Search

It is well-known that identity-unrelated variations (e.g., viewpoint or illumination) degrade the performances of face recognition methods. In order to handle this challenge, a robust method for disentangling the identity and view representations has drawn an attention in the machine learning area. However, existing methods learn discriminative features which require a manual supervision of such factors of variations. In this paper, we propose a novel disentangling framework through modeling three representations of identity, viewpoint, and residues (i.e., identity and pose unrelated) which do not require supervision of the variations. By jointly modeling the three representations, we enhance the disentanglement of each representation and achieve robust face recognition performance. Further, the learned viewpoint representation can be utilized for pose estimation or editing of a posed facial image. Extensive quantitative and qualitative evaluations verify the effectiveness of our proposed method which disentangles identity, viewpoint, and residues of facial images.

Few-shot classification is defined as a task where the network aims to classify unseen classes given only a few samples. Recent approaches, especially metric-based methods, have great progress in few-shot classification. However, the existing metric-based methods have a limitation in deploying discriminative features for elaborate comparison. They usually extract features from the embedding network without direct consideration of the relationship between support and query sets. To address the relationship, we propose a novel architecture, Style Correlated Module (SCM) to learn style correlation between support and query sets for few-shot classification. The proposed module leads support and query feature maps to focus on significant style correlated features and encourage the metric network to conduct an elaborate comparison. Furthermore, the proposed module can be generally applied to the existing metric-based approaches by adding the SCM behind the embedding network. We evaluate our proposed method with comprehensive experiments on two publicly available datasets and demonstrate its effectiveness with comparable results.

Supervised visual perception models require a large number of annotated inputs so that a model can be trained and validated. However, manual annotation of large-scale datasets can be prohibitive in terms of time and expense. We present an automatic fashion image tagging algorithm that uses detailed item descriptions and carefully selected category information produced by a service provider. Compared to previous work, which used a relatively simple frequency-based heuristic, a set of image attributes extracted by our proposed algorithm could be used to create a model (ResNet-18 architecture) with exhibited 8% higher accuracy than a model created with attributes suggested by DeepFashion. This in-progress work identifies opportunities that carefully chosen visually recognizable attributes can be used to help build a more general and descriptive model.

Ambulatory blood pressure monitoring (ABPM) has become essential to diagnose the hypertension. In the fast few decades, commercialized blood pressure monitor used oscillometric cuff-based methods which is cumbersome and unsuitable for ABPM due to the discomfort it cause to the patients. In this study, we developed the prototype of cuff-less blood pressure monitor to validate cuff-less BP estimation algorithm. Sixteen subjects were participated in the experiment to test the accuracy of the device compared to the cuff-based BP monitor using oscillometric method. Experimental results using Bland-Altman plot showed that the mean and standard deviation of error of systolic BP (SBP) and diastolic BP (DBP) were 0.3±6.6 and 1.7±9.1 respectively.

We studied limitation factors of high efficiency Cu 2 ZnSn(S,Se) 4 (CZTSSe) solar cells. Depending on the S/(S}+Se) ratio of absorber layer, different limitation factors are observed. The Se-rich CZTSSe solar cells exhibited larger conduction band offset (CBO) between the absorber and the buffer, whereas Sincreased solar cells showed deep defects in the absorber. The larger CBO in Se-rich CZTSSe solar cell and deep defects in Sincreased CZTSSe solar cell are observed even in $\sim12$ % efficiency solar cells. Thus, to obtain efficiency more than 12% in kesterite solar cell, these possible limitation factors should be resolved.

Due to the advancement of CMOS image sensors (CIS), camera module on the mobile platform has paved way for very high quality photos and video shooting capability. In order to improve picture quality, the CIS technology has been also scaling aggressively to provide more Mega-Pixels (MP) but it also must be less susceptible and immune to noise sources, particle and defect-free, and highly reliable. In this report, we'll discuss the reliability characterization done on the 3D stack sensor with in-pixel DTI. By optimization of both process and design, noise immune and very high quality advanced CIS products have been successfully qualified and manufactured for volume production.

