fine-grained performance analysis. 2) To reveal ‘true’ performances algorithms practice, adopts...
Fracture is one of the most common and unexpected traumas. If not treated in time, it may cause serious consequences such as joint stiffness, traumatic arthritis, nerve injury. Using computer vision technology to detect fractures can reduce workload misdiagnosis also improve fracture detection speed. However, there are still some problems sternum detection, low rate small occult fractures. In this work, authors have constructed a dataset with 1227 labelled X-ray images for detection. The...
The visual cues from multiple support regions of different sizes and resolutions are complementary in classifying a candidate box object detection. Effective integration local contextual these has become fundamental problem In this paper, we propose gated bi-directional CNN (GBD-Net) to pass messages among features during both feature learning extraction. Such message passing can be implemented through convolution between neighboring two directions conducted various layers. Therefore,...
The goal of multiple object tracking (MOT) is to estimate the locations objects and maintain their identities consistently yield individual trajectories. MOT has been developed enormously, but it still a challenging work due similar appearances different occlusion by other or background in complex scene. In this study, authors propose confidence score‐based appearance model learning hierarchical data association for MOT. First, score used divide associated tracklet‐detection first stage into...
Graph convolutional neural networks~(GCNs) have recently demonstrated promising results on graph-based semi-supervised classification, but little work has been done to explore their theoretical properties. Recently, several deep networks, e.g., fully connected and with infinite hidden units proved be equivalent Gaussian processes~(GPs). To exploit both the powerful representational capacity of GCNs great expressive power GPs, we investigate similar properties infinitely wide GCNs. More...
In recent years, convolutional neural networks (CNNs) have been widely used for visual object tracking, especially in combination with correlation filters (CFs). However, the increasing complex CNN models introduce more useless information, which may decrease tracking performance. This study proposes an online feature map selection method to remove noisy and irrelevant maps from different layers of CNN, can reduce computation redundancy improve accuracy. Furthermore, a novel appearance model...