- Advanced Image and Video Retrieval Techniques
- Image Retrieval and Classification Techniques
- Video Analysis and Summarization
- Text and Document Classification Technologies
- Machine Learning and Data Classification
- Topic Modeling
- Music and Audio Processing
- Face recognition and analysis
- Natural Language Processing Techniques
- Face and Expression Recognition
- Advanced Text Analysis Techniques
- Video Surveillance and Tracking Methods
- Multimodal Machine Learning Applications
- Digital Marketing and Social Media
- Domain Adaptation and Few-Shot Learning
- Algorithms and Data Compression
- Information Retrieval and Search Behavior
- Machine Learning and ELM
- Neural Networks and Applications
- Human Pose and Action Recognition
- Sentiment Analysis and Opinion Mining
- Complex Network Analysis Techniques
- Machine Learning and Algorithms
- Speech Recognition and Synthesis
- Glaucoma and retinal disorders
Huazhong University of Science and Technology
1994-2025
Shanghai University of Finance and Economics
2024
Tongji Hospital
2021-2023
Union Hospital
2022
Anhui Medical University
2020-2021
Wuhan Union Hospital
2020
Inner Mongolia University
2007-2020
Zhengzhou University
2018
Tsinghua University
2016
Snap (United States)
2015
Many multimedia applications can benefit from techniques for adapting existing classifiers to data with different distributions. One example is cross-domain video concept detection which aims adapt across various domains. In this paper, we explore two key problems classifier adaptation: (1) how transform classifier(s) into an effective a new dataset that only has limited number of labeled examples, and (2) select the best adaptation. For first problem, propose Adaptive Support Vector...
Social advertising uses information about consumers' peers, including peer affiliations with a brand, product, organization, etc., to target ads and contextualize their display. This approach can increase ad efficacy for two main reasons: peers' reflect unobserved consumer characteristics, which are correlated along the social network; inclusion of cues (i.e., association brand) alongside affect responses via influence processes. For these reasons, may be increased when multiple signals...
This paper is concerned with the problem of mining social emotions from text. Recently, fast development web 2.0, more and documents are assigned by users emotion labels such as happiness, sadness, surprise. Such can provide a new aspect for document categorization, therefore help online to select related based on their emotional preferences. Useful it is, ratio manual still very tiny comparing huge amount web/enterprise documents. In this paper, we aim discover connections between affective...
Many data mining applications can benefit from adapt- ing existing classifiers to new with shifted distribu- tions. In this paper, we present Adaptive Support Vector Machine (Adapt-SVM) as an efficient model for adapting a SVM classifier trained one dataset where only limited labeled examples are available. By in- troducing regularizer into SVM's objective function, Adapt-SVM aims minimize both the classification error over training examples, and discrepancy between adapted original...
A number of researchers have been building high-level semantic concept detectors such as outdoors, face, building, etc., to help with video retrieval. Using the TRECVID collection and LSCOM truth annotations from 300 concepts, we simulate performance retrieval under different assumptions detection accuracy. Even low accuracy provides good results, when sufficiently many concepts are used. Considering this extrapolation reasonable assumptions, paper arrives at conclusion that "concept-based"...
Video information retrieval requires a system to find relevant query which may be represented simultaneously in different ways through text description, audio, still images and/or video sequences. We present novel approach that uses pseudo-relevance feedback from retrieved items are NOT similar the without further inquiring user feedback. provide insight into this using statistical model and suggest score combination scheme via posterior probability estimation. An evaluation on 2002 TREC...
An approach using many intermediate semantic concepts is proposed with the potential to bridge gap between what a color, shape, and texture-based ldquolow-levelrdquo image analysis can extract from video users really want find, most likely text descriptions of their information needs. Semantic such as cars, planes, roads, people, animals, different types scenes (outdoor, night time, etc.) be automatically detected in reasonable accuracy. This leads us ask how they used does user (or...
