A5062359795- Advanced Image and Video Retrieval Techniques
- Visual Attention and Saliency Detection
- Generative Adversarial Networks and Image Synthesis
- Advanced Neural Network Applications
- Adversarial Robustness in Machine Learning
- Face recognition and analysis
- Video Surveillance and Tracking Methods
- Digital Media Forensic Detection
- Multimodal Machine Learning Applications
- Human Pose and Action Recognition
- Video Analysis and Summarization
- Image Enhancement Techniques
- Anomaly Detection Techniques and Applications
- Face and Expression Recognition
- AI in cancer detection
- Image Retrieval and Classification Techniques
- Forensic and Genetic Research
- Image and Video Quality Assessment
- Advanced Vision and Imaging
- Robotics and Sensor-Based Localization
- Augmented Reality Applications
- Music and Audio Processing
- Domain Adaptation and Few-Shot Learning
- 3D Shape Modeling and Analysis
- Industrial Vision Systems and Defect Detection
Ho Chi Minh City University of Science
2012-2025
Vietnam National University Ho Chi Minh City
2012-2024
National Institute of Informatics
2020-2024
The University of Tokyo
2002-2023
The Graduate University for Advanced Studies, SOKENDAI
2016-2019
National Graduate Institute for Policy Studies
2019
Korea University
2010
Moi University
2009
Purdue University West Lafayette
2009
Regenstrief Institute
2009
Camouflaged objects are generally difficult to be detected in their natural environment even for human beings. In this paper, we propose a novel bio-inspired network, named the MirrorNet, that leverages both instance segmentation and mirror stream camouflaged object segmentation. Differently from existing networks segmentation, our proposed network possesses two streams: main corresponding with original image its flipped image, respectively. The output is then fused into stream's result...
This paper presents a method for detecting salient objects in videos, where temporal information addition to spatial is fully taken into account. Following recent reports on the advantage of deep features over conventional handcrafted features, we propose new set spatiotemporal (STD) that utilize local and global contexts frames. We also conditional random field (STCRF) compute saliency from STD features. STCRF our extension CRF domain describes relationships among neighboring regions both...
The proliferation of deepfake media is raising concerns among the public and relevant authorities. It has become essential to develop countermeasures against forged faces in social media. This paper presents a comprehensive study on two new countermeasure tasks: multi-face forgery detection segmentation in-the-wild. Localizing multiple human unrestricted natural scenes far more challenging than traditional recognition task. To promote these tasks, we have created first large-scale dataset...
Deep neural networks are vulnerable to adversarial examples (AEs), which have transferability: AEs generated for the source model can mislead another (target) model's predictions. However, transferability has not been understood in terms of class target predictions were misled (i.e., class-aware transferability). In this paper, we differentiate cases a predicts same wrong as ("same mistake") or different ("different analyze and provide an explanation mechanism. We find that (1) tend cause...
This paper pushes the envelope on decomposing camouflaged regions in an image into meaningful components, namely, instances. To promote new task of instance segmentation in-the-wild images, we introduce a dataset, dubbed CAMO++, that extends our preliminary CAMO dataset (camouflaged object segmentation) terms quantity and diversity. The substantially increases number images with hierarchical pixel-wise ground truths. We also provide benchmark suite for segmentation. In particular, present...
The reliability of remote identity-proofing systems (i.e., electronic Know Your Customer, or eKYC, systems) is challenged by the development deepfake generation tools, which can be used to create fake videos that are difficult detect using existing detection models and indistinguishable facial recognition systems. This poses a serious threat eKYC danger individuals' personal information property. Existing datasets not particularly appropriate for developing evaluating systems, require...
Few-shot instance segmentation extends the few-shot learning paradigm to task, which tries segment objects from a query image with few annotated examples of novel categories. Conventional approaches have attempted address task via prototype learning, known as point estimation. However, this mechanism depends on prototypes (e.g. mean K-shot) for prediction, leading performance instability. To overcome disadvantage estimation mechanism, we propose approach, dubbed MaskDiff, models underlying...
