Search

Temporal action proposal generation (TAPG) aims to estimate temporal intervals of actions in untrimmed videos, which is a challenging yet plays an important role in many tasks of video analysis and understanding. Despite the great achievement in TAPG, most existing works ignore the human perception of interaction between agents and the surrounding environment by applying a deep learning model as a black-box to the untrimmed videos to extract video visual representation. Therefore, it is beneficial and potentially improve the performance of TAPG if we can capture these interactions between agents and the environment. In this paper, we propose a novel framework named Agent-Aware Boundary Network (ABN), which consists of two sub-networks (i) an Agent-Aware Representation Network to obtain both agent-agent and agents-environment relationships in the video representation, and (ii) a Boundary Generation Network to estimate the confidence score of temporal intervals. In the Agent-Aware Representation Network, the interactions between agents are expressed through local pathway, which operates at a local level to focus on the motions of agents whereas the overall perception of the surroundings are expressed through global pathway, which operates at a global level to perceive the effects of agents-environment. Comprehensive evaluations on 20-action THUMOS-14 and 200-action ActivityNet-1.3 datasets with different backbone networks (i.e C3D, SlowFast and Two-Stream) show that our proposed ABN robustly outperforms state-of-the-art methods regardless of the employed backbone network on TAPG. We further examine the proposal quality by leveraging proposals generated by our method onto temporal action detection (TAD) frameworks and evaluate their detection performances. The source code can be found in this URL https://github.com/vhvkhoa/TAPG-AgentEnvNetwork.git., Comment: Accepted in the journal of IEEE Access Vol. 9

Identifying stress level can provide valuable data for mental health analytics as well as labels for annotation systems. Although much research has been conducted into stress detection models using heart rate variability at a higher cost of data collection, there is a lack of research on the potential of using low-resolution Electrodermal Activity (EDA) signals from consumer-grade wearable devices to identify stress patterns. In this paper, we concentrate on performing statistical analyses on the stress detection capability of two popular approaches of training stress detection models with stress-related biometric signals: user-dependent and user-independent models. Our research manages to show that user-dependent models are statistically more accurate for stress detection. In terms of effectiveness assessment, the balanced accuracy (BA) metric is employed to evaluate the capability of distinguishing stress and non-stress conditions of the models trained on either low-resolution or high-resolution Electrodermal Activity (EDA) signals. The results from the experiment show that training the model with (comparatively low-cost) low-resolution EDA signal does not affect the stress detection accuracy of the model significantly compared to using a high-resolution EDA signal. Our research results demonstrate the potential of attaching the user-dependent stress detection model trained on personal low-resolution EDA signal recorded to collect data in daily life to provide users with personal stress level insight and analysis.

Cross-data analytics and retrieval have gained significant improvement recently. People can now extract more data insights precisely and quickly towards having many excellent applications serving human lives. Since people create multimedia and other types of data that reflect the diverse perspectives of human lives, these data are just pieces of the puzzle of the world's pictures. Hence, it is necessary to assembly all these pieces towards having a better solution for human-centered problems. Hence, the workshop welcomes those who work with multimedia and others and come from diverse research domains and disciplines to work on intelligent cross-data analytics and retrieval to bring a smart, sustainable society to human beings. The research domain can vary from well-being, disaster prevention and mitigation, mobility to food computing, to name a few.

Camouflaged objects attempt to conceal their texture into the background and discriminating them from the background is hard even for human beings. The main objective of this paper is to explore the camouflaged object segmentation problem, namely, segmenting the camouflaged object(s) for a given image. This problem has not been well studied in spite of a wide range of potential applications including the preservation of wild animals and the discovery of new species, surveillance systems, search-and-rescue missions in the event of natural disasters such as earthquakes, floods or hurricanes. This paper addresses a new challenging problem of camouflaged object segmentation. To address this problem, we provide a new image dataset of camouflaged objects for benchmarking purposes. In addition, we propose a general end-to-end network, called the Anabranch Network, that leverages both classification and segmentation tasks. Different from existing networks for segmentation, our proposed network possesses the second branch for classification to predict the probability of containing camouflaged object(s) in an image, which is then fused into the main branch for segmentation to boost up the segmentation accuracy. Extensive experiments conducted on the newly built dataset demonstrate the effectiveness of our network using various fully convolutional networks. \url{https://sites.google.com/view/ltnghia/research/camo}, Comment: Published in CVIU 2019. Project page: https://sites.google.com/view/ltnghia/research/camo

Most popular systems for object instance search are based on the bag-of-visual-word model. The inherent weaknesses of this standard model such as quantization error, unstructured representation, burstiness phenomenon are to some extent solved. However, it has a serious problem of searching small objects on a database with cluttered background. In many situations, even the irrelevant objects which share the same texture or shape with a query object get higher score than relevant ones. To overcome this problem, we propose a novel fusion method to significantly boost the accuracy of instance search systems. Firstly, we use the state-of-the-art object detector with denser feature for finding object bounding box and similarity score. Secondly, to exploit the spatial relationship of each visual word with an object proposal, a detected area that might contain a query object, we define three categories of visual word pairs, i.e., discriminative, weak relevant, and context inferred ones. Finally, we propose a new re-ranking scheme with three weighting functions corresponding to the three categories of visual word pairs to compute the final similarity score between a query topic and a video shot. To illustrate the efficiency of the proposed method, we conduct experiments on datasets which have a wide variety of types of query objects. Experimental results on TRECVID Instance Search datasets (INS2013 and INS2014) show the superiority of our proposed method over the state-of-the-art approaches.