The mesh network has propagated rapidly, because of the Internet. All over the world, many people cannot use the Internet, because the Internet cable is not installed. So many mesh network communities try to set up the mesh network nodes and manage it. By its effort, there are more than 50 thousand mesh nodes in the world. However, there are no enough people who can make the mesh network and can manage it. By this reason, we develop open source GOTHAM. It has three main part. There are “Self-Organizing”, “Visualization” and “Content-Centric”. G.O.T.H.A.M is designed for people who don't know the network. So, it focuses simplicity, familiarity and automaticity. In this paper we explain original G.O.T.H.A.M and new G.O.T.H.A.M with Hazelcast including three functions. And also, we describe the program which runs with GOTHAM. All of the software in this paper is available on Github.

In these days, the Internet is growing and the amount of user to use network increases rapidly. The mesh(ad-hoc) network has studied for many areas where people can't use the Internet. Also, mesh(ad-hoc) network is used for special purposes, and it expands its domain recently. However, low transfer rate and small coverage are a critical issue in mesh(ad-hoc) network. In this paper, it suggests improving transfer rate using BATMAN, Open vSwitch, and Dijkstra algorithm. This paper has three section. In the first section, it suggests network architecture for improving transfer rate. This structure consists of network, controller, and storage. In the second section, it explains the algorithm for finding the optimal path using modified Dijkstra algorithm. Finally, it shows the testbed to realize this proposal. We describe some advantage of controlling traffic flow.

In these days, wireless AP's signal is all around us. Multiple APs are set up at a big area, enterprise, university, movie theater, etc. and small area, such as a private house. So wireless APs are not special to us. However, the APs with several tools that can manage multiple APs are hard to find and also too expensive to the public. To solve this problem, this paper discusses how to install the multiple APs quickly and inexpensively and how to manage APs simply. There is no open source project about this problem ever before. Specifically, we proposed two ways. First, we explain a method, setting up the APs for a public network with the Internet and an agent program that manage APs. Second, we handle a method, setting up the APs for private(Internet-less) mesh network using P2P application at meeting room or concert hole. And also, we discuss the GOTHAM program which can visualize the network topology. All of the program in this paper, are open source software, HOSTAPD, DHCPD, OpenWRT, BATMAN, TOX, etc. And the programs, we made, are open source. So you can download all programs on Internet.

Existing NAT solves to IP address exhaustion problem binding private IP address and public IP address, and NAT traversal such as hole punching scheme enables to communicate End-to-End devices located in different private networks. However, such technologies are centralized the workload at NAT gateway and increase transmission delay caused by packet modification per packet. In this paper, we propose an SDN based fully distributed NAT traversal scheme, which can distribute the workload of NAT processing to devices and reduce transmission delay by packet switching instead of packet modification. Furthermore, we describe SDN based IoT connectivity management architecture for supporting IoT global connectivity and enhanced real-time and security.

Recent trends are to build tall buildings in big cities as a way out of the current housing overpopulation problem. These new structures unveil problems that if not addressed in time could cause catastrophes of unimaginable impact. One of those problems is the incidence of a fire threat happening upstairs in one of those buildings. Research solutions include implementation of multi-agent fire safety systems for fire threats in high rise buildings. Humans, agents, robots, machines and sensors communication model is used for indistinguishability of actors. This work discusses the implementation of the unmanned ground vehicles to spot the real location of the fire.

In this paper, we present an automatic piano tutoring system that uses a consumer-level depth camera to recognize keys that are touched by a performer in real-time. Accordingly, the system provides feedback to the performer as to whether the appropriate keys for playing a music score at the proper tempo are being touched. Our system can be installed on any piano system with a simple calibration process. As a result, any conventional acoustic piano can be converted into a piano tutor system with the addition of a low-cost depth camera.

RFID(Radio-Frequency Identification) is widely used in many industries. Furthermore, the many researches for RFID antenna have been studied. If the RFID antennas designed to increase RCS optimally, the performance of the RFID will be improved. Therefore, we need to analyze RFID tag chip having nonlinear characteristic. In this paper, we analyze the strip antenna using the MoM(Method of Moments). Then the strip antenna is converted into equivalent source model. Finally, the RF circuit including a nonlinear component is analyzed with harmonic balance method.