This paper addresses accident detection where we not only detect objects with classes, but also recognize their characteristic properties. More specifically, aim at simultaneously detecting object class bounding boxes on roads and recognizing status such as safe, dangerous, or crashed. To achieve this goal, construct a new dataset propose baseline method for benchmarking the task of detection. We design an network, called Attention R-CNN, which consists two streams: one is classes property...
The regularising parameter of the energy function in graph-cut based image segmentation methods should be carefully determined since it strongly affects result. Presented is a modified technique using novel without for segmentation. In addition, mixture colour model employed to apply proposed effectively. Experimental results show that method outperforms conventional algorithm terms accuracy and ease use.
The fashion e-commerce industry has witnessed significant growth in recent years, prompting exploring image-based virtual try-on techniques to incorporate Augmented Reality (AR) experiences into online shopping platforms. However, existing research primarily overlooked a crucial aspect - the runtime of underlying machine-learning model. While methods prioritize enhancing output quality, they often disregard execution time, which restricts their applications on limited range devices. To...
A model compound of Tyr244-His240 bovine cytochrome c oxidase was synthesized and examined with UV resonance Raman (UVRR) as well absorption spectroscopy pH titration. Owing to the covalent linkage between imidazole phenol, pKa phenolic OH imidazolic NδH groups were lowered by 1.1 2.3, respectively. UVRR measurements ortho-imidazole-bound para-cresol (Im⧫CrOH), its deprotonated anion (Im⧫CrO-), neutral radical (Im⧫CrO•), their perdeuterated, cresol 18O derivatives allowed assignments bands...
Amount and variety of training data drastically affect the performance CNNs. Thus, annotation methods are becoming more critical to collect efficiently. In this paper, we propose a simple yet efficient Interactive Self-Annotation framework cut down both time human labor cost for video object bounding box annotation. Our method is based on recurrent self-supervised learning consists two processes: automatic process interactive process, where aims build supported detector speed up process....
Traffic flow analysis is essential for intelligent transportation systems. In this paper, we introduce our Intelligent Analysis Software Kit (iTASK) to tackle three challenging problems: vehicle counting, re-identification, and abnormal event detection. For the first problem, propose real-time track vehicles moving along desired direction in corresponding motion-of-interests (MOIs). second consider each as a document with multiple semantic words (i.e., attributes) transform given problem...
This paper presents the first adversarial example based method for attacking human instance segmentation networks, namely person networks in short, which are harder to fool than classification networks. We propose a novel Fashion-Guided Adversarial Attack (FashionAdv) framework automatically identify attackable regions target image minimize effect on quality. It generates textures learned from fashion style images and then overlays them clothing original make all persons invisible The...
Finger vein recognition (FVR) systems have been commercially used, especially in ATMs, for customer verification. Thus, it is essential to measure their robustness against various attack methods, when a handcrafted FVR system used without any countermeasure methods. In this paper, we are the first literature introduce master attacks which craft vein-looking image so that can falsely match with as many identities possible by systems. We present two methods generating veins use attacking these...
Masked face counting is the of faces at various crowd densities and discriminating between masked unmasked faces, which generally considered to be an object (i.e., face) detection task. Counting accuracy limited, especially higher densities, when are relatively small, unclear, viewed angles. Furthermore, it costly create ground-truth bounding boxes needed train methods. We formulate as a fine-grained crowd-counting task, appropriate for tackling this challenging task used with density map...
The AMPATH program is a leading initiative in rural Kenya providing healthcare services to combat HIV. Malnutrition and food insecurity are common among patients the Nutritional Information System (NIS) was designed, with cross-functional collaboration between engineering medical communities, as comprehensive electronic system record assist effective distribution region poor infrastructure.The NIS designed modularly support urgent need of for growing program. manages ordering, storage,...