Asbestos is a known human carcinogen and the chief known cause of mesothelioma. In 1997, a group of experts developed the Helsinki Criteria, which established criteria for attribution of mesothelioma to asbestos. The criteria include two methods for causation attribution: 1) a history of significant occupational, domestic, or environmental exposure and/or 2) pathologic evidence of exposure to asbestos. In 2014, the Helsinki Criteria were updated, and these attribution criteria were not changed. However, since the Helsinki Criteria were first released in 1997, some pathologists, cell biologists, and others have claimed that a history of exposure cannot establish causation unless the lung asbestos fiber burden exceeds “the background range for the laboratory in question to attribute mesothelioma cases to exposure to asbestos.” This practice ignores the impact on fiber burden of clearance/translocation over time, which in part is why the Helsinki Criteria concluded that a history of exposure to asbestos was independently sufficient to attribute causation to asbestos. After reviewing the Helsinki Criteria, we conclude that their methodology is fatally flawed because a quantitative assessment of a background lung tissue fiber level cannot be established. The flaws of the Helsinki Criteria are both technical and substantive. The 1995 paper that served as the scientific basis for establishing background levels used inconsistent methods to determine exposures in controls and cases. In addition, historic controls cannot be used to establish background fiber levels for current cases because ambient exposures to asbestos have decreased over time and control cases pre-date current cases by decades. The use of scanning electron microscope (SEM) compounded the non-compatibility problem; the applied SEM cannot distinguish talc from anthophyllite because it cannot perform selected area electron diffraction, which is a crucial identifier in ATEM for distinguishing the difference between serpentine asbestos, amphibole asbestos, and talc.

Language prior plays an important role in the way humans detect and recognize text in the wild. Current scene text recognition methods do use lexicons to improve recognition performance, but their naive approach of casting the output into a dictionary word based purely on the edit distance has many limitations. In this paper, we present a novel approach to incorporate a dictionary in both the training and inference stage of a scene text recognition system. We use the dictionary to generate a list of possible outcomes and find the one that is most compatible with the visual appearance of the text. The proposed method leads to a robust scene text recognition model, which is better at handling ambiguous cases encountered in the wild, and improves the overall performance of state-of-the-art scene text spotting frameworks. Our work suggests that incorporating language prior is a potential approach to advance scene text detection and recognition methods. Besides, we contribute VinText, a challenging scene text dataset for Vietnamese, where some characters are equivocal in the visual form due to accent symbols. This dataset will serve as a challenging benchmark for measuring the applicability and robustness of scene text detection and recognition algorithms. Code and dataset are available at https://github.com/VinAIResearch/dict-guided.

Multi-Camera Multiple Object Tracking (MC-MOT) is a significant computer vision problem due to its emerging applicability in several real-world applications. Despite a large number of existing works, solving the data association problem in any MC-MOT pipeline is arguably one of the most challenging tasks. Developing a robust MC-MOT system, however, is still highly challenging due to many practical issues such as inconsistent lighting conditions, varying object movement patterns, or the trajectory occlusions of the objects between the cameras. To address these problems, this work, therefore, proposes a new Dynamic Graph Model with Link Prediction (DyGLIP) approach to solve the data association task. Compared to existing methods, our new model offers several advantages, including better feature representations and the ability to recover from lost tracks during camera transitions. Moreover, our model works gracefully regardless of the overlapping ratios between the cameras. Experimental results show that we outperform existing MC-MOT algorithms by a large margin on several practical datasets. Notably, our model works favorably on online settings but can be extended to an incremental approach for large-scale datasets., accepted at CVPR 2021

Traffic event retrieval is one of the important tasks for intelligent traffic system management. To find accurate candidate events in traffic videos corresponding to a specific text query, it is necessary to understand the text query’s attributes, represent the visual and motion attributes of vehicles in videos, and measure the similarity between them. Thus we propose a promising method for vehicle event retrieval from a natural-language-based specification. We utilize both appearance and motion attributes of a vehicle and adapt the COOT model to evaluate the semantic relationship between a query and a video track. Experiments with the test dataset of Track 5 in AI City Challenge 2021 show that our method is among the top 6 with a score of 0.1560.

