- Human Pose and Action Recognition
- Video Surveillance and Tracking Methods
- Multimedia Communication and Technology
- Video Analysis and Summarization
- Advanced Image and Video Retrieval Techniques
- Multimodal Machine Learning Applications
- Advanced Neural Network Applications
- Digital Rights Management and Security
- Anomaly Detection Techniques and Applications
- Domain Adaptation and Few-Shot Learning
- Cloud Computing and Remote Desktop Technologies
- Network Security and Intrusion Detection
- Smart Grid Security and Resilience
- Visual Attention and Saliency Detection
- IoT and Edge/Fog Computing
- Advanced Vision and Imaging
- Mobile Agent-Based Network Management
- Artificial Intelligence in Healthcare
- Advanced Malware Detection Techniques
- Brain Tumor Detection and Classification
- Context-Aware Activity Recognition Systems
- Nuclear Materials and Properties
- Innovation in Digital Healthcare Systems
- Peer-to-Peer Network Technologies
- Generative Adversarial Networks and Image Synthesis
Seoul National University of Science and Technology
2020-2025
Electronics and Telecommunications Research Institute
2011-2020
Korea Electric Power Corporation (South Korea)
2014
Chungnam National University
2012-2014
Green Cross (South Korea)
2009
Gwangju Institute of Science and Technology
2002
We propose a simple yet efficient anchor-free instance segmentation, called CenterMask, that adds novel spatial attention-guided mask (SAG-Mask) branch to one stage object detector (FCOS) in the same vein with Mask R-CNN. Plugged into FCOS detector, SAG-Mask predicts segmentation on each box attention map helps focus informative pixels and suppress noise. also present an improved backbone networks, VoVNetV2, two effective strategies: (1) residual connection for alleviating optimization...
As DenseNet conserves intermediate features with diverse receptive fields by aggregating them dense connection, it shows good performance on the object detection task. Although feature reuse enables to produce strong a small number of model parameters and FLOPs, detector backbone rather slow speed low energy efficiency. We find linearly increasing input channel connection leads heavy memory access cost, which causes computation overhead more consumption. To solve inefficiency DenseNet, we...
We present a novel image editing system that generates images as the user provides free-form masks, sketches and color inputs. Our consists of an end-to-end trainable convolutional network. In contrast to existing methods, our utilizes entirely input in terms shape. This allows respond user's sketch inputs, using them guidelines generate image. this work, we trained network with additional style loss, which made it possible realistic results despite large portions being removed. proposed...
From a streaming video, online action detection aims to identify actions in the present. For this task, previous methods use recurrent networks model temporal sequence of current frames. However, these overlook fact that an input image includes background and irrelevant as well interest. detection, paper, we propose novel unit explicitly discriminate information relevant ongoing from others. Our unit, named Information Discrimination Unit (IDU), decides whether accumulate based on its...
In this paper, we propose a new framework for detecting the unauthorized dumping of garbage in real‐world surveillance camera. Although several action/behavior recognition methods have been investigated, these studies are hardly applicable to scenarios because they mainly focused on well‐refined datasets. Because actions take variety forms, building method disclose instead exploiting previous approaches is better strategy. We detected action by change relation between person and object being...
In the current landscape where cybersecurity threats are escalating in complexity and frequency, traditional defense mechanisms like rule-based firewalls signature-based detection proving inadequate. The dynamism sophistication of modern cyber-attacks necessitate advanced solutions that can evolve adapt real-time. Enter field deep reinforcement learning (DRL), a branch artificial intelligence has been effectively tackling complex decision-making problems across various domains, including...
An effective way to reduce emotional distress is by sharing negative emotions with others. This why counseling a virtual counselor an emerging methodology, where the sharer can consult freely anytime and anywhere without having fear being judged. To improve effectiveness, most studies so far have focused on designing verbal compassion for counselors. However, recent showed that counselors' nonverbal through eye contact, facial mimicry, head-nodding also significant impact overall experience....
Generative adversarial networks (GANs) have demonstrated remarkable performance in various fashion-related applications, including virtual try-ons, compatible clothing recommendations, fashion editing, and the generation of items. Despite this progress, limited research has addressed specific challenge generating a item with an ensemble consisting distinct categories, such as tops, bottoms, shoes. In response to gap, we propose novel GANs framework, termed Deep Fashion Designer Adversarial...
