GPUs Research Articles

Spatiotemporal Pixelization to Increase the Recognition Score of Characters for Retinal Prostheses.

Most of the retinal prostheses use a head-fixed camera and a video processing unit. Some studies proposed various image processing methods to improve visual perception for patients. However, previous studies only focused on using spatial information. The present study proposes a spatiotemporal pixelization method mimicking fixational eye movements to generate stimulation images for artificial retina arrays by combining spatial and temporal information. Input images were sampled with a resolution that was four times higher than the number of pixel arrays. We subsampled this image and generated four different phosphene images. We then evaluated the recognition scores of characters by sequentially presenting phosphene images with varying pixel array sizes (6 × 6, 8 × 8 and 10 × 10) and stimulus frame rates (10 Hz, 15 Hz, 20 Hz, 30 Hz, and 60 Hz). The proposed method showed the highest recognition score at a stimulus frame rate of approximately 20 Hz. The method also significantly improved the recognition score for complex characters. This method provides a new way to increase practical resolution over restricted spatial resolution by merging the higher resolution image into high-frame time slots.

Spatiotemporal Pixelization to Increase the Recognition Score of Characters for Retinal Prostheses

Most of the retinal prostheses use a head-fixed camera and a video processing unit. Some studies proposed various image processing methods to improve visual perception for patients. However, previous studies only focused on using spatial information. The present study proposes a spatiotemporal pixelization method mimicking fixational eye movements to generate stimulation images for artificial retina arrays by combining spatial and temporal information. Input images were sampled with a resolution that was four times higher than the number of pixel arrays. We subsampled this image and generated four different phosphene images. We then evaluated the recognition scores of characters by sequentially presenting phosphene images with varying pixel array sizes (6 × 6, 8 × 8 and 10 × 10) and stimulus frame rates (10 Hz, 15 Hz, 20 Hz, 30 Hz, and 60 Hz). The proposed method showed the highest recognition score at a stimulus frame rate of approximately 20 Hz. The method also significantly improved the recognition score for complex characters. This method provides a new way to increase practical resolution over restricted spatial resolution by merging the higher resolution image into high-frame time slots.

Identification of the intensity of air pollution sources based on hybrid computer systems

The method of identification of the intensity of the sources of chemically interacting pollutants is presented. The implemented model includes the phenomenon of self-purification in reaching the limit concentration. For computational implementation the possibility is shown of using parallel methods based on Nvidia CUDA graphic processing units. The method of source identification combined with the parallel computing implementation using the modified red-black ordering (D4) method reduces simulation time by 12 times and the RAM usage by 30% when using the Nvidia c2050 graphics accelerator in comparison with the node of the NTUU "KPI" cluster.

Communication and Load Balancing Optimization for Finite Element Electromagnetic Simulations Using Multi-GPU Workstation

This paper considers a method for accelerating finite-element simulations of electromagnetic problems on a workstation using graphics processing units (GPUs). The focus is on finite-element formulations using higher order elements and tetrahedral meshes that lead to sparse matrices too large to be dealt with on a typical workstation using direct methods. We discuss the problem of rapid matrix generation and assembly, as well as accelerating preconditioned iterative solvers in the context of limited on-board GPU memory, and we show how to mitigate some of these problems using multiple GPUs. We propose a new fast data-distribution technique for multi-GPU platforms that allows optimal splitting of finite-element method (FEM) matrices between graphics accelerators. The technique draws upon the graph partitioning approach used in nonoverlapping domain-decomposition methods and provides information that drives the FEM matrix-generation and assembly process in such a way that it produces data structures for each GPU; this not only ensures load balancing and minimizes communication between GPUs, but also reflects the hierarchy of the basis functions. The concepts proposed in this paper are illustrated with examples involving sparse matrices of up to 13.9 million rows and over a billion nonzero elements.