This paper pushes the envelope on decomposing camouflaged regions in an image into meaningful components, namely, camouflaged instances. To promote the new task of camouflaged instance segmentation of in-the-wild images, we introduce a dataset, dubbed CAMO++, that extends our preliminary CAMO dataset (camouflaged object segmentation) in terms of quantity and diversity. The new dataset substantially increases the number of images with hierarchical pixel-wise ground truths. We also provide a benchmark suite for the task of camouflaged instance segmentation. In particular, we present an extensive evaluation of state-of-the-art instance segmentation methods on our newly constructed CAMO++ dataset in various scenarios. We also present a camouflage fusion learning (CFL) framework for camouflaged instance segmentation to further improve the performance of state-of-the-art methods. The dataset, model, evaluation suite, and benchmark will be made publicly available on our project page: https://sites.google.com/view/ltnghia/research/camo_plus_plus, TIP acceptance. Project page: https://sites.google.com/view/ltnghia/research/camo_plus_plus

The talc industry and Food and Drug Administration (FDA) have asserted that talc has been asbestos-free since 1976 when the industry created a voluntary specification for the asbestos content of cosmetic talc. However, recent evidence reveals that cosmetic talc is not and never was asbestos-free. This narrative review examines the talc industry’s role in delaying and ultimately blocking federal regulation of cosmetic talc from the 1970s to today. We review primary source material, including corporate documents released in recent litigation and FDA documents released in response to Freedom of Information Act requests. Our results indicate that the talc industry exerted considerable influence over three key areas: regulatory proceedings at the FDA; testing methods and the manipulation of test results (including undisclosed results); and press coverage and the medical literature. The talc companies’ actions and FDA indifference have had a lasting effect on consumer health, including the regulation of talc by other government agencies.

Gastrointestinal (GI) endoscopy has been an active field of research motivated by the large number of highly lethal GI cancers. Early GI cancer precursors are often missed during the endoscopic surveillance. The high missed rate of such abnormalities during endoscopy is thus a critical bottleneck. Lack of attentiveness due to tiring procedures, and requirement of training are few contributing factors. An automatic GI disease classification system can help reduce such risks by flagging suspicious frames and lesions. GI endoscopy consists of several multi-organ surveillance, therefore, there is need to develop methods that can generalize to various endoscopic findings. In this realm, we present a comprehensive analysis of the Medico GI challenges: Medical Multimedia Task at MediaEval 2017, Medico Multimedia Task at MediaEval 2018, and BioMedia ACM MM Grand Challenge 2019. These challenges are initiative to set-up a benchmark for different computer vision methods applied to the multi-class endoscopic images and promote to build new approaches that could reliably be used in clinics. We report the performance of 21 participating teams over a period of three consecutive years and provide a detailed analysis of the methods used by the participants, highlighting the challenges and shortcomings of the current approaches and dissect their credibility for the use in clinical settings. Our analysis revealed that the participants achieved an improvement on maximum Mathew correlation coefficient (MCC) from 82.68% in 2017 to 93.98% in 2018 and 95.20% in 2019 challenges, and a significant increase in computational speed over consecutive years.

In everyday life, authentication is an indispensable process of human activities. Bio-metric authentication system is one of the effective solutions, because it uses human-based features, instead of other traditional features, such as pin, password, etc. However, to apply a face authentication system in practical applications, we need to ensure that the system must not try to recognize the face of an unknown person into known categories, meaning we need to reject faces of unknown people in our application. In this paper, we present the limitations of recent Deep Learning based methods in Face Recognition tasks. We then propose two methods helping Face Recognition system have the ability to reject faces from unknown people by using Open-Set concepts. We conduct the experiments on a subset of CASIA-WebFace dataset, with a train set that includes 7000 images of 100 known people and a test set that includes both known and unknown people. Without rejecting unknown faces, the regular face recognition, i.e. the baseline method, yields the accuracy of only 45.9%, as the method tries to classify all face photos into known classes. Our proposed methods, which are combined deep network of Facenet system with recent Open Set methods, are called Learning Placeholder on Facenet (P-Facenet) and Facenet with OpenMax (O-Facenet). They achieve the accuracy of 83.6% and 88.5% respectively. This is a potential approach for authentication with face recognition to decrease the error rate of the model when recognizing faces of unknown people in the wild.

In the context of modern Android application development, security issues and secure programming are considered unignorable aspects to ensure the safety of Android applications while still guarantee the development speed. The lack of attention to security factors in the software development process or the delay of traditional security assurance methods are the main causes of unsafe Android software. Those unsafe Android applications contain many vulnerabilities and a high risk of leaking user information, especially since Android applications are rapidly developed and published. Developers must adhere to a secure development process to counter Android application risks to avoid data leakage or access control flaws. Security has to be integrated throughout the application development process to secure the software development life cycle. This paper presents two main research contributions: summarizing common security issues in Android applications and developing a plug-in for Android Studio to support secure programming, 9Fix. The low-time-cost 9Fix plug-in can inspect your vulnerable code and instantly suggest an alternative secure code for developers in programming time that helps to improve the security and instruct the developers on how to write a secure code. Moreover, the developers can add their own security rules to 9Fix so 9Fix can adapt smoothly in a specific situation. We also demonstrate the effectiveness and the convenience based on the student feedback by experimenting with the 9Fix plugin.

