Portrait Segmentation Using Ensemble of Heterogeneous Deep-Learning Models

Image segmentation plays a central role in a broad range of applications, such as medical image analysis, autonomous vehicles, video surveillance and augmented reality. Portrait segmentation, which is a subset of semantic image segmentation, is widely used as a preprocessing step in multiple applications such as security systems, entertainment applications, video conferences, etc. A substantial amount of deep learning-based portrait segmentation approaches have been developed, since the performance and accuracy of semantic image segmentation have improved significantly due to the recent introduction of deep learning technology. However, these approaches are limited to a single portrait segmentation model. In this paper, we propose a novel approach using an ensemble method by combining multiple heterogeneous deep-learning based portrait segmentation models to improve the segmentation performance. The Two-Models ensemble and Three-Models ensemble, using a simple soft voting method and weighted soft voting method, were experimented. Intersection over Union (IoU) metric, IoU standard deviation and false prediction rate were used to evaluate the performance. Cost efficiency was calculated to analyze the efficiency of segmentation. The experiment results show that the proposed ensemble approach can perform with higher accuracy and lower errors than single deep-learning-based portrait segmentation models. The results also show that the ensemble of deep-learning models typically increases the use of memory and computing power, although it also shows that the ensemble of deep-learning models can perform more efficiently than a single model with higher accuracy using less memory and less computing power.     

W 1, p(·)-Regularity for Elliptic Equations with Measurable Coefficients in Nonsmooth Domains

We establish global W ^{1, p(·)}-estimates for second order elliptic equations in divergence form under the natural assumption that p(·) is log-Hölder continuous. To this end, we assume that the coefficients are measurable in one variable and have small BMO semi-norms in the other variables and the boundary of the domain is Reifenberg flat. Our work is an optimal and natural extension of W ^1,p -regularity for such equations with merely measurable coefficients beyond Lipschitz domains.     

Synthesis and characterization of chemically modified polystyrene as processable carbon fiber precursors

In this study, the chemically modified polystyrene was studied for improvement of polystyrene which had low carbon yield, as a carbon fiber precursor. The polystyrene was synthesized with divinylbenzene which was used as a cross-linking agent by the solution polymerization method. Then the synthesized polystyrene was nitrated with sulfuric and nitric acids solution (H₂SO₄/HNO₃) followed by reduction to form nitrogen-functional groups. The surface properties of the modified polystyrene were investigated by Fourier transfer infrared spectroscopy to confirm the introduction of functional groups on the modified polystyrene surfaces. The thermal properties of the modified polystyrene were measured by thermogravimetric analysis and differential scanning calorimetry. The morphologies of the electrospun polystyrene fibers by a chemical modification were observed by scanning electron microscopy. From the results, the nitrogen-functional groups were introduced on the modified polystyrene surfaces which affected the quantity of functional groups. Also, the chemical treatment affected the carbon yield of the polystyrene owing to the introduction of nitrogen-functional groups on polystyrene surfaces. Consequently, it was concluded that the chemical treatment of polystyrene fibers enables it to be used as a possible carbon fiber precursor.     

Glyco‐pseudopolyrotaxanes: Carbohydrate Wheels Threaded on a Polymer String and Their Inhibition of Bacterial Adhesion

We report glyco‐pseudopolyrotaxanes composed of cucurbit[6]uril‐based mannose wheels (ManCB[6]) threaded on polyviologen (PV), which not only effectively induce bacterial aggregation, but also exhibit high inhibitory activity against bacterial binding to host cells. Three glyco‐pseudopolyrotaxanes ( 1 – 3 ), which have 10, 5, and 3 ManCB[6] wheels, respectively, on a PV string, were prepared and characterized. Bacterial aggregation assays and hemagglutination inhibition assays illustrated the specific and multivalent interaction between the glyco‐pseudopolyrotaxanes and E. coli ORN178. Compound 3 was especially effective at inducing bacterial aggregation and showed 300 times higher inhibitory potency than monomeric methyl‐α‐mannoside (Me‐αMan) for ORN178‐induced hemagglutination. Furthermore, we demonstrated their inhibitory activities for the adhesion of ORN178 bacteria to urinary epithelial cells as a model of urinary tract infection. Our findings suggest that these supramolecular carbohydrate clusters are potentially useful in antiadhesion therapy.     

Ambient xenon sampling using an Ag/ZSM-5 zeolite

Journal of Radioanalytical and Nuclear Chemistry - An ambient xenon sampling system using Ag/ZSM-5 was designed, and its performance was tested. The xenon collection efficiency of the system was...     

TMR of double spin-valve type AF/FM/I/FM/I/FM/AF magnetic tunneling junctions

An unusually large enhancement of TMR at 77 K was observed in double barrier tunnel junctions (DBTJ). This is explained with extended Julliere's model which yields a twice larger TMR value. When the spin coherence length is much smaller at higher temperature, DBTJ is shown to work as a series of two single barrier tunnel junctions.     

Switchable erbium-doped fiber ring laser based on Sagnac loop mirror incorporating few-mode high birefringence fiber

Without the need of any other additional filtering device, a triple-wavelength switchable erbium-doped fiber ring laser based on Sagnac loop mirror incorporating a piece of few-mode high birefringence fiber is newly proposed and experimentally demonstrated. Three potential lasing wavelengths at about 2.0 nm interval with the side mode suppression ratio higher than 40 dB can be switched mainly by adjusting a polarization controller next to the few-mode high birefringence fiber in the Sagnac loop. In addition to the stable single-line lasing operation, simultaneous dual- and triple-line oscillations are also highly stable at room temperature with suppressing the intrinsic homogeneous gain broadening of the erbium-doped fiber.