Search

Semantic segmentation is essential in computer vision for various applications, yet traditional approaches face significant challenges, including the high cost of annotation and extensive training for supervised learning. Additionally, due to the limited predefined categories in supervised learning, models typically struggle with infrequent classes and are unable to predict novel classes. To address these limitations, we propose MaskDiffusion, an innovative approach that leverages pretrained frozen Stable Diffusion to achieve open-vocabulary semantic segmentation without the need for additional training or annotation, leading to improved performance compared to similar methods. We also demonstrate the superior performance of MaskDiffusion in handling open vocabularies, including fine-grained and proper noun-based categories, thus expanding the scope of segmentation applications. Overall, our MaskDiffusion shows significant qualitative and quantitative improvements in contrast to other comparable unsupervised segmentation methods, i.e. on the Potsdam dataset (+10.5 mIoU compared to GEM) and COCO-Stuff (+14.8 mIoU compared to DiffSeg). All code and data are released at https://github.com/Valkyrja3607/MaskDiffusion.

Semantic segmentation is a crucial task in computer vision, where each pixel in an image is classified into a category. However, traditional methods face significant challenges, including the need for pixel-level annotations and extensive training. Furthermore, because supervised learning uses a limited set of predefined categories, models typically struggle with rare classes and cannot recognize new ones. Unsupervised and open-vocabulary segmentation, proposed to tackle these issues, faces challenges, including the inability to assign specific class labels to clusters and the necessity of user-provided text queries for guidance. In this context, we propose a novel approach, TAG which achieves Training, Annotation, and Guidance-free open-vocabulary semantic segmentation. TAG utilizes pre-trained models such as CLIP and DINO to segment images into meaningful categories without additional training or dense annotations. It retrieves class labels from an external database, providing flexibility to adapt to new scenarios. Our TAG achieves state-of-the-art results on PascalVOC, PascalContext and ADE20K for open-vocabulary segmentation without given class names, i.e. improvement of +15.3 mIoU on PascalVOC.

Malicious hosts have come to play a significant and varied role in today’s cyber attacks. Some of these hosts are equipped with a technique called cloaking, which discriminates between access from potential victims and others and then returns malicious content only to potential victims. This is a serious threat because it can evade detection by security vendors and researchers and cause serious damage. As such, cloaking is being extensively investigated, especially for phishing sites. We are currently engaged in a long-term cloaking study of a broader range of threats. In the present study, we implemented Stargazer, which actively monitors malicious hosts and detects geographic and temporal cloaking, and collected 30,359,410 observations between November 2019 and February 2022 for 18,397 targets from 13 sites where our sensors are installed. Our analysis confirmed that cloaking techniques are widely abused, i.e., not only in the context of specific threats such as phishing. This includes geographic and time-based cloaking, which is difficult to detect with single-site or one-shot observations. Furthermore, we found that malicious hosts that perform cloaking include those that survive for relatively long periods of time, and those whose contents are not present in VirusTotal. This suggests that it is not easy to observe and analyze the cloaking malicious hosts with existing technologies. The results of this study have deepened our understanding of various types of cloaking, including geographic and temporal ones, and will help in the development of future cloaking detection methods.

Recent innovations in wearable devices have expanded the usage opportunities of single-channel electrocardiography (ECG) recordings in a daily life environment and enabled a variety of indirect daily activity monitoring based on heart rate variability (HRV). In general, wearable ECGs rarely undergo visual inspection by medical experts and therefore may contain noise or artifacts. Although noise/artifact-induced changes in ECG waveforms are known to cause misdetection of the QRS complex (i.e., the most distinguishable ECG components comprised of Q wave, R wave, and S wave), its complete suppression might be technically impossible. Since misdetection occurs in the QRS complex unit, we propose reframing the traditional HRV analysis flow by subdividing the R-R interval (RRI) editing into four steps in accordance with the processing detail (i.e., identification and editing) and its target unit (i.e., QRS complex or RRI). In addition, as a dubious QRS complex identification method for practical use, we utilize the amplitude at the detected point assuming the use of a single-channel wearable ECG without a reference. Initial evaluations using pseudo/real ECG datasets including ECGs with noise/artifacts show that the proposed processing/unit-based subdivision is theoretically effective for improving HRV calculation accuracy, and that the dubious QRS complex identification method for practical use also maintains this effect. Our study starting from practical HRV analysis using single-channel wearable ECG devices encourages reexamining each step in HRV analysis through the interdisciplinary research of clinical medicine and engineering/informatics that reveals the relationship of every two adjacent steps from the perspective of theory and practice.