Gesture recognition is a fundamental tool to enable novel interaction paradigms in a variety of application scenarios like Mixed Reality environments, touchless public kiosks, entertainment systems, and more. Recognition of hand gestures can be nowadays performed directly from the stream of hand skeletons estimated by software provided by low-cost trackers (Ultraleap) and MR headsets (Hololens, Oculus Quest) or by video processing software modules (e.g. Google Mediapipe). Despite the recent advancements in gesture and action recognition from skeletons, it is unclear how well the current state-of-the-art techniques can perform in a real-world scenario for the recognition of a wide set of heterogeneous gestures, as many benchmarks do not test online recognition and use limited dictionaries. This motivated the proposal of the SHREC 2021: Track on Skeleton-based Hand Gesture Recognition in the Wild. For this contest, we created a novel dataset with heterogeneous gestures featuring different types and duration. These gestures have to be found inside sequences in an online recognition scenario. This paper presents the result of the contest, showing the performances of the techniques proposed by four research groups on the challenging task compared with a simple baseline method.

Record companies invest billions of dollars in new talent around the globe each year. Gaining insight into what actually makes a hit song would provide tremendous benefits for the music industry. In this research, we tackle this question by focusing on predicting rank of hit songs in the next 6 months. Our dataset is used in ZALO AI CHALLENGE 2019 in Hit Song Prediction problem including not only songs but also its information such as composer, artist name, released date, etc. Because of that, while most previous work formulates hit song prediction as a regression or classification problem, we present in this paper how to apply Gradient Boosting technique to treat it as a ranking problem. The resulting best model has a good performance when predicting whether a song is a top 10 dance hit versus a lower listed position with 1.48815 Root Mean Square Error - our result dominates most of the solution in this competition (better than 3rd ranked solution of 87 in total). Moreover, it is possible to further improve by extracting chords, tones and more information from each song to obtain the highlights of songs and by using linguistics model to offer high-level features of metadata.

Lifelog can provide useful insights of our daily activities. It is essential to provide a flexible way for users to retrieve certain events or moments of interest, corresponding to a wide variation of query types. This motivates us to develop FIRST, a Flexible Interactive Retrieval SysTem, to help users to combine or integrate various query components in a flexible manner to handle different query scenarios, such as visual clustering data based on color histogram, visual similarity, GPS location, or scene attributes. We also employ personalized concept detection and image captioning to enhance image understanding from visual lifelog data, and develop an autoencoder-like approach for query text and image feature mapping. Furthermore, we refine the user interface of the retrieval system to better assist users in query expansion and verifying sequential events in a flexible temporal resolution to control the navigation speed through sequences of images.

Traffic flow analysis is essential for intelligent transportation systems. In this paper, we introduce our Intelligent Traffic Analysis Software Kit (iTASK) to tackle three challenging problems: vehicle flow counting, vehicle re-identification, and abnormal event detection. For the first problem, we propose to real-time track vehicles moving along the desired direction in corresponding motion-of-interests (MOIs). For the second problem, we consider each vehicle as a document with multiple semantic words (i.e., vehicle attributes) and transform the given problem to classical document retrieval. For the last problem, we propose to forward and backward refine anomaly detection using GAN-based future prediction and backward tracking completely stalled vehicle or sudden-change direction, respectively. Experiments on the datasets of traffic flow analysis from AI City Challenge 2020 show our competitive results, namely, S1 score of 0.8297 for vehicle flow counting in Track 1, mAP score of 0.3882 for vehicle re-identification in Track 2, and S4 score of 0.9059 for anomaly detection in Track 4. All data and source code are publicly available on our project page. 1

Recognition across domains has recently become an active topic in the research community. However, it has been largely overlooked in the problem of recognition in new unseen domains. Under this condition, the delivered deep network models are unable to be updated, adapted, or fine-tuned. Therefore, recent deep learning techniques, such as domain adaptation, feature transferring, and fine-tuning, cannot be applied. This paper presents a novel approach to the problem of domain generalization in the context of deep learning. The proposed method1 is evaluated on different datasets in various problems, i.e. (i) digit recognition on MNIST, SVHN, and MNIST-M, (ii) face recognition on Extended Yale-B, CMU-PIE and CMU-MPIE, and (iii) pedestrian recognition on RGB and Thermal image datasets. The experimental results show that our proposed method consistently improves performance accuracy. It can also be easily incorporated with any other CNN frameworks within an end-to-end deep network design for object detection and recognition problems to improve their performance.

This paper presents an overview of the ImageCLEF 2020 lab that was organized as part of the Conference and Labs of the Evaluation Forum - CLEF Labs 2020. ImageCLEF is an ongoing evaluation initiative (first run in 2003) that promotes the evaluation of technologies for annotation, indexing and retrieval of visual data with the aim of providing information access to large collections of images in various usage scenarios and domains. In 2020, the 18th edition of ImageCLEF runs four main tasks: (i) a medical task that groups three previous tasks, i.e., caption analysis, tuberculosis prediction, and medical visual question answering and question generation, (ii) a lifelog task (videos, images and other sources) about daily activity understanding, retrieval and summarization, (iii) a coral task about segmenting and labeling collections of coral reef images, and (iv) a new Internet task addressing the problems of identifying hand-drawn user interface components. Despite the current pandemic situation, the benchmark campaign received a strong participation with over 40 groups submitting more than 295 runs.

