A5101539923- Advanced Image and Video Retrieval Techniques
- Advanced Neural Network Applications
- Domain Adaptation and Few-Shot Learning
- Multimodal Machine Learning Applications
- COVID-19 diagnosis using AI
- Image Retrieval and Classification Techniques
- Radiomics and Machine Learning in Medical Imaging
- Robotics and Sensor-Based Localization
- Video Surveillance and Tracking Methods
- Image Processing Techniques and Applications
- Advanced X-ray and CT Imaging
- Adversarial Robustness in Machine Learning
- Gait Recognition and Analysis
- Retinal Diseases and Treatments
- Advanced Vision and Imaging
- Robotic Path Planning Algorithms
- Chronic Obstructive Pulmonary Disease (COPD) Research
- Remote Sensing and LiDAR Applications
- Aeolian processes and effects
- Error Correcting Code Techniques
- Glaucoma and retinal disorders
- Visual Attention and Saliency Detection
- Hand Gesture Recognition Systems
- Retinal Imaging and Analysis
- Human Pose and Action Recognition
Beijing Normal University
2024
Center for High Pressure Science & Technology Advanced Research
2024
Yunnan University
2020-2021
Ningxia Center for Diseases Prevention and Control
2020
Sinosteel (China)
2020
The University of Adelaide
2015-2019
University of Nebraska–Lincoln
2014-2016
Institute of Automation
2013-2016
Chinese Academy of Sciences
2012-2016
Universität der Bundeswehr München
2015
This paper studies an approach to gait based human identification via similarity learning by deep convolutional neural networks (CNNs). With a pretty small group of labeled multi-view walking videos, we can train recognize the most discriminative changes patterns which suggest change identity. To best our knowledge, this is first work on CNNs for recognition in literature. Here, provide extensive empirical evaluation terms various scenarios, namely, cross-view and cross-walking-condition,...
Depth estimation from single monocular images is a key component in scene understanding. Most existing algorithms formulate depth as regression problem due to the continuous property of depths. However, value input data can hardly be regressed exactly ground-truth value. In this paper, we propose pixelwise classification task. Specifically, first discretize depths into several bins and label according their ranges. Then, solve by training fully convolutional deep residual network. Compared...
The trend towards increasingly deep neural networks has been driven by a general observation that increasing depth increases the performance of network. Recently, however, evidence amassing simply may not be best way to increase performance, particularly given other limitations. Investigations into residual have also suggested they in fact operating as single network, but rather an ensemble many relatively shallow networks. We examine these issues, and doing so arrive at new interpretation...
Image classification is a hot topic in computer vision and pattern recognition. Feature coding, as key component of image classification, has been widely studied over the past several years, number coding algorithms have proposed. However, there no comprehensive study concerning connections between different methods, especially how they evolved. In this paper, we first make survey on various feature including their motivations mathematical representations, then exploit relations, based which...
We propose an approach to instance-level image segmentation that is built on top of category-level segmentation. Specifically, for each pixel in a semantic category mask, its corresponding instance bounding box predicted using deep fully convolutional regression network. Thus it follows different pipeline the popular detect-then-segment approaches first predict instances' boxes, which are current state-of-the-art show that, by leveraging strength our models, proposed method can achieve...
This paper proposes to learn features from sets of labeled raw images. With this method, the problem over-fitting can be effectively suppressed, so that deep CNNs trained scratch with a small number training data, i.e., 420 albums about 30 000 photos. method deal images, no matter if bear temporal structures. A typical approach sequential image analysis usually leverages motions between adjacent frames, while proposed focuses on capturing co-occurrences and frequencies features....
We propose a method for high-performance semantic image segmentation (or pixel labelling) based on very deep residual networks, which achieves the state-of-the-art performance. A few design factors are carefully considered to this end. make following contributions. (i) First, we evaluate different variations of fully convolutional network so as find best configuration, including number layers, resolution feature maps, and size field-of-view. Our experiments show that further enlarging...
Unsupervised domain adaptation (UDA) in semantic segmentation is a fundamental yet promising task relieving the need for laborious annotation works. However, shifts/discrepancies problem this compromise final performance. Based on our observation, main causes of shifts are differences imaging conditions, called image-level shifts, and object category configurations category-level shifts. In paper, we propose novel UDA pipeline that unifies alignment feature distribution regularization...
We propose an approach to semantic (image) segmentation that reduces the computational costs by a factor of 25 with limited impact on quality results. Semantic has number practical applications, and for most such applications are critical. The method follows typical two-column network structure, where one column accepts input image, while other half-resolution version image. By identifying specific regions in full-resolution image can be safely ignored, as well carefully tailoring we process...
We propose a foreground segmentation method based on convolutional networks. To predict the label of pixel in an image, model takes hierarchical context as input, which is obtained by combining multiple patches different scales. Short range contexts depict local details, while long capture object-scene relationships image. Early means that we combine into one before any trainable layers are learned, i.e., early-combing. In contrast, late-combing combination occurs later, e.g., when feature...
To reduce the rate of collision accidents between mobile cranes and immobile obstacles in construction sites during lift operations, this paper develops a real-time automated anticollision system that can warn crane operators about potential collisions automatically implement collision-avoidance strategies. This does not require additional devices be installed existing controllers. Before operation, location shape data all objects work area are collected by boom head stored track-sector...
Depth estimation from single monocular images is a key component of scene understanding and has benefited largely deep convolutional neural networks (CNN) recently. In this article, we take advantage the recent residual propose simple yet effective approach to problem. We formulate depth as pixel-wise classification task. Specifically, first discretize continuous values into multiple bins label according their range. Then train fully predict each pixel. Performing discrete instead value...
Objective: To assess the performance of a novel deep learning (DL)-based artificial intelligence (AI) system in classifying computed tomography (CT) scans pneumonia patients into different groups, as well to present an effective clinically relevant machine (ML) based on medical image identification and clinical feature interpretation assist radiologists triage diagnosis. Methods: The 3,463 CT images used this multi-center retrospective study were divided four categories: bacterial ( n =...
Visual interpretation is extremely important in human geography as the primary technique for geographers to use photograph data identifying, classifying, and quantifying geographic topological objects or regions. However, it also time-consuming requires overwhelming manual effort from professional geographers. This paper describes our interdisciplinary team's efforts integrating computer vision models with geographers' visual image process reduce their workload interpreting images. Focusing...
Background Given the high prevalence of fibrotic interstitial lung abnormalities (ILAs) post-COVID-19, this study aims to evaluate effectiveness quantitative CT features in predicting ILAs at 3-month follow-up. Methods This retrospective utilized cohorts from distinct clinical settings: training dataset comprised individuals presenting fever clinic and emergency department, while validation included patients hospitalized with COVID-19 pneumonia. They were classified into group nonfibrotic...
Ensuring the safety and efficiency of crane operation is challenging due to complexity lifting operation. The real-time path planning system developed in this paper aims provide an optimized, collision-free for mobile operators. first contribution take advantage mounted sensors components as hardware collect object information. No additional device needs be purchased. Secondly, data storage, planning, optimizing, visualizing functions are designed minimize required computer memory so that...