A5101565109- Advanced Neural Network Applications
- Video Surveillance and Tracking Methods
- Face and Expression Recognition
- Human Pose and Action Recognition
- CCD and CMOS Imaging Sensors
- Image and Video Stabilization
- Biometric Identification and Security
- Engineering Technology and Methodologies
- Diabetic Foot Ulcer Assessment and Management
- Embedded Systems Design Techniques
- Educational Technology and Assessment
- Advanced Surface Polishing Techniques
- Image and Object Detection Techniques
- Advanced Malware Detection Techniques
- Advanced Image and Video Retrieval Techniques
- Face recognition and analysis
- Surface Roughness and Optical Measurements
- Hand Gesture Recognition Systems
- AI and Multimedia in Education
- Welding Techniques and Residual Stresses
- Industrial Vision Systems and Defect Detection
- Software Engineering Research
- Traditional Chinese Medicine Studies
- Educational Technology and Pedagogy
- Higher Education Learning Practices
Shenzhen Polytechnic
2021-2023
Shanghai Jiao Tong University
2023
Chongqing University
2023
Beijing Microelectronics Technology Institute
2019-2022
Beijing University of Posts and Telecommunications
2020
Urumqi 4th People's Hospital
2020
The Fourth People's Hospital
2020
Jilin Province Science and Technology Department
2019
Jilin University
2019
Heilongjiang University
2015-2017
3D human pose estimation from a monocular RGB image is challenging task in computer vision because of depth ambiguity single image. As most methods consider joint locations independently which can lead to an overfitting problem on specific datasets, it's crucial the plausibility poses terms their overall structures. In this paper, we present Generative Adversarial Networks (GANs) for estimation, learn plausible body representations by adversarial training. GANs, generator regresses positions...
In this paper, a total of 20 sites single nucleotide polymorphisms (SNPs) on the serotonin 3 receptor A gene (HTR3A) and B (HTR3B) are used for feature fusion with age, education marital status information, grid search-support vector machine (GS-SVM), convolutional neural network (CNN) combined long short-term memory (CNN-LSTM) to classify discriminate between alcohol-dependent patients (AD) non-alcohol-dependent control group. The results show that 19 SNPs academic qualifications have best...
Governance includes the state, private sector, and civil society, all of which play an important role in sustained human development. The state constructs a feasible political legal environment, sector creates employment income opportunities, society assists social interaction, mobilizes various forces to participate economic, social, activities. Vocational education is product development, development civilization, can also be said self-development. And it special period. benefits benefit...
According to the survey, off-line examination is still main method in universities, primary and secondary schools. The grading processing of time-consuming. Besides, since subjective, it error-prone. In order address challenges examinations schools, very urgent improve efficiency grading. realize intelligent for examinations, we exploit deep learning techniques First, propose an image English letters. Second, a recognition based on Third, lightweight framework Based above designs, design...
Recently, convolutional neural network (CNN) has been widely implemented in the compute vision, nature language processing and automatic driving. However, it makes much difficulties to employ embedded system because of limit memory storage computation bandwidth. To address those limitations, we explore a two-stage approach compression for scene, object detection. In this paper, first propose an effective pruning on trained network, achieve total 81.86%-91.54% sparse rate with accuracy losing...
In recent years, convolutional neural network (CNN) has become widely universal in large number of applications including computer vision, natural language processing and automatic driving. However, the CNN-based methods are computational-intensive resource-intensive, thus hard to integrate into embedded systems such as smart phones, driving robots. To address limitation, various deep learning accelerators have been proposed implement on field programmable gate array (FPGA) platform, because...
As deep neural networks have been performing better and on various tasks, their number of parameters has increasing, the demand for computing power storage increasing. In 2016, Joseph Redmon et al. proposed a one-stage target detection method: You Only Look Once [1], which widely used worldwide in past 5 years. However, such scale network is not possible to be directly applied mobile devices nowadays [2]. order reduce model, energy consumption improve computational speed, we consider both...
Estimating 3D human poses from a single RGB image is challenging task in computer vision due to depth ambiguity and lack of unconstrained datasets. Poor exploiting cues will lead inaccurate pose predictions. Previous works try constrain illegal spatial relations exploit 2D annotations in-the-wild images as weak labels. In this paper, we propose use multi-scale recalibration with stronger geometry constraints regress pose. The overall network end-to-end which consists estimation sub-network...
Deep learning target detection has always been a major research direction in the field of artificial intelligence. Its results are widely used fields automatic driving, security system and medical treatment. This paper proposes method to improve effect small targets, which realizes objects different scales input image, especially small-scale targets. Before collected data set is sent neural network for training, it first divided into three according size be detected image set. Then one or...
The large scale floating point matrix operations are widely used in numerical analysis, image processing and signal processing. This paper studies the design implementation of complex floating-point multiplication based on high level synthesis (HLS), mainly including operation architecture, large-scale RAM model, random verification platform. complex-floating-point accumulation array is designed by HLS approach, matrix-vector multiply implemented instancing paralleling arrays, chip...