Chest X-ray examination plays an important role in lung disease detection. The more accuracy of this task, the more experienced radiologists are required. After ChestX-ray14 dataset containing over 100,000 frontal-view X-ray images of 14 diseases was released, several models were proposed with high accuracy. In this paper, we develop a work flow for lung disease diagnosis in chest X-ray images, which can improve the average AUROC of the state-of-the-art model from 0.8414 to 0.8445. We apply image preprocessing steps before feeding to the 14 diseases detection model. Our project includes three models: the first one is DenseNet-121 to predict whether a processed image has a better result, a convolutional auto-encoder model for bone shadow exclusion is the second one, and the last is the original CheXNet., Comment: KSE 2018 long paper

Objective Asbestos is a known cause of ovarian cancer. We report 10 cases of serous ovarian cancer among users of Johnson & Johnson (J&J) asbestos-containing "cosmetic" talc products. Methods We conducted an asbestos exposure assessment during talc application and analyzed surgical tissues and talc containers for asbestos and talc. Results Talc was found in all cases and tremolite and/or anthophyllite asbestos was found in 8/10 cases. The asbestos fibers found in the "cosmetic" talc containers matched those found in tissues. We estimated inhaled asbestos dose ranged from 0.38 to 5.18 fiber years. Conclusion We provide evidence that the inhaled dose of asbestos/fibrous talc from "cosmetic" talc use causes ovarian cancer. The unique combination of the types of asbestiform minerals detected in cancerous tissue and "cosmetic" talc is a fingerprint for exposure to asbestos-containing talc.

Inspired by recent advances in computer vision and deep learning, we propose new enhancements to tackle problems appearing in endoscopic image analysis, especially abnormality finding and anatomical landmark detection. In details, a combination of Residual Neural Network and Faster R-CNN are jointly applied in order to take all of their advantages and improve the overall performance. Nevertheless, novel data augmentation is designed and adapted to corresponding domains. Our approaches prove their competitive results in term of not only the accuracy but also the inference time in Medico: The 2018 Multimedia for Medicine Task and The Biomedia ACM MM Grand Challenge 2019. These results show the great potential of the collaborating between deep learning models and data augmentation in medical image analysis applications. Especially, more than 4900 bounding boxes localizing the symptom of some classes from KVASIR dataset that we annotated and used in this project are shared online for future research.

Despite recent advances in deep learning, the crucial task of nuclear segmentation for computational pathology remains challenging. Recently, deep learning, and specifically U-Nets, have shown significant improvements for this task, but there is still room for improvement by further enhancing the design and training of U-Nets for nuclear segmentation. Specifically, we consider enforcing rotation equivariance in the network, the placement of residual blocks, and applying novel data augmentation designed specifically for histopathology images, and show the relative improvement and merit of each. Incorporating all of these enhancements in the design and training of a U-Net yields significantly improved segmentation results while still maintaining a speed of inference that is sufficient for real-world applications, in particular, analyzing whole-slide images (WSIs). Code for our enhanced U-Net is available at https://github.com/thatvinhton/G-U-Net.

With non-stop development of e-commerce and different internet-based applications’ demands, service providers include many physical servers scattering all over the world. If some users would like to use various services, they have to repeatedly register. Therefore, they must remember all these information such as, user-names or passwords. To solve this problem, many authentication schemes in multi-server environment are proposed. Furthermore, to prevent the adversary from keeping track of another user when he or she logins, many schemes apply dynamic identity, but they have some limitations with popular kinds of attacks such as, replay attack, impersonation attack, or man-in-the-middle attack... In 2014, Yeh (Department of Information Management, National Dong Hwa University) proposed multi-server scheme using Rivest–Shamir–Adleman (RSA). However, we discover this scheme cannot achieve two basic properties, mutual authentication and session-key agreement. In this paper, we concretely demonstrate that discovery and propose a different provable version using elliptic curve cryptosystem in multi-server environment which overcomes the limitation of Yeh’s scheme and satisfies security and efficiency.

Aims: To determine serum leptin concentration, insulin resistance and the relationship with other factors in pre-diabetes population. Methods: A total 275 prediabetic subjects were included in this study. Collected data were: anthropometric characteristics, blood pressure, family history and history of metabolic diseases. Fasting blood samples were obtained in the early morning and assayed for serum leptin, blood glucose, HbA1c, insulin, lipid profile and hsCRP. Results: Serum leptin levels were significant higher (4.57 (0.04 – 61.57) ng/mL) in prediabetes group. The prevalence of hyperleptinemia was 41% and significant associated with female gender, overweight, viseral obesity, insulin, HOMA-IR and hsCRP. The prevalence of insulin resistance was 61% and leptin was an independent risk factor of insulin resistance. Conclusion: Prediabetic patients have higher leptin concentration than normal subjects and hyperleptinemia is associated insulin resistance.