GPU-Accelerated LOBPCG Method with Inexact Null-Space Filtering for Solving Generalized Eigenvalue Problems in Computational Electromagnetics Analysis with Higher-Order FEM

AbstractThis paper presents a GPU-accelerated implementation of the Locally Optimal Block Preconditioned Conjugate Gradient (LOBPCG) method with an inexact nullspace filtering approach to find eigenvalues in electromagnetics analysis with higher-order FEM. The performance of the proposed approach is verified using the Kepler (Tesla K40c) graphics accelerator, and is compared to the performance of the implementation based on functions from the Intel MKL on the Intel Xeon (E5-2680 v3, 12 threads) central processing unit (CPU) executed in parallel mode. Compared to the CPU reference implementation based on the Intel MKL functions, the proposed GPU-based LOBPCG method with inexact nullspace filtering allowed us to achieve up to 2.9-fold acceleration.

Resolving the Trade-off Between Visual Sensitivity and Spatial Acuity-Lessons from Hawkmoths.

The visual systems of many animals, particularly those active during the day, are optimized for high spatial acuity. However, at night, when photons are sparse and the visual signal competes with increased noise levels, fine spatial resolution cannot be sustained and is traded-off for the greater sensitivity required to see in dim light. High spatial acuity demands detectors and successive visual processing units whose receptive fields each cover only a small area of visual space, in order to reassemble a finely sampled and well resolved image. However, the smaller the sampled area, the fewer the photons that can be collected, and thus the worse the visual sensitivity becomes-leading to the classical trade-off between sensitivity and resolution. Nocturnal animals usually resolve this trade-off in favour of sensitivity, and thus have lower spatial acuity than their diurnal counterparts. Here we review results highlighting how hawkmoths, a highly visual group of insects with species active at different light intensities, resolve the trade-off between sensitivity and spatial resolution. We compare adaptations both in the optics and retina, as well as at higher levels of neural processing in a nocturnal and a diurnal hawkmoth species, and also give a perspective on the behavioral consequences. We broaden the scope of our review by drawing comparisons with the adaptive strategies used by other nocturnal and diurnal insects.

Independent sampling for Bayesian normal conditional autoregressive models with OpenCL acceleration

Abstract A new computational strategy produces independent samples from the joint posterior distribution for a broad class of Bayesian spatial and spatiotemporal conditional autoregressive models. The method is based on reparameterization and marginalization of the posterior distribution and massive parallelization of rejection sampling using graphical processing units (GPUs) or other accelerators. It enables very fast sampling for small to moderate-sized datasets (up to approximately 10,000 observations) and feasible sampling for much larger datasets. Even using a mid-range GPU and a high-end CPU, the GPU-based implementation is up to 30 times faster than the same algorithm run serially on a single CPU, and the numbers of effective samples per second are orders of magnitude higher than those obtained with popular Markov chain Monte Carlo software. The method has been implemented in the R package CARrampsOcl. This work provides both a practical computing strategy for fitting a popular class of Bayesian models and a proof of concept that GPU acceleration can make independent sampling from Bayesian joint posterior densities feasible.

Interactive Web-Based 3D Wellbore Viewer Enables Collaborative Analysis

Summary Wellbore visualizations are often depicted by use of 2D perspective views. This approach is limited and potentially misleading. Of particular interest to rod-pumping and progressing-cavity pumps is the amount of side loading placed on the rod string throughout the wellbore. By illustrating the trajectory of a wellbore in 2D, some nuanced geometries can be easily lost. For example, a “corkscrew” wellbore, or any other complex wellbore geometry, can be very difficult to depict through simple perspective views. A corkscrew is best visualized at a specific camera angle where the shape is apparent. This best viewing angle is generally not at the simple 90° angle where typical 2D perspective views are generated. Distributed wellbore-visualization tools can be applied to both the design and analysis of pump installations. Interactive and shared views help to identify points in the well where problems may occur, or have occurred. Strategic placement of downhole equipment such as pump depth relative to perforations and historical fluid levels can be performed in a more intuitive way, taking into account a more accurate understanding of the geometry in a given well. The use of standards-based modern Web technologies enables the rapid deployment of such a tool to both desktop and mobile devices. The 3D graphics acceleration was previously only available in the Web browser through plugins and other third-party software installations. The 3D graphics can now be rendered natively in most modern Web browsers without the need for any additional software installation. Examples of such native visualizations can be found in Fig. 1.