Despite constant advances in X-ray crystallography, the resolution of the acquired electron density maps still poses a serious limit on protein protein structural model building efforts. Furthermore, the currently available toolkits require hours or even days for model building. Methods capable of processing a large volume of samples in a short time which can also handle low resolution samples are needed. This work proposes a neural network-based approach to locate and classify residues in crystallographic electron density maps automatically in a single forward pass without relying on the protein’s residue sequence. Our proposed method shows an average 23.53% increase in accuracy over our previous approach and also compares favorably to currently available toolkits. It can process protein samples in seconds on consumer-grade hardware saving significant time and resources.

Digital identities, including names and age, provide modern information systems with valuable data. Identity management is a security feature that enables users to manage and utilize their digital identities when interacting with online services. A self-sovereign identity (SSI) system is a cutting-edge identity model that uses blockchain technology to foster peer-to-peer trust in service authentication and authorization. SSI systems typically adhere to a set of functional- and quality-related principles and properties. However, we discovered that the current information security and privacy principles and properties of SSI systems are not compliant with the laws, regulations, and technical standards. A compliance set of principles and properties must be established to support the implementation of SSI systems and to improve security and privacy. In this article, we propose CSSPS, a new compliance SSI system property set that expands upon the missing security and privacy controls specified in applicable laws, regulations, and technical standards. We used systematic comparative analysis and systematic review to identify inconsistent content from a vast collection of applicable sources, and then used them to extend the current properties of the SSI system. The proposed CSSPS increases the consistency of security and privacy controls, and is applicable in accordance with the functionality of real SSI systems, as determined by a qualitative evaluation. The proposed CSSPS contributes to SSI system implementation by facilitating correct implementation, while adhering to applicable information security and privacy sources. In addition, the proposed CSSPS can indirectly enhance the security and privacy of SSI systems.

In this study, cosimulation with Q3D and Simplorer were adopted to design a single-phase power stack with equal output current on three paralleled IGBT modules in high power application, therefore, the derating of total current will be reduced, which can improve the current utilization and power density of power stack. The performance of the designed stack was verified using a double-pulse test (DPT). The circuit used in the DPT regarded the power stack output current as static current, which is a critical index to evaluate current sharing. The static current sharing among the IGBT modules was mainly dependent on the stray inductance of the DC busbar, IGBT modules, and phase output bar (POB), which is used to connect the IGBT modules to load, in parallel paths of the power stack. Cosimulation was performed to determine the stray inductance of the IGBT modules, DC busbar and POB, and a constant-current-slope method was used to verify the inductance. Subsequently, design the shapes of DC busbar and POB for attaining maximum ratio of imbalanced current (MRIC) of the three IGBT modules is within 3% under 200% (2400A) rated current. The cosimulation results indicated that the MRIC of the current paths in power stack were 0.8%. Finally, three IGBT modules and a gate driver were used to construct power stack with 1000V DC link. The experimental results obtained by DPT indicated that the MRIC was 0.9% under 200% rated current, demonstrating the effectiveness of the proposed stray inductance design method.

This work proposes a neural network architecture, called 3D FC-DenseNet, for assigning amino acid labels to X-ray crystallographic electron density maps without relying on the amino acid sequence of proteins. The 3DFC-DenseNet is able to treat the task as a 3D semantic segmentation problem, assigning amino acid labels directly to protein electron density maps. By creating dedicated data sets and models for high, medium and low resolution samples, our method matches the performance of crystallographic toolkits for primary structure assignment at high resolutions. Furthermore, it outperforms them at medium resolution and functions at low resolutions where current toolkits and human ability fails.