A robotic spatial augmented reality (RSAR) system combines robotics with spatial augmented reality (SAR) where cameras are used to recognize real objects, and projectors augment information and user interface directly on the surface of the real objects, rather than relying on mobile or wearable display devices. Hence, the control of a RSAR system requires handling of different types of control schemes at once such as classical inverse kinematics of simply linked bodies, inverse projections to find appropriate internal/external parameters of a projector, and geometric manipulation of a projection source image to increase the flexibility in control. In this paper, we outline a hybrid approach to control relevant control components in a coordinated manner, specially focused on application in a prototype RSAR system developed in ETRI.

We present a simple reconstruction algorithm of a plane-dominant 3D environment for the robotic spatial augmented reality (RSAR). In spatial augmented reality, a projector renders virtual objects onto 3D objects in the real space. To watch the augmented virtual objects from a viewpoint without distortions, the final image should be pre-distorted based on the geometry information of the 3D objects in the real world. In our RSAR setting, we assume that a robot is equipped with the devices such as a projector, low-cost depth cameras, and the sensor of capturing 3D position of the viewpoint. The robot captures the 3D environment with as a point cloud and reconstructs the geometry of 3D objects in the real world with a set of planes. In order that the viewer can see the distortion-free virtual objects, we compute the pre-warped images of virtual objects for a projector. Finally, we provide an efficient algorithm using GPU shaders to compute the pre-warped images for a projector. In experiments, we provide some results of preliminary simulations for our RSAR scenario.

Most loci discovered through genome-wide association analyses are predicted to affect gene expression, the integrative approach of genome-wide analysis with gene expression data is becoming essential procedure for discovering genetic effect of disease development. Many studies have been performed to discover significant SNP-gene associations, but most of them are limited to consider only cis-associations and neglect trans-territory. In this study, we explored the effect of trans-eSNP associations that may underlie the alternation of gene expressions. Through the integrative genome-wide association analysis considering the entire SNP-gene associations, we identified numerous trans associations which are significantly associated with gene expression, even more than cis associations in quantity and significance. Our findings revealed the necessity of reconsidering trans-association effect from integrative genome-wide association analysis and provided novel insights to find undiscovered genetic causalities.

Recently, ITS (Intelligent Transport System) has been studied actively which is useful for smooth traffic and safety drive of the various vehicles. Though the laser or imaging camera sensor was used for ITS, they have unstable performances and high complexity. Therefore, FMCW radar using millimeter wave has been proposed recently. Actually, 47, 60, 77, 94 and 139 GHz are assigned for the vehicle radar frequency in Europe and Japan. In this paper, FMCW radar was simulated and modeled with interference. Also, RER(Range Error Rate) was proposed to estimate the performance error due to the interference.

Direction estimation of signal of interest has been an important issue in radar and communication system. Generally, DOA(Direction Of Arrival) methods have been researched in the field of signal processing with ideal array sensors. However, there are some problems in array antennas such as the input signal distortions in amplitude and phase, due to the mutual coupling between array elements. Therefore, the compensation techniques of mutual coupling effects in the array antenna are required for the accurate estimation of DOA. In this paper, we propose a new method of the compensation of the mutual coupling effects for the DOA estimation in the dipole array antenna by using the method of moment (MoM). Also, the proposed method is applied to the estimation of azimuth (Ф) and elevation (θ) angles using uniformly linear dipole array under noisy environments.

We studied CuInS 2 (CIS) film growth using high electrostatic field assisted ultrasonic spray deposition. All the sprayed CIS films were observed to be grown with mixed ordering mode, where chalcopyrite (CH) and CuAu (CA) orderings coexisted. For the high efficiency solar cell, CH-ordered CIS film is desirable. We performed post-sulfurization for all as-sprayed films to induce recrystallization. After post-sulfurization of as-sprayed films, CH-ordered phase was significantly increased. For Cu-rich CIS films, the CH-fraction was increased to be ∼ 0.8. The scanning electron microscopy measurement revealed that grain size was also increased after post-sulfurization. These results indicate that spray deposition could be utilized for the fabrication of high quality CIS photo absorbing films.