OpenCL implementation of a high performance 3D Peridynamic model on graphics accelerators

Parallel processing is one of the major trends in the computational mechanics community. Due to inherent limitations in processor design, manufacturers have shifted towards the multi- and many-core architectures. The graphics processing units (GPUs) are gaining more and more popularity due to high availability and processing power as well as maturity of development tools and community experience. In this research we describe a rather general approach to using OpenCL implementation of 3D Peridynamics model on GPU platform. Peridynamics is a non-local continuum theory for describing the behavior of material used especially when damage and crack nucleation or propagation is of interest. The steps taken for developing an OpenMP code from the serial one as well as the comparison between OpenCL and OpenMP codes are provided. Optimization techniques and their effects on the performance of the code are described. The implementations are tested on some 3D benchmarks with hundred of thousands to millions of nodes. The behavior of codes in terms of being memory or compute bound are analyzed. In all test cases reported, the OpenCL implementation consistently outperforms serial and OpenMP ones and paves the road for the development of high performance Peridynamics codes.

A real-time image optimization strategy based on global saliency detection for artificial retinal prostheses

Current retinal prostheses can only generate low-resolution visual percepts constituted of inadequate phosphenes which are elicited by a limited number of stimulating electrodes and with unruly color and restricted grayscale. Fortunately, for most retinal prostheses, an external camera and a video processing unit are employed to be essential components, and allow image processing to improve visual perception for recipients. At present, there have been some studies that use a variety of sophisticated image processing algorithms to improve prosthetic vision perception. However, most of them cannot achieve real-time processing due to the complexity of the algorithms and the limitation of platform processing power. This greatly curbs the practical application of these algorithms on the retinal prostheses. In this study, we propose a real-time image processing strategy based on a novel bottom-up saliency detection algorithm, aiming to detect and enhance foreground objects in a scene. Results demonstrate by verification of conducting two eye-hand-coordination visual tasks that under simulated prosthetic vision, our proposed strategy has noticeable advantages in terms of accuracy, efficiency, and head motion range. The study aims to help develop image processing modules in retinal prostheses, and is hoped to provide more benefit towards prosthesis recipients.

PolarGlobe: A web-wide virtual globe system for visualizing multidimensional, time-varying, big climate data

ABSTRACTThe increasing research interest in global climate change and the rise of the public awareness have generated a significant demand for new tools to support effective visualization of big climate data in a cyber environment such that anyone from any location with an Internet connection and a web browser can easily view and comprehend the data. In response to the demand, this paper introduces a new web-based platform for visualizing multidimensional, time-varying climate data on a virtual globe. The web-based platform is built upon a virtual globe system Cesium, which is open-source, highly extendable and capable of being easily integrated into a web environment. The emerging WebGL technique is adapted to support interactive rendering of 3D graphics with hardware graphics acceleration. To address the challenges of transmitting and visualizing voluminous, complex climate data over the Internet to support real-time visualization, we develop a stream encoding and transmission strategy based on video-compression techniques. This strategy allows dynamic provision of scientific data in different precisions to balance the needs for scientific analysis and visualization cost. Approaches to represent, encode and decode processed data are also introduced in detail to show the operational workflow. Finally, we conduct several experiments to demonstrate the performance of the proposed strategy under different network conditions. A prototype, PolarGlobe, has been developed to visualize climate data in the Arctic regions from multiple angles.

An investigation of GPU-based stiff chemical kinetics integration methods

A fifth-order implicit Runge–Kutta method and two fourth-order exponential integration methods equipped with Krylov subspace approximations were implemented for the GPU and paired with the analytical chemical kinetic Jacobian software pyJac. The performance of each algorithm was evaluated by integrating thermochemical state data sampled from stochastic partially stirred reactor simulations and compared with the commonly used CPU-based implicit integrator CVODE. We estimated that the implicit Runge–Kutta method running on a single Tesla C2075 GPU is equivalent to CVODE running on 12–38 Intel Xeon E5-4640 v2 CPU cores for integration of a single global integration time step of 10−6s with hydrogen and methane kinetic models. In the stiffest case studied—the methane model with a global integration time step of 10−4s—thread divergence and higher memory traffic significantly decreased GPU performance to the equivalent of CVODE running on approximately three CPU cores. The exponential integration algorithms performed more slowly than the implicit integrators on both the CPU and GPU. Thread divergence and memory traffic were identified as the main limiters of GPU integrator performance, and techniques to mitigate these issues were discussed. Use of a finite-difference Jacobian on the GPU—in place of the analytical Jacobian provided by pyJac—greatly decreased integrator performance due to thread divergence, resulting in maximum slowdowns of 7.11−240.96×; in comparison, the corresponding slowdowns on the CPU were just 1.39−2.61×, underscoring the importance of use of an analytical Jacobian for efficient GPU integration. Finally, future research directions for working towards enabling realistic chemistry in reactive-flow simulations via GPU/SIMT accelerated stiff chemical kinetics integration were identified.