Autonomous driving systems are highly expected to be used on public roads to improve traffic throughput and road safety, but it will likely take a long transition period before all human-driven vehicles can be replaces with autonomous vehicles. Hence, CAVs have to safely cooperate with the surrounding human drivers, in order to gain the benefits of autonomous driving technologies during such transition periods. In this paper, we present a Distributed Synchronous Intersection Protocol (DSIP) and a Cooperative Perception-based High-Definition Map (CP-HD Map) for mixed traffic environments, in which autonomous vehicles co-exist with human-driven vehicles. First, in DSIP, each CAV utilizes dynamic decision-making mechanisms to adaptively change the vehicle behaviors based on the surrounding environments by using vehicle states and vehicle mode. In addition, in CP-HD Map, each CAV uses Vehicle-to-Vehicle (V2V) communications to share the information of detected objects to improve the road safety in the mixed traffic environments. Under these protocols, human-driven vehicles simply follow the traffic lights just like they do today. Finally, we show that DSIP and CP-HD Map increase the traffic throughput around the road intersections when we compared to existing signalized intersections and other V2V communications-based protocols.

We have shown that the flow velocity of Shirase Glacier can be estimated by applying the offset tracking method to the amplitude images derived from the phased array type L-band synthetic aperture radar type-2 (PALSAR-2) onboard the Advanced Land Observing Satellite-2. Although the offset tracking is widely used, the method is sometimes not applicable effectively and the reason is not always clear. Here, we consider a possibility of ionospheric contamination and show the validation results of the estimated flow velocity applied the offset tracking method to the PALSAR-2 data, based on in-situ measurement of the flow velocity from the GNSS receiver on the autonomous phase-sensitive radio echo sounder simultaneously with the PALSAR-2 observation. As a result, the ionospheric contamination can yield a spurious error of ±0.2 km a−1. After statistically removing the erroneous pairs, the RMSE was 0.049 km a−1 derived from the flow velocity error between the in-situ measurement results and estimated from satellite, and the estimated flow velocity obtained using PALSAR-2 data is proved to be in good agreement with the ground truth. Obtained spatial and temporal variations reveal signature of glacier dynamics, which prove the efficacy of the offset tracking method. The obtained ice-flow velocity increases rapidly from the upstream region to the coast, but its velocity is roughly constant over a region, 10-km long about the grounding line (GL), then gradually tends to increase again downstream from the GL. This trend has continued largely unchanged over24 years since 1996.

Generative image extension has the advantage of extending the overall image size while preserving the target image because, unlike other image extensions using interpolation, it completes the surroundings of the target image. However, existing generative image extension methods tend to have poor quality in the generation of outer pixels. One method deals only with a limited number of scene classes because the extension repeats the same semantics. We propose a mirrored input, which sandwiches the extended region by mirroring a part of the target image. This replaces generative image extension with an image inpainting problem and thus helps to achieve higher quality pixel generation and can extend semantics with more complex shapes than horizontal repetition. Experimental results show that our proposed method achieves a scenery image extension that exceeds the state-of-the-art generative image extension methods in both visual quality and FID score for datasets containing diverse scenes.

Recent innovations in wearable electrocardiogram (ECG) devices have enabled various personal healthcare applications based on heart rate variability (HRV). However, wearable ECGs rarely undergo visual inspection by medical experts, hence may contain noise and artifacts. Because apparent changes in the recorded ECGs caused by noise and artifacts may hamper the extraction of QRS complexes, an R-R interval (RRI) estimation algorithm tolerant to these measurement faults is required as the initial step toward HRV analysis using wearable ECGs. This paper proposes a semi-real-time RRI estimation for wearable ECGs utilizing a two-stage structure. In the preprocessing stage, we use a complex-valued wavelet that can adaptively fit to morphological variations of the QRS complex while retaining computing resources for extracting the QRS complex features. In the decision stage, we make use of complex-valued features and select appropriate QRS complexes in consideration of three features: peak magnitude, peak location, and peak morphology (phase). Initial evaluations show that the QRS complex detection performance of the proposed method achieved the F1 score of 0.952 ± 0.040 when targeting pseudo ECG data created from open data assuming wearable ECGs, and of 0.986 ± 0.018 when targeting actual ECG data recorded by a shirt-type wearable ECG device during an exercise activity. Furthermore, the proposed method was able to suppress overlook or misdetection of QRS complexes, so the obtained RRIs are closer to the reference RRIs. The proposed method therefore contributes to achieving accurate HRV analysis using wearable ECGs in terms of obtaining accurate RRIs.