We demonstrated the organic light-emitting diodes (OLED) with additive such as 2-propanol (IPA) in poly(3,4-ethylenedioxythiophene): polystyrene sulphonate (PEDOT:PSS). IPA mixed PEDOT: PSS showed a significantly decreased contact angle on ITO compared to those without IPA. On the other hand, the luminous efficiency of the devices using PEDOT: PSS mixed with 10 % and 20 % IPA were decreased due to the diluted solution of PEDOT: PSS. However, it solved by increasing the conductivity of PEDOT: PSS film with addition of dimethyl sulfoxide (DMSO).

In this paper, we propose a novel, systematic structure for quasi-orthogonal space-time block codes (QO-STBCs) achieving full rate for even number of transmit antennas over quasi-static Rayleigh fading channels. Especially, the new codes are delay optimal unlike conventional QO-STBCs and thus have a great advantage in ML decoding complexity at a receiver. In order to exploit full spatial diversity, we utilize a constellation rotation method and also examine optimal rotation angles through brute-force search for some cases of modulations and transmit antennas. Computer simulation results show that the proposed codes exhibit average error performance almost the same as or better than the existing QO-STBCs.

Ricci flow is a powerful curvature flow method in geometric analysis. This work is the first application of surface Ricci flow in computer vision. We show that previous methods based on conformal geometries, such as harmonic maps and least-square conformal maps, which can only handle 3D shapes with simple topology are subsumed by our Ricci flow based method which can handle surfaces with arbitrary topology. Because the Ricci flow method is intrinsic and depends on the surface metric only, it is invariant to rigid motion, scaling, and isometric and conformal deformations. The solution to Ricci flow is unique and its computation is robust to noise. Our Ricci flow based method can convert all 3D problems into 2D domains and offers a general framework for 3D surface analysis. Large non-rigid deformations can be registered with feature constraints, hence we introduce a method that constrains Ricci flow computation using feature points and feature curves. Finally, we demonstrate the applicability of this intrinsic shape representation through standard shape analysis problems, such as 3D shape matching and registration.

Chalcogen photo thin film transistor has been proposed. Using chalcogen alloys, to form a photoactive channel layer, high and low resistive CuInSe2 films are grown by thermal annealing and a three step process, respectively. Optical and electrical properties of the fabricated CuInSe2 films are investigated.

Multiresolution geometry streaming has been well studied in recent years. The client can progressively visualize a triangle mesh from the coarsest resolution to the finest one while a server successively transmits detail information. However, the streaming order of the detail data usually depends only on the geometric importance, since basically a mesh simplification process is performed backwards in the streaming. Consequently, the resolution of the model changes globally during streaming even if the client does not want to download detail information for the invisible parts from a given view point. In this paper, we introduce a novel framework for view-dependent streaming of multiresolution meshes. The transmission order of the detail data can be adjusted dynamically according to the visual importance with respect to the client's current view point. By adapting the truly selective refinement scheme for progressive meshes, our framework provides efficient view-dependent streaming that minimizes memory cost and network communication overhead. Furthermore, we reduce the per-client session data on the server side by using a special data structure for encoding which vertices have already been transmitted to each client. Experimental results indicate that our framework is efficient enough for a broadcast scenario where one server streams geometry data to multiple clients with different view points.

An optical recording device with an atomic force microscope (AFM) tip, having a sub wavelength aperture, has been considered as one of candidates for Th/in/sup 2/ data density in near-field optical storage technology. However, the low optical throughput of a near field aperture limits the practical use of the near-field optical storage technique. This limit can be avoided by the enhancement of the throughput of the near-field apertured tip and the development of new recording media with low phase change temperature. To increase the optical throughput of an AFM/near field apertured tip, several fabrication techniques have been proposed so far. In this paper we present a simple fabrication process of a near-field high throughput apertured tip using a bird's beak effect. Using the bird's beak effect and field oxidation, a higher throughput aperture and larger cone angled tip are fabricated, simultaneously.

Signal integrity (SI) is affected by the location of via in the vicinity of slot. In multiple planar layer circuitry, when a signal trace changes layer through via, whether there is a slot in the reference plane beneath the signal trace or not has effects on signal integrity (SI). In this paper, the improved efficiency on signal integrity (SI) according to the location of via in the vicinity of slot is seen through simulation and measurement. The reduction of slot efficiency to provide isolation of a noise source from the rest of the PCB is also shown when a via is placed through the slot. Finally, design rule for the location of via in the vicinity of slot in the nearest reference plane is also discussed.