GPU-based Implementation Research Articles

Efficient image decolorization with a multimodal contrast-preserving measure

Decolorization is widely used in monotonic displaying, black-and-white printing, single-channel image processing, and many computer vision applications. The conversion of a color image into a grayscale one suffers from data loss. This paper presents a novel parallel decolorization method which effectively preserves not only the spatially local color contrast but also the dominant non-local color contrast. A new multimodal contrast-preserving measure with a multimodal Gaussian distribution is proposed to relax the constraint of color contrast. The dominant non-local color pair set is constructed by taking advantage of a linear bounding volume hierarchy while the local color pair set is produced by removing duplicate instances of local color pairs. The whole pipeline design is highly parallel, allowing for a real-time GPU-based implementation. Experimental results and comparisons show that the proposed method can produce plausible decolorized images. A number of single-channel image processing applications, including edge detection, image segmentation, color augmentation, and image stylization, are demonstrated to verify the feasibility of the proposed decolorization method.

GPU-based implementation of a diagnostic wind field model used in real-time prediction of atmospheric dispersion of radionuclides

In order to improve the prediction of atmospheric dispersion of radionuclides (ADR) in vicinity of Central Nuclear Almirante Alvaro Alberto (CNAAA) Brazilian Nuclear Power Plants (NPP), a more refined computational system is under development. To achieve desired refinement, the required computational effort increases in such a way that system's execution by current computers leads to prohibitive processing time. Aiming to accelerate execution of such refined system, allowing its effective use in real-time prediction of ADR, a GPU-based parallel approach has been proposed. Basically, the ADR system used in CNAAA is comprised by 4 main modules (programs): Source Term, Wind Field, Plume Dispersion and Plume Projection modules.This work is focused on the Wind Field module, which uses a mass-consistent approach, based on Winds Extrapolated from Stability and Terrain (WEST) model. Due to the strong sequential nature of the algorithm, domain decomposition by a 3D-Red-Black partitioning was proposed and a new parallel GPU-based algorithm was implemented using the Compute Unified Device Architecture (CUDA) and C programming language. As a result, the execution time of a fine-grained simulation decreased from about 450 s (running on an Intel-I7) to 18 s (running on a GTX-680 GPU). Here, the most important issues of the parallel implementation, as well as comparative results, are presented and discussed.

Adding Approximate Counters

We describe a general framework for adding the values of two approximate counters to produce a new approximate counter value whose expected estimated value is equal to the sum of the expected estimated values of the given approximate counters. (To the best of our knowledge, this is the first published description of any algorithm for adding two approximate counters.) We then work out implementation details for five different kinds of approximate counter and provide optimized pseudocode. For three of them, we present proofs that the variance of a counter value produced by adding two counter values in this way is bounded, and in fact is no worse, or not much worse, than the variance of the value of a single counter to which the same total number of increment operations have been applied. Addition of approximate counters is useful in massively parallel divide-and-conquer algorithms that use a distributed representation for large arrays of counters. We describe two machine-learning algorithms for topic modeling that use millions of integer counters and confirm that replacing the integer counters with approximate counters is effective, speeding up a GPU-based implementation by over 65% and a CPU-based implementation by nearly 50%, as well as reducing memory requirements, without degrading their statistical effectiveness.

Online tensor methods for learning latent variable models

We introduce an online tensor decomposition based approach for two latent variable modeling problems namely, (1) community detection, in which we learn the latent communities that the social actors in social networks belong to, and (2) topic modeling, in which we infer hidden topics of text articles. We consider decomposition of moment tensors using stochastic gradient descent. We conduct optimization of multilinear operations in SGD and avoid directly forming the tensors, to save computational and storage costs. We present optimized algorithm in two platforms. Our GPU-based implementation exploits the parallelism of SIMD architectures to allow for maximum speed-up by a careful optimization of storage and data transfer, whereas our CPU-based implementation uses efficient sparse matrix computations and is suitable for large sparse data sets. For the community detection problem, we demonstrate accuracy and computational efficiency on Facebook, Yelp and DBLP data sets, and for the topic modeling problem, we also demonstrate good performance on the New York Times data set. We compare our results to the state-of-the-art algorithms such as the variational method, and report a gain of accuracy and a gain of several orders of magnitude in the execution time.