Distribution networks are suffering a transformation process with the insertion of distributed energy resources, changing from a passive part of the power grid into an active system. Aiming to reduce the challenges for the distribution system operators to keep the power grid operating inside of the quality levels it is possible to propose some ancillary services. Optimum Power flow is an optimization method used to plan the distribution grid operation. Additionally, the distributed energy resources can be modeled on it and ancillary services provisioning can be considered, as proposed in this work. In this way, the objective of the present work is to formulate a Multiperiod Optimum Power Flow (MPOPF) with the insertion of distributed generation, energy storage systems, microgrids, and electric vehicles at the distribution grid. This MPOPF considers the provisioning of ancillary services by the inverters associated with the equipment connected to the main grid. In the formulation, the entire grid is modeled, considering the placement of classic equipment as a voltage regulator and capacitor banks, in addition to modern technologies as DFACTS (Distribution - Flexible AC Transmission System) and four-quadrant inverters. The MPOPF was simulated for several scenarios considering a 90-bus test feeder and a real distribution grid from Curitiba – Brazil. From the results, the MPOPF proved to be highly robust, being able to simulate the grid with all the equipment connected simultaneously, performing the optimal dispatch of active and reactive power, as well as allowing the operation of ancillary services such as voltage support, peak-shaving, and demand-side management.

This paper describes a deep learning approach to classify physically fatigued and non-fatigued gait cycles via a recurrent neural network (RNN), where each gait cycle is represented as a time series of three-dimensional coordinates of body joints. Gait cycles inherently have large intra-class variations caused by gait stance differences (e.g., which foot is supporting/swinging) at the beginning of each gait cycle, which makes it difficult to identify subtle differences induced by fatigue. To overcome these difficulties, we introduce a supporting foot-aware RNN model in a multi-task learning framework for better fatigue detection. More specifically, the RNN model has two branches of layers: one is assigned to the main task of fatigue classification and the other is assigned to the auxiliary task of estimating the first supporting foot in the gait cycles. We collected physically fatigued and non-fatigued gait cycles from eight subjects and conducted experiments to evaluate the accuracies of the proposed multi-task model in comparison to a single-task model. As a result, the proposed method achieved an overall area under curve (AUC) of 0.860 for fatigue classification in a leave-one-subject-out cross-validation, and an AUC of 0.915 in a leave-one-day-out evaluation. It can be concluded from the experimental results that a fatigue detection system for daily use, especially for screening purposes, is very feasible on the basis of the proposed approach.

The Mini-Mental State Examination (MMSE) is widely used in clinics to screen for low cognitive status. However, it is limited in that it requires examiners to be present; and has fixed questions that constrain its repeated use. Thus, the MMSE cannot be used as a daily assessment to facilitate early detection of cognitive impairment. To address this issue, we developed an automated system to detect older adults with lower MMSE scores by analyzing performance during a dual task involving stepping and calculation, which can be used repeatedly because its questions were randomly created. Leveraging this advantage, this paper proposes a learning-based method to detect subjects with lower MMSE scores using multiple trials with the dual-task system. We investigated various patterns for effectively combining the features acquired during multiple continuous trials, and analyzed the sensitivity of the number $N$ of trials on detection performance to find the optimal $N$ via experiments. We compared our approach with previous methods and demonstrated the superiority of our strategy. Using the cross-trial feature, our approach achieved an overall performance (sensitivity + specificity) as high as 1.79 for detecting older adults whose MMSE score is equal to or less than 23 (indicate a relatively high probability of dementia), and 1.75 for detecting older adults whose MMSE score is equal to or less than 27 (indicative of a relatively high probability of mild cognitive impairment (MCI)).

