- Video Surveillance and Tracking Methods
- Face recognition and analysis
- Gait Recognition and Analysis
- Domain Adaptation and Few-Shot Learning
- Multimodal Machine Learning Applications
- Time Series Analysis and Forecasting
- Human Pose and Action Recognition
- Business Process Modeling and Analysis
- Software System Performance and Reliability
- Context-Aware Activity Recognition Systems
- Topic Modeling
- Mobile Health and mHealth Applications
- Advanced Text Analysis Techniques
- Digital Platforms and Economics
- Advanced Image and Video Retrieval Techniques
- COVID-19 diagnosis using AI
- Multimedia Communication and Technology
- Educational Systems and Policies
- Robotic Mechanisms and Dynamics
- Robotic Path Planning Algorithms
- Complex Network Analysis Techniques
- ERP Systems Implementation and Impact
- Cloud Computing and Resource Management
- Data Visualization and Analytics
- Digital Mental Health Interventions
Soonchunhyang University
2023-2024
Yonsei University
2021-2023
Gyeongsang National University
2010
Video-based person re-identification (reID) aims to retrieve videos with the same identity as a query across multiple cameras. Spatial and temporal distractors in videos, such background clutter partial occlusions over frames, respectively, make this task much more challenging than image-based reID. We observe that spatial appear consistently particular location, show several patterns, e.g., occur first few where patterns provide informative cues for predicting which frames focus on (i.e.,...
We present a novel unsupervised domain adaption method for person re-identification (reID) that generalizes model trained on labeled source to an unlabeled target domain. introduce camera-driven curriculum learning (CaCL) framework leverages camera labels of images transfer knowledge from domains progressively. To this end, we divide dataset into multiple subsets based the labels, and initially train our with single subset (i.e., captured by camera). then gradually exploit more training,...
Time series data is essential in various applications, including climate modeling, healthcare monitoring, and financial analytics. Understanding the contextual information associated with real-world time often for accurate reliable event predictions. In this paper, we introduce TimeCAP, a time-series processing framework that creatively employs Large Language Models (LLMs) as contextualizers of data, extending their typical usage predictors. TimeCAP incorporates two independent LLM agents:...
AIoT-based smart healthcare system that utilize devices can provide personalized, proactive care to patients, reducing the reliance on limited medical service capacities and promoting everyday health management. However, lack of compatibility in data formats services across from different manufacturers hinders growth services. To address this issue, we propose a unified for collection raw sensor (i.e., smartphones, smartwatches) various as well survey data. The application part our is...
Time series data is essential in various applications, including climate modeling, healthcare monitoring, and financial analytics. Understanding the contextual information associated with real-world time often for accurate reliable event predictions. In this paper, we introduce TimeCAP, a time-series processing framework that creatively employs Large Language Models (LLMs) as contextualizers of data, extending their typical usage predictors. TimeCAP incorporates two independent LLM agents:...
We address the problem of person re-identification (reID), that is, retrieving images from a large dataset, given query image interest. A key challenge is to learn representations robust intra-class variations, as different persons could have same attribute, and persons' appearances look different, e.g., with viewpoint changes. Recent reID methods focus on learning features discriminative only for particular factor variations (e.g., human pose), which also requires corresponding supervisory...
We present an empirical study of collecting healthrelated data like sensor signals and lifelog from wearables mobile devices. have developed a health-related collection framework for everyday lives, called Health24, composed in standard IoT platform Galaxy watch, Apple Fitbit watch smartphones iPhone Android phones. Raw signal is collected accelerometers, gyroscopes, pressure sensors, lifelogs such as activities sleep time are extracted smartphone- or cloud-supported APIs. Our designed to...
We present a novel unsupervised domain adaption method for person re-identification (reID) that generalizes model trained on labeled source to an unlabeled target domain. introduce camera-driven curriculum learning (CaCL) framework leverages camera labels of images transfer knowledge from domains progressively. To this end, we divide dataset into multiple subsets based the labels, and initially train our with single subset (i.e., captured by camera). then gradually exploit more training,...
로봇에 대한 필요성이 더 이상 산업용 국한되지 않고 서비스 로봇 혹은 의료 로봇으로 확대됨에 따라 사람과의 공존을 위해 외부 환경에 즉각적으로 대응이 가능한 궤적 생성 방법이 요구되고 있다. 이에 본 논문에서는 컨볼루션 연산을 이용한 실시간으로 변경 방법을 제시한다. 논문에서 제시하는 방법은 기존의 방법과 같이, 시스템의 운동학적 제약 조건 내에서의 궤적을 생성하며 기존 방법의 모든 특성을 만족한다. 또한, 항상 사다리꼴 모양으로 궤적이 생성되는 특성으로 인한 특정 상황에서 비효율적으로 생성될 수 있는 단점을 개선시키는 새로운 모의 실험을 통해 제안하는 유효성과 적합성을 보이며, 방법과의 비교를 그 효율성을 보인다.
We present a novel unsupervised domain adaptation method for semantic segmentation that generalizes model trained with source images and corresponding ground-truth labels to target domain. A key adaptive is learn domain-invariant discriminative features without labels. To this end, we propose bi-directional pixel-prototype contrastive learning framework minimizes intra-class variations of the same object class, while maximizing inter-class different ones, regardless domains. Specifically,...
Sets have been used for modeling various types of objects (e.g., a document as the set keywords in it and customer items that she has purchased). Measuring similarity Jaccard Index) between sets key building block wide range applications, including, plagiarism detection, recommendation, graph compression. However, grown numbers sizes, computational cost storage required computation become substantial, this led to development hashing sketching based solutions. In work, we propose Set2Box,...
Video-based person re-identification (reID) aims to retrieve videos with the same identity as a query across multiple cameras. Spatial and temporal distractors in videos, such background clutter partial occlusions over frames, respectively, make this task much more challenging than image-based reID. We observe that spatial appear consistently particular location, show several patterns, e.g., occur first few where patterns provide informative cues for predicting which frames focus on (i.e.,...
Group interactions arise in our daily lives (email communications, on-demand ride sharing, comment on online communities, to name a few), and they together form hypergraphs that evolve over time. Given such temporal hypergraphs, how can we describe their underlying design principles? If sizes time spans are considerably different, compare structural characteristics? In this work, define 96 hypergraph motifs (TH-motifs), propose the relative occurrences of instances as an answer above...