Design and Implementation of a Stateful Network Packet Processing Framework for GPUs

Graphics processing units (GPUs) are a powerful platform for building the high-speed network traffic processing applications using low-cost hardware. The existing systems tap the massively parallel architecture of GPUs to speed up certain computationally intensive tasks, such as cryptographic operations and pattern matching. However, they still suffer from significant overheads due to critical-path operations that are still being carried out on the CPU, and redundant inter-device data transfers. In this paper, we present GASPP, a programmable network traffic processing framework tailored to modern graphics processors. GASPP integrates optimized GPU-based implementations of a broad range of operations commonly used in the network traffic processing applications, including the first purely GPU-based implementation of network flow tracking and TCP stream reassembly. GASPP also employs novel mechanisms for tackling the control flow irregularities across SIMT threads, and for sharing the memory context between the network interfaces and the GPU. Our evaluation shows that GASPP can achieve multigigabit traffic forwarding rates even for complex and computationally intensive network operations, such as stateful traffic classification, intrusion detection, and packet encryption. Especially when consolidating multiple network applications on the same system, GASPP achieves up to $16.2\times $ speedup compared with different monolithic GPU-based implementations of the same applications.

A Resistive CAM Processing-in-Storage Architecture for DNA Sequence Alignment

A novel processing-in-storage (PRinS) architecture based on Resistive CAM (ReCAM) is described and proposed for Smith-Waterman (S-W) sequence alignment. The ReCAM massively-parallel compare operation finds matching base-pairs in a fixed number of cycles, regardless of sequence length. The ReCAM PRinS S-W algorithm is simulated and compared to FPGA, Xeon Phi and GPU-based implementations, showing at least 4.7x higher throughput and at least 15x lower power dissipation.

GPU accelerated Foreign Object Debris Detection on Airfield Pavement with visual saliency algorithm

We present a GPU-based implementation of visual saliency algorithm to detect foreign object debris(FOD) on airfield pavement with effectiveness and efficiency. Visual saliency algorithm is introduced in FOD detection for the first time. We improve the image signature algorithm to target at FOD detection in complex background of pavement. First, we make pooling operations in obtaining saliency map to improve recall rate. Then, connected component analysis is applied to filter candidate regions in saliency map to get the final targets in original image. Besides, we map the algorithm to GPU-based kernels and data structures. The parallel version of the algorithm is able to get the results with 23.5 times speedup. Experimental results elucidate that the proposed method is effective to detect FOD real-time.

Combined Visualization of Vessel Deformation and Hemodynamics in Cerebral Aneurysms.

We present the first visualization tool that combines patient-specific hemodynamics with information about the vessel wall deformation and wall thickness in cerebral aneurysms. Such aneurysms bear the risk of rupture, whereas their treatment also carries considerable risks for the patient. For the patient-specific rupture risk evaluation and treatment analysis, both morphological and hemodynamic data have to be investigated. Medical researchers emphasize the importance of analyzing correlations between wall properties such as the wall deformation and thickness, and hemodynamic attributes like the Wall Shear Stress and near-wall flow. Our method uses a linked 2.5D and 3D depiction of the aneurysm together with blood flow information that enables the simultaneous exploration of wall characteristics and hemodynamic attributes during the cardiac cycle. We thus offer medical researchers an effective visual exploration tool for aneurysm treatment risk assessment. The 2.5D view serves as an overview that comprises a projection of the vessel surface to a 2D map, providing an occlusion-free surface visualization combined with a glyph-based depiction of the local wall thickness. The 3D view represents the focus upon which the data exploration takes place. To support the time-dependent parameter exploration and expert collaboration, a camera path is calculated automatically, where the user can place landmarks for further exploration of the properties. We developed a GPU-based implementation of our visualizations with a flexible interactive data exploration mechanism. We designed our techniques in collaboration with domain experts, and provide details about the evaluation.

A comparison of various schemes for solving the transport equation in many-core platforms

This paper addresses the resolution of the transport (advection---diffusion) equation in 3D by the use of the finite difference method as a discretization technique. In particular, our focus is the comparison of different time schemes (implicit, explicit and predictor-corrector) from the viewpoint of the numerical accuracy and the performance scalability