Due to the declining population and the aging of the farmer population in the hilly and mountainous areas of Japan, it is necessary for the elderly to carry out the mowing work on the ridges and slopes, which is traditionally regarded as heavy labor as part of paddy farming. One of the most important causes of these accidents is an incorrect mowing posture; therefore, it is important and necessary to identify effective and safe working patterns during inclined plane mowing. In this paper, we designed and implemented a set of mowing experiments in a terraced field area in Hiroshima to collect information on the body motion of experienced elderly mowing workers via a high-precision motion capture device that supports the collection of information from 23 joints. According to an analysis that calculated the angles of the workers' joints during mowing, we confirmed the characteristics of the mowing workers' working patterns in three different situations (typical inclined plane mowing (TI), top-down mowing (TD) and bottom-up mowing (BU)). The comparative analysis indicates that the basic actions “c” (cutting) and “t” (throwing) are basically the same in terms of body posture for the situation TI, but for situation TD, the difference was observed with respect to the workers' right ankle. Moreover, based on the comparation analysis for mowing action “c” (cutting), we confirmed that mowing workers should: keep their lower bodies as still as possible to keep balance for ensuring safety while working on inclined plane (TI); keep careful even if they are working on the flat ground (TD); and do not exert their utmost strength to mow, unless they are standing on the flat ground (BU). The findings of this work should be emphasized in the future development of mowing support systems and training programs for new mowing workers.

Energy and economy are increasing the relationship over the years, where the energy becomes a significant resource to keep a country developing, and it supports its economy. Then, more reliable the energy should become, especially the distribution network, to keep the entire process running. In this level of energy distribution, where residential consumers and medium and small industries are supplied, the number of interconnections of the network is enormous. However, for economic and environmental aspects, these complex systems, which are operating close to their capacity, needs to increase the automation, appearing the concept of smart grids and the Advanced Distribution Management System (ADMS) and its methods to control. Inside of the ADMS, there are a lot of essential techniques. Among them, there are two techniques which are the most relevant for this paper: the self-healing and load management. In an ADMS system, these two techniques are treated separately, but the best solution occurs when they are computed together. In this paper, it is proposed an approach that can address both problems at the same time or individually, i.e., in place to have a sequential method to solve step-by-step the issues in the networks. The proposed algorithm, through reinforcement learning technique, can handle both problems together. The proposed approach is tested in a real urban distribution network with some created scenarios to compare the results with outages and overloads. Some comparisons with other methods are carried out.

The dual-task paradigm is a promising procedure for estimating cognitive status and may also be collaterally used to reduce cognitive decline and prevent dementia. In this paper, we use the mini-mental state exam (MMSE) to the assess cognitive status in the elderly as a reference and investigate the potential of using machine learning for early detecting cognitive impairment in the elderly. Although many studies have suggested that dual-task performance, in which participants perform a cognitive task while walking, is associated with cognition, they only considered the correlation between cognitive parameters and simple gait feature, such as gait speed, through the statistical analysis. We instead use a Kinect sensor to capture participants' whole-body movements and extract a rich gait feature that has the ability to exhibit different tendencies of movements between healthy and cognitive-impaired elderlies. In our experiments, a classifier based on the dual-task gait feature achieved a higher performance than the one based on the single-task feature; the performance of the rich gait feature was better than that of a simple one, and; an optimal detection performance was achieved with an MMSE cutoff score of 25. We positively validated that the proposed method could early detect elderly with lower MMSE scores based on dual-task gait feature with a promising performance. Our approach can support early and automated diagnosis of cognitive impairment.

Gaze estimation using monocular cameras has significant commercial applicability, and many studies have been undertaken on head pose-invariant and calibration-free gaze estimation. The head positions in existing data sets used in these studies are, however, limited to the vicinity of the camera, and methods trained on such data sets are not applicable when subjects are at greater distances from the camera. In this paper, we create a room-scale gaze data set with large variations in head poses to achieve robust gaze estimation across a broader range of widths and depths. The head positions are much farther from the camera, and the resolution of the eye image is lower than in conventional data sets. To address this issue, we propose a likelihood evaluation method based on edge gradients with dense particles for iris tracking, which achieves robust tracking at low-resolution eye images. Cross-validation experiments show that our proposed method is more accurate than conventional methods on all the individuals in our data set.