Dr. Victor Roggli republished a "literature review and critical analysis" of the toxicity of short fiber asbestos. His paper was originally prepared and presented at a conference of asbestos defense lawyers. His review omitted published papers that indicate that short fiber asbestos is a carcinogen. We critically review his paper.

City anomalies such as residential fires or urban floods can cause a lot of damage each year from loss of human lives to negative effects on the productivity of the city. One possible way to deal with these anomalies in a quick and effective manner is to use the information provided by the network of citizens that is likely present at the scene of any serious anomaly. However, this source of information may be of varying quality and there may be too much data for human operators to manually inspect in a speedy manner. This motivates us to propose an architecture to effectively make use of the network of citizens to deal with city anomalies. At the center of our architecture is a neural network that automatically classify incoming image data and use this information to assist anomaly handling efforts. In order for our classifier to work well, we have also collected a dataset of city anomalies specific to Vietnam, consisting of 9963 images. We experiment with several neural network models on our collected dataset. In overall, they all have decent accuracies with the best being MobileNet with 92.3% accuracy.

Searchable Symmetric Encryption (SSE) enables clients to securely store data on untrusted server while keeping the ability to search and update over the encrypted data efficiently. For practical purposes, we also need several additional properties to make our search function more versatile. In this paper, we introduce an extension type of SSE which we called Keyword-search Interval Dynamic SSE (KIDSSE). In KIDSSE we can search for encrypted files containing a keyword that occurs in a queried range. We also construct a solution for KIDSSE and show that our scheme maintains all important security properties such as forward privacy, backward privacy.

Generalized nucleus segmentation techniques can contribute greatly to reducing the time to develop and validate visual biomarkers for new digital pathology datasets. We summarize the results of MoNuSeg 2018 Challenge whose objective was to develop generalizable nuclei segmentation techniques in digital pathology. The challenge was an official satellite event of the MICCAI 2018 conference in which 32 teams with more than 80 participants from geographically diverse institutes participated. Contestants were given a training set with 30 images from seven organs with annotations of 21,623 individual nuclei. A test dataset with 14 images taken from seven organs, including two organs that did not appear in the training set was released without annotations. Entries were evaluated based on average aggregated Jaccard index (AJI) on the test set to prioritize accurate instance segmentation as opposed to mere semantic segmentation. More than half the teams that completed the challenge outperformed a previous baseline. Among the trends observed that contributed to increased accuracy were the use of color normalization as well as heavy data augmentation. Additionally, fully convolutional networks inspired by variants of U-Net, FCN, and Mask-RCNN were popularly used, typically based on ResNet or VGG base architectures. Watershed segmentation on predicted semantic segmentation maps was a popular post-processing strategy. Several of the top techniques compared favorably to an individual human annotator and can be used with confidence for nuclear morphometrics.

In this paper, we propose our method for vehicle detection with multiple adaptive vehicle detectors and velocity estimation with landmark-based scanlines. Inspired by the idea for tiny object detection, we use Faster R-CNN with Resnet-101 to create different specialized vehicle detectors corresponding to different levels of details and poses. We propose a heuristic to check the fitness of a particular vehicle detector to a specific region in camera's view by the mean velocity direction and the mean object size. By this way, we can determine an adaptive set of appropriate vehicle detectors for each region in camera's view. Thus our system is expected to detect vehicles with high accuracy, both in precision and recall, even with tiny objects. We exploit the U.S. road rules for the length and distance of broken white lines on roads to propose our method for vehicle's velocity estimation using such landmarks. We determine equally-distributed scanlines, virtual parallel lines that are nearly-perpendicular to the road direction, with reference to the line connecting the corresponding ends of multiple broken white lines. From the timespan for a vehicle to cross two consecutive virtual scanlines, we can calculate the average vehicle's velocity within that road segment. We also refine the speed estimation by detecting when a vehicle stops at a traffic light, and smooth the results with a moving average filter. Experiments on the dataset of Traffic Flow Analysis from NVIDIA AI City Challenge 2018 show that our method achieves the perfect detect rate of 100%, the average velocity difference of 6.9762 mph on freeways, and 8.9144 mph on both freeways and urban roads.

At DAVIS-2016 Challenge, many state-of-art video segmentation methods achieve potential results, but they still much depend on annotated frames to distinguish between background and foreground. It takes a lot of time and efforts to create these frames exactly. In this paper, we introduce a method to segment objects from video based on keywords given by user. First, we use a real-time object detection system - YOLOv2 to identify regions containing objects that have labels match with the given keywords in the first frame. Then, for each region identified from the previous step, we use Pyramid Scene Parsing Network to assign each pixel as foreground or background. These frames can be used as input frames for Object Flow algorithm to perform segmentation on entire video. We conduct experiments on a subset of DAVIS-2016 dataset in half the size of its original size, which shows that our method can handle many popular classes in PASCAL VOC 2012 dataset with acceptable accuracy, about 75.03%. We suggest widely testing by combining other methods to improve this result in the future.

