A5100419527- Human Pose and Action Recognition
- Anomaly Detection Techniques and Applications
- Video Surveillance and Tracking Methods
- Advanced Neural Network Applications
- Video Analysis and Summarization
- Multimodal Machine Learning Applications
- Machine Learning and Data Classification
- Gait Recognition and Analysis
- Brain Tumor Detection and Classification
- Machine Learning and ELM
- Mechanical and Optical Resonators
- Immune cells in cancer
- Adversarial Robustness in Machine Learning
- Domain Adaptation and Few-Shot Learning
- CCD and CMOS Imaging Sensors
- Atherosclerosis and Cardiovascular Diseases
- Ideological and Political Education
- Pancreatic and Hepatic Oncology Research
- Imbalanced Data Classification Techniques
- Tensor decomposition and applications
- Advanced Graph Neural Networks
- Renal and related cancers
- Recommender Systems and Techniques
- Healthcare Education and Workforce Issues
- Advancements in Transdermal Drug Delivery
Peking University
2012-2025
Peng Cheng Laboratory
2019-2025
King University
2025
Shanghai Fudan Microelectronics (China)
2024
Fudan University
2024
Beijing Electronic Science and Technology Institute
2023
Bengbu Medical College
2023
Beijing Academy of Artificial Intelligence
2022
University of Chinese Academy of Sciences
2022
Chinese Academy of Sciences
2015-2022
Video anomaly detection under weak labels is formulated as a typical multiple-instance learning problem in previous works. In this paper, we provide new perspective, i.e., supervised task noisy labels. such viewpoint, long cleaning away label noise, can directly apply fully action classifiers to weakly detection, and take maximum advantage of these well-developed classifiers. For purpose, devise graph convolutional network correct Based upon feature similarity temporal consistency, our...
Graph convolutional networks (GCNs) are potentially short of the ability to learn hierarchical representation for graph embedding, which holds them back in classification task. Here, we propose AttPool, is a novel pooling module based on attention mechanism, remedy problem. It able select nodes that significant adaptively, and generate features via aggregating attention-weighted information nodes. Additionally, devise prediction architecture sufficiently leverage facilitate model learning....
In the image super-resolution (SR) field, recovering missing high-frequency textures has always been an important goal. However, deep SR networks based on pixel-level constraints tend to focus stable edge details and cannot effectively restore random textures. It was not until emergence of generative adversarial network (GAN) that GAN-based models achieved realistic texture restoration quickly became mainstream method for SR. still have some drawbacks, such as relying a large number...
Spatial-temporal action detection in videos is a challenging problem that has attracted considerable attention recent years. Most current approaches address as an object problem, which utilizes successful frameworks such Faster R-CNN to operate at every single frame first, and then generates tubes by linking bounding boxes across the whole video offline fashion. However, unlike static images, temporal context information vital for videos. Therefore, we propose online model leverages...
A report is presented on the modelling and verification of microneedle-skin interactions. nonlinear finite element model based micro biomechanical properties skin was established to simulate a microneedle being inserted into skin. The deformation force-displacement behaviour could be obtained. accuracy experimentally verified by measuring relationship between force displacements during insertion mouse With this model, influences different geometries fracture were discussed, which useful...
Currently, an increasing number of model pruning methods are proposed to resolve the contradictions between computer powers required by deep learning models and resource-constrained devices. However, for simple tasks like robotic detection, most traditional rule-based network cannot reach a sufficient compression ratio with low accuracy loss time consuming as well laborious. In this article, we propose automatic blockwise channelwise (ABCP) jointly search action detection reinforcement...
Abstract The low temperature superconducting nanowire single-photon detectors have become a key infrared photon counting technology in communication and astronomy applications. However, the constrained physical space of devices demands for high-performance capable operation at higher temperature. To date, fabrication high superconductor nanowires is still facing seriously uneven lateral damage during ion etching process. In this work, we report promising fabricating method high-temperature...
Motion information is a key factor for action recognition and has been eagerly pursued decades. How to effectively learn motion features in Convolutional Networks (ConvNets) remains an open issue. Prevalent ConvNets often take several full frames of video as input at time, which can be heavy burden network training. In this paper, we introduce novel framework called Tube ConvNets, by substituting tubes reduce burden. focus on the regions interest (ROI) where motions occur, thus eliminate...
Transferring complex computing to the cloud server side leverages cloud-based intelligent service robots that are capable of highly tasks such as video analysis. In practical behavior surveillance applications, captured videos from continuous. Action extraction continuous unconstrained is an important prerequisite for action analysis, classification and recognition, abnormal event detection, crowd emotion sensing. This paper proposes a novel approach in video, which has three parts: spatial...
At present, spatio-temporal action detection in the video is still a challenging problem, considering complexity of background, variety or change viewpoint unconstrained environment. Most current approaches solve problem via two-step processing: first detecting actions at each frame; then linking them, which neglects continuity and operates an offline batch processing manner. In this paper, we attempt to build online model that introduces coherence existed among regions when performing...
Abnormal event detection plays an important role in video surveillance and smart camera systems. Existing methods the literature are usually not object-aware, where different objects distinguished processing. In this work, we propose efficient object-aware anomaly scheme, specifically focusing on certain object categories, such as pedestrians. We first perform a block-based foreground segmentation to confine our analysis moving avoid irrelevant background dynamics. Then discard uninterested...
In this paper, a method is proposed to search for spatio-temporal path action localization in unconstrained videos. We mainly focus on two requirements, i.e., accurate human extraction and speeding generation of proposal. The approach first generates proposals at the frame level, then scores them based complementary parts, posteriori probability evaluated via fine-tuned Faster-RCNN template-matching similarity spatiotemporal continuity. Finally, proposal formulated as Max-Path discovery...
Indium–gallium–zinc oxide (IGZO) as a star material has been broadly applied in multiple functional devices, including planar displays, flexible electronic and photoelectronics. In recent years, the development of artificial intelligence great data also extends application IGZO-based thin film transistors (TFTs) to memory, memristors, neuromorphic computing. Thus, research high performance reliable IGZO TFTs attracts tremendous attention worldwide. Herein, high-quality was deposited via...
Abstract Gathering complete aircraft types for close-set tasks is challenging and costly in fine-grained classification, resulting encountering unknown class images real-world models. To address this problem, we propose a network called multi-reference source classification (MRSN) to explicitly implicitly distinguish known boundaries embed classes into the sample space. Specifically, an embedding module (UCPE) synthesize prototype facilitate modeling of data distribution. Besides, introduce...
Oriented object detection in aerial images is a crucial link earth observation. As special parameter oriented representation, angle the key to achieving high-precision detection. However, widely-used regression-based methods suffer from boundary discontinuity problem due periodicity of angle. To address this issue, we proposed novel prediction method called Fixed Step Trigonometric Coder (FSTC). Exploiting innate trigonometric functions, FSTC can encode angles cyclically succinct,...
Despite tremendous progress achieved in temporal action detection, state-of-the-art methods still suffer from the sharp performance deterioration when localizing starting and ending boundaries. Although most apply boundary regression paradigm to tackle this problem, we argue that direct lacks detailed enough information yield accurate In paper, propose a novel Boundary Likelihood Pinpointing (BLP) network alleviate deficiency of improve localization accuracy. Given loosely localized search...