Recently, the IT industry has become larger, and cloud service rapidly increased; thus cybersecurity to protect sensitive data from attacks an important factor. However, services have making surface area a complex cyber environment leads difficulty managing defending. With rise of artificial intelligence, applying intelligence automatically detect respond cyberattacks begun get attention. In order apply in environments, simulation framework that is easily applicable can represent real...
Precision medicine is a new approach to understanding health and disease based on patient-specific data such as medical diagnoses; clinical phenotype; biologic investigations laboratory studies imaging; environmental, demographic, lifestyle factors. The importance of machine learning techniques in healthcare has expanded quickly the last decade owing rising availability vast multi-modality developed computational models algorithms. Reinforcement an appealing method for developing efficient...
Cybersecurity is a growing concern in today's interconnected world. Traditional cybersecurity approaches, such as signature-based detection and rule-based firewalls, are often limited their ability to effectively respond evolving sophisticated cyber threats. Reinforcement learning (RL) has shown great potential solving complex decision-making problems various domains, including cybersecurity. However, there significant challenges overcome, the lack of sufficient training data difficulty...
In visual surveillance systems, it is necessary to recognize the behavior of people handling objects such as a phone, cup, or plastic bag. this paper, address problem, we propose new framework for recognizing object-related human actions by graph convolutional networks using and object poses. framework, construct skeletal graphs reliable poses selectively sampling informative frames in video, which include joints with high confidence scores obtained pose estimation. The generated from...
Recent temporal action proposal generation approaches have suggested integrating segment- and snippet score-based methodologies to produce proposals with high recall accurate boundaries. In this paper, different from such a hybrid strategy, we focus on the potential of approach. Specifically, propose new method, named Snippet Relatedness-based Generator (SRG), novel concept "snippet relatedness". relatedness represents which snippets are related specific instance. To effectively learn...
We present a novel image editing system that generates images as the user provides free-form mask, sketch and color an input. Our consist of end-to-end trainable convolutional network. Contrary to existing methods, our wholly utilizes input with shape. This allows respond user's input, using it guideline generate image. In particular work, we trained network additional style loss which made possible realistic results, despite large portions being removed. proposed architecture SC-FEGAN is...
The significance of machine-learning approaches in the healthcare domain has grown rapidly owing to existence enormous amounts data and well-established simulation models algorithms. digitization health-related data, as well rapid technological advancements are accelerating development application machine learning healthcare, particularly precision medicine. ultimate goal medicine is provide personalized medicine, which requires tailoring medical decisions each patient based on their...
The detection of abnormal postures, such as that a reclining person, is crucial part visual surveillance. Further, even regular poses can appear rotated because incongruity between the image and angle pre-installed camera. However, most existing human pose estimation methods focus on small rotational changes, i.e., those less than 50 degrees, they seldom consider robust for more drastic changes. To best our knowledge, there have been no reports robustness changes through large angles. In...
The welding of zirconium alloy components is one the most critical processes in fabrication nuclear fuel rods used pressurized water reactors. For this, various processes, such as gas tungsten arc welding, electron beam laser and resistance pressure (RPW), are around world. In Korea, RPW process being to fabricate assembly rods. This study investigated changes weldment shape owing conditions current, force, overlapping. soundness was evaluated by hydraulic burst test. temperature weld zone...
In terms of safety and the efficient management spent fuel storage, detecting failed is one most important tasks in a CANada Deuterium Uranium (CANDU) reactor operation. It has been successfully demonstrated that CANDU reactor, on-power detection location systems, along with alarm area gamma monitors, can detect locate defective suspect bundles before discharging them from to storage bay. reception bay, however, only visual inspection used identify bundles. Gaseous fission product delayed...
Visual relationship detection is an intermediate image understanding task that detects two objects and classifies a predicate explains the between in image. The three components are linguistically visually correlated (e.g. "wear" related to "person" "shirt", while "laptop" "table" "on") thus, solution space huge because there many possible cases them. Language visual modules exploited sophisticated spatial vector proposed. models this work outperformed state of arts without costly linguistic...