This paper presents a novel tunable dual-band bandstop filter based on doubly tuned RF transformers. This design results in two distinct notch frequencies in a single resonator using silicon varactor diodes. The 2-pole tunable dual-band bandstop filters are implemented using printed circuit board (PCB) transformers and air-coil transformers. With PCB transformers, the lower frequency can be tuned at 513–845 MHz while the higher frequency can be tuned at 715–1151 MHz with a ${\hbox{notch level}} > 16~{\hbox{dB}}$ . With air-coil transformers, a single-band 2-pole notch filter with wide rejection bandwidth is achieved. The design results in 604–982 MHz tuning with a 20-dB rejection bandwidth of 27–45 MHz. Also, by implementing a series varactor in a transformer, the separation between two coupled frequencies can be changed. The topology can be easily extended to higher order filters and design equations are presented.

Large-scale image retrieval has been shown remarkable potential in real-life applications. The standard approach is based on Inverted Indexing, given images are represented using Bag-of-Words model. However, one major limitation of both Inverted Index and Bag-of-Words presentation is that they ignore spatial information of visual words in image presentation and comparison. As a result, retrieval accuracy is decreased. In this paper, the authors investigate an approach to integrate spatial information into Inverted Index to improve accuracy while maintaining short retrieval time. Experiments conducted on several benchmark datasets (Oxford Building 5K, Oxford Building 5K+100K and Paris 6K) demonstrate the effectiveness of our proposed approach.

Paraphrase identification is a core component for many significant tasks in natural language processing (e.g., text summarization, headline generation). A method suggested by Bach et al. for detecting Vietnamese paraphrase text using nine similarity metrics. The authors state that it is the first method for Vietnamese text. They evaluated the method on vnPara corpus with 3000 sentence pairs. However, this corpus is limited by collecting from few Vietnamese websites. Most other methods have focused on the English text. For instance, our previous method detected paraphrasing sentences by matching identical phrases and close words using Wordnet similarity. This method is unsuitable for Vietnamese due to the restriction of Wordnet corpora and morphological words in Vietnamese. Therefore, we extend the method to identify the paraphrase by proposing a SimVN metric which measures the similarity of two Vietnamese words. We evaluated the proposed method on the vnPara corpus. The result shows that the method achieves better accuracy (97.78%) comparing with the state-of-the-art method (accuracy = 89.10%). The proposed method then creates a high diversity paraphrase corpus with 3134 sentence pairs in eight main topics from the top fifteen popular Vietnamese news websites.

News classification is among essential needs for people to organize, better understand, and utilize information from the Internet. This motivates the authors to propose a novel method to classify news from social media. First, we propose to vectorize an article with TD2V, our pre-trained Twitter-based universal document representation following Doc2Vec approach. We then define Modified Distance to better measure the semantic distance between two document vectors. Finally, we apply retrieval and automatic query expansion to get the most relevant labeled documents in a corpus to determine the category for a new article. As our TD2V is created from 297 million words in 420,351 news articles from more than one million tweets in Twitters from 2010 to 2017, it can be used as one of the efficient pre-trained models for English document representation in various applications. Experiments on datasets from different online sources show that our method achieves the classification accuracy better than existing methods, specifically 98.4±0.3% (BBC dataset), 98.9±0.7% (BBC Sport dataset), 94.1±0.2% (Amazon4 dataset), and 78.6% (20NewsGroup dataset). Furthermore, in the classification training process, we just encode all articles in the training set with TD2V, not to train a dedicated classification model for each of these datasets.

Document sensitivity classification is essential to prevent potential sensitive data leakage for individuals and organizations. As most of existing methods use regular expressions or data fingerprinting to classify sensitive documents, they may not fully exploit the semantic and content of a document, especially with informal messages and files. This motivates the authors to propose a novel method to classify document sensitivity in realtime with better semantic and content analysis. Taking advantages of deep learning in natural language processing, we use our pre-trained Twitter-based document embedding TD2V to encode a document or a text fragment into a fixed length vector of 300 dimensions. Then we use retrieval and automatic query expansion to retrieve a re-ranked list of semantically similar known documents, and determine the sensitivity score for a new document from those of the retrieved documents in this list. Experimental results show that our method can achieve classification accuracy of more than 99.9% for 4 datasets (snowden, Mormon, Dyncorp, TM) and 98.34% for Enron dataset. Furthermore, our method can early predict a sensitive document from a short text fragment with the accuracy higher than 98.84%.