This paper is concerned with the problem of social affective text mining, which aims to discover connections between emotions and terms based on user-generated emotion labels. We propose a joint emotion-topic model by augmenting latent Dirichlet allocation an additional layer for modeling. It first generates set topics from emotions, followed generating each topic. Experimental results online news collection show that proposed can effectively identify meaningful emotion. Evaluation...
This paper presents a system for protecting the privacy of specific individuals in video recordings. We address following two problems: automatic people identification with limited labeled data, and human body obscuring preserved structure motion information. In order to first problem, we propose new discriminative learning algorithm improve accuracy using training data from original imperfect pairwise constraints face obscured data. employ robust detection tracking obscure faces video. Our...
Labeling faces in news video with their names is an interesting research problem which was previously solved using supervised methods that demand significant user efforts on labeling training data. In this paper, we investigate a more challenging setting of the where there no complete information data labels. Specifically, by exploiting uniqueness face's name, formulate as special multi-instance learning (MIL) problem, namely exclusive MIL or eMIL so it can be tackled model trained partial...
Pervasive activity monitoring in a skilled-nursing facility helps capture continuous audio and video record. The CareMedia project analyzes this information by automatically tracking people, helping to efficiently label individuals, characterizing selected activities actions.
The first VideOlympics brings content-based analysis to the archive and allows for many-to- many communication between video search engines their audience It was a great Success. provided excitement of competition without associated stress on participants. For time, able compare different multimedia retrieval systems same tasks see how they performed with unrehearsed topics. Many members felt understood technology's capabilities after seeing it in live action several system variations.
Abstract Laser communication technology with characteristics of large capacity, high rate, low power consumption, and strong anti-jamming ability, shows great advantage on satellite communication. Meanwhile, the stability uniformity indexes temperature laser antenna must be strictly controlled in order to ensure duration quality. This paper proposes an expanded configuration layout heat radiator + pipe type thermal control design scheme, takes a DFH-4E platform terminal scheme as example,...
Face-based Voice Conversion (FVC) is a novel task that leverages facial images to generate the target speaker's voice style. Previous work has two shortcomings: (1) suffering from obtaining embeddings are well-aligned with identity information, and (2) inadequacy in decoupling content speaker information audio input. To address these issues, we present FVC method, Identity-Disentanglement (ID-FaceVC), which overcomes above limitations. More precisely, propose an Identity-Aware Query-based...
Abstract Objectives To develop an occupational exposure risk assessment scale for nursing staff during major public health emergencies based on the Likelihood Exposure Consequence (LEC) method. The purpose is to provide managers with a reliable tool assessing faced by and serve as reference formulation of protection standards. Methods item pool factors was screened using LEC accident cause theory. This achieved through comprehensive literature review, semi-structured interviews, group...
We aim to investigate the prevalence and associated factors for compassion fatigue among nurses in Fangcang Shelter Hospitals Wuhan. Studies have shown that was more common than other health-care providers, its predictors were also different. In recent years, most studies investigated emergency oncology nurses, whereas there is little information on from frontline of during COVID-19 pandemic.A descriptive, cross-sectional design used this study. An online survey conducted (n = 972) five...
The Carnegie Mellon University Informedia group has enjoyed consistent success with TRECVID interactive search using traditional storyboard interfaces for shot-based retrieval. For 2006 the output of automatic was included first time storyboards, both as an option user and in a different run sole means access. makes use relevance-based probabilistic retrieval models to determine weights combining sources when addressing given topic. Storyboard-based access outperformed extreme video manual...
In this paper we introduce a learning approach to improve the efficiency of manual image annotation. Although important in practice, annotation has rarely been studied quantitative way. We propose formal models characterize times for two commonly used approaches, i.e., tagging and browsing. The make clear complementary properties these inspire learning-based hybrid algorithm. Our experiments show that proposed algorithm can achieve up 50% reduction time over baseline methods.