TOAST: Automatic tiling for iterative stencil computations on GPUs

SummaryThe stencil pattern is important in many scientific and engineering domains, spurring great interest from researchers and industry. In recent years, various optimizations have been proposed for parallel stencil applications running on graphics processing units (GPUs). In particular, tiling is a technique that can significantly enhance application performance by improving data locality and by reducing the volume of communication between host memory and GPU. In addition, tiling enables stencil applications to process inputs that are larger than the physical GPU memory. However, implementing tiling efficiently is complex, time‐consuming, and error‐prone. In this paper, we propose transparently optimized automatic stencil tiling (TOAST), an automatic tiling mechanism for iterative stencil computations running on GPUs; TOAST has 3 main benefits: (1) It incorporates an optimization model that seeks to maximize data reuse within tiles while respecting the amount of dynamically available GPU memory; (2) it offers a virtualized GPU memory for stencil computations, allowing for large input data; and (3) it performs optimal tiling transparently to the developer of the parallel stencil application. The current implementation of TOAST augments the PSkel framework with an internal solver based on genetic algorithms. Our experimental results show that TOAST improves the performance of iterative stencil applications by up to 13 × compared with their multithreaded (central processing unit–based) optimized versions and up to 48 × compared with a naive tiling approach on GPU. The TOAST mechanism is able to automatically achieve a low percentual overhead of data management compared with actual stencil computation.

Copy–move forgery detection based on multi‐radius PCET

Copy–move, which copies part of an image and pastes to another part of the same image, is one of the most commonly used image tampering operations. The copied part may suffer to post-processing operations such as rotation, scaling and blur to make the forgery visually convincing. To detect the copied parts with large-scale rotation and scaling, this study proposes a copy–move forgery detection method based on multi-radius polar complex exponential transform (PCET). First, the multi-radius PCET with graphic processing unit acceleration is used to extract the rotational invariant and multi-scale features. Then, the lexicographical order matching algorithm, optimised with minimum heap is applied to get a coarse match result. After that, the accurate detection based on radius ratio and position information is used to get the accurate detection result. Compared with the state-of-the-art methods, the proposed method can detect the copied parts with large-scale rotation or scaling and is robust against Joint Photographic Experts Group (JPEG) compression, smoothing and noise degrading.

RETRACTED ARTICLE: A novel registration and super-resolution jointed paradigm for medical images under internet of thing environment

This paper proposes a novel registration and super-resolution jointed paradigm for medical images under the Internet of thing environment. In the medical image processing, the matching issue is one catches wide attention with the domain of research. Image registration technique can be divided into similarity measure, optimization, geometric transformation, and interpolation, etc. As the first essential clue of our model, we propose the novel registration algorithm based on energy feature extraction. Generally, the matching energy function by the similarity measurement and a penalty constitution is called the external force and endogenic force separately. The matching is an external force and endogenic force mutual competition, eventually achieves the balanced process. Furtherly, we integrate the game analysis and area feature selection to achieve the better image super-resolution mode through the pretreatment of the image to change the initial value, so as to achieve the purpose of improving the performance. Besides the algorithm level innovation, we integrate the GPU and the IOT to construct the hardware based implementation of the proposed medical image processing system. The latency of registers to read and write data across a GPU’s entire storage system is minimal, it is private to each thread, and can only be accessed by its owning thread. For each thread, the local memory is also private and it is often used to deal with the problem of overflow register, reducing the buffer overflow caused by the entire application of a substantial decline in the possibility and shared memory is visible to all threads within the thread block. We then achieve the optimal integration of IOT and GPU. The experimental result proves the robustness of the method.