Gait recognition has been considered as a new promising approach for biometric-based authentication. Gait signals are commonly obtained by collecting motion data from inertial sensors (accelerometers, gyroscopes) integrated in mobile and wearable devices. Motion data is subsequently transformed to a feature space for recognition procedure. One fashionable, effective way to extract features automatically is using conventional Convolutional Neural Networks (CNN) as feature extractors. In this paper, we propose DeepSense-Inception (DSI), a new method inspired from DeepSense, to recognize users from their gait features using Inception-like modules for better feature extraction than conventional CNN. Experiments for user identification on UCI Human Activity Recognition dataset demonstrate that our method not only achieves an accuracy of 99.9%, higher than that of DeepSense (99.7%), but also uses only 149K parameters, less than one third of the parameters in DeepSense (529K parameters). Thus, our method can be implemented more efficiently in limited resource systems.

Wireless environments such as GSM, 3G, and 4G are more and more popular. Consequently, communications in such networks need to be guarded. It is necessary to have a secure mutual authentication scheme to defend transactions between user and service provider against illegitimate adversaries. Especially, users are those vulnerable to attacks and there are many authentication schemes with smart-card proposed to protect them. Recently, Yung-Cheng Lee has suggested a dynamic identity based user authentication scheme to resist smart-card-theft attack. Nevertheless, he assumed that smart-card is tamperproof. In our opinion, this is not appropriate because Kocher and Messerges pointed that smart-card's confidential information could be extracted by physically monitoring its power consumption. Therefore, design of Yung-Cheng Lee cannot withstand this kind of attack. In addition, anyone who is a legal member can masquerade server or other legal users in his scheme. Moreover, legitimacy verification only starting from server side truly makes Lee's scheme be impractical. In this paper, we present an improvement to his scheme to isolate such problems.

With non-stop growth in network environments, communication security is necessary. A strong protocol guarantees that users and service providers are secure against many kinds of attacks, such as impersonation and replay attack. Sood et al. proposed an authentication scheme based on dynamic identity to prevent transactions from being intercepted by malicious users. Although they claimed that their scheme has advantages over previous schemes with the same approach, we prove that their scheme is vulnerable to impersonation attack and stolen verification attack, and can be affected by clock synchronization. Therefore we propose a novel authentication scheme to enhance security and overcome limitations existing in Sood's scheme. Our security analysis shows that our proposed method can efficiently resist known types of attacks. Experimental results also show that the method can be implemented and processed in real-time thus applicable for not only regular computers but also mobile devices.

As inertial sensors are low-cost, easy-to-use, and can be integrated in wearable devices, they can be used to establish as a new modality for user authentication in the smart environment in which computing systems can recognize persons implicitly by their walking patterns. This motivates our proposal to use multi-region size Convolutional Neural Network to recognize users from their gait patterns recorded from accelerometers and gyroscopes in mobile and wearable devices.

Human Activity Recognition is one of the attractive topics to develop smart interactive environment in which computing systems can understand human activities in natural context. Besides traditional approaches with visual data, inertial sensors in wearable devices provide a promising approach for human activity recognition. In this paper, we propose novel methods to recognize human activities from raw data captured from inertial sensors using convolutional neural networks with either 2D or 3D filters. We also take advantage of hand-crafted features to combine with learned features from Convolution-Pooling blocks to further improve accuracy for activity recognition. Experiments on UCI Human Activity Recognition dataset with six different activities demonstrate that our method can achieve 96.95%, higher than existing methods.

Face clustering is a task to partition facial images into disjoint clusters. In this paper, we investigate a specific problem of face clustering in videos. Unlike traditional face clustering problem with a given collection of images from multiple sources, our task deals with set of face tracks with information about frame ID. Thus, we can exploit two kinds of prior knowledge about the temporal and spatial information from face tracks: sequence of faces in the same track and contemporary faces in the same frame. We utilize this forehand lore and characteristic of low rank representation to introduce a new light weight but effective method entitled Very Fast Sparse Clustering (VFSC). Since the superior speed of VFSC, the method can be adapted into large scale real-time applications. Experimental results with two public datasets (BF0502 and Notting-Hill), on which our proposed method significantly breaks the limits of not only speed but also accuracy clustering of state-of-the-art algorithms (up to 250 times faster and 10% higher in accuracy), reveal the imminent power of our approach.

In this paper, we propose two approaches to deal with the problems of video searching: ad-hoc video search and known item search. First, we propose to combine multiple semantic concepts extracted from multiple networks trained on many data domains. Second, to help user find exactly video shot that has been shown before, we propose a sketch based search system which detects and indexes many objects proposed by an object proposal algorithm. By this way, we not only leverage the concepts but also the spatial relations between them.

People frequently post their photos on social network sites (e.g. Facebook, Instagram) as a way to share memorable moments, emotions, or locations visited. While sharing photos with the same subjects as past photos could lead to user reminiscence and potentially create valuable benefits, existing products still cannot support this adequately. Based on a survey on user habits in sharing and revisiting photos on social network sites, we propose NowAndThen, a photo recommendation concept and tool that assists reminiscence when sharing photos on social network sites. By combining visual features of the photos and associated tags, a prototype we developed can recommend old photos that have common subjects with the user's current photos of interest. Our study with the prototype shows that this approach can help users positively revive past memories and connections with their friends. In addition, our results include various design insights and implications for future reminiscence-support systems.