Research on the Modeling Method for Digital Weaving Based on the Information of Physical Yarns and Fabric Pattern

The modeling method of digital weaving is one of the key technologies in digital textile, which is considered a great value in the textile industry. Based on the real yarn image information and the specified weave structure, this paper presents a method to realize that modeling for digital weaving. The imaging system has been designed to extract the information of yarn appearance of geometry characteristic. The basic cycle matrix is used to represents weave structure and the weaving matrix is for modeling. Weaving matrix can be obtained by extending the basic cycle matrix. According to the weaving matrix, the weaving network is built by the selected number of warp and weft yarns and their order of arrangement. Then the physical yarn image is divided into a set of nodes and is filled to the weaving network. By extracting the connecting matrix and stress matrix of warp and weft yarns respectively, the visual processing units (VPU) can be built. Finally, all VPUs are drawn according to the corresponding shape and illumination model. Through image splicing for all VPUs according to their corresponding position in weaving network, the modeling of digital weaving can be realized. Experiments demonstrate the validity of the method proposed in this paper.

3D model of flow in micro channels

The micro channels are important components in many micro - electro - mechanical systems (MEMS) and technical systems. They are used to transport a gas or a liquid into devices and the systems. In our work, the gas flows into a micro channel have been studied for various shapes and sizes of the micro channels. Numerical simulations of compressible flows through a micro channel are performed by solving the Navier–Stokes equations and the quasigasdynamic equations (QGD). The numerical approaches are realized on hybrid parallel computer systems. The software includes the Domain Decomposition technique, the Message Passing Interface (MPI) for organization of interprocess data exchange and an application programming interface Open Multi-Processing (OpenMP). The calculations are performed on the Heterogeneous Computer Systems with Video Processing Units (VPU) and the classical microprocessors - Central Processing Unit (CPU).

Mathematical Modeling of Building Constructions Using Hybrid Computing Systems

Consideration is given to algorithm for solving on computers of hybrid architecture (with multi-core processors and graphics accelerators) the systems of linear algebraic equations with band symmetrical positive definite matrices arising in the numerical modeling of 3D building constructions using the finite element method. Some results of the proposed hybrid algorithms testing for solving problems of static strength analysis of some building objects and individual constructions are presented.

Investigation of secondary cooling design enhancements in thermally limited compact notebooks

Thermal design enhancements in a thermally limited compact notebook system are investigated in this paper. System temperature, power, and fan speed are characterized under a range of activity levels. A finite element model is developed, and validated against measurements. Design enhancements improve cooling with minimum intrusion to the existing mechanical design. A passive secondary heat pipe in the system reduces the CPU temperature by 5 $^{\circ}$C, and improves the system performance through increased CPU + Graphics and Memory Controller Hub (GMCH) thermal design power (TDP) by 6.4%. When such a secondary heat pipe is considered with an integrated off-the-shelf Peltier cooler, the CPU temperature is only reduced by 2.3 $^{\circ}$C and CPU+GMCH TDP is improved only by 4.9%. Although Peltier integration provides no benefit to thermals, it can be advantageous in generating small amount of thermoelectric power in conditions when the system is not executing thermally limited applications. Calculations suggest that a 10% increase in Seebeck coefficient and consequently a 5.5% increase in coefficient of performance (COP) of off-the-shelf thermoelectric materials can increase the TDP envelop by 7.1% using the Peltier-integrated secondary heat pipe scheme.

On the Use of a GPU-Accelerated Mobile Device Processor for Sound Source Localization

The growing interest to incorporate new features into mobile devices has increased the number of signal processing applications running over processors designed for mobile computing. A challenging signal processing field is acoustic source localization, which is attractive for applications such as automatic camera steering systems, human-machine interfaces, video gaming or audio surveillance. In this context, the emergence of systems-on-chip (SoC) that contain a small graphics accelerator (or GPU), contributes a notable increment of the computational capacity while partially retaining the appealing low-power consumption of embedded systems. This is the case, for example, of the Samsung Exynos 5422 SoC that includes a Mali-T628 MP6 GPU. This work evaluates an OpenCL-based implementation of a method for sound source localization, namely, the Steered-Response Power with Phase Transform (SRP-PHAT) algorithm, on GPUs of this type. The results show that the proposed implementation, given the audio samples, is able to perform audio localization in real time with high-resolution spatial grids using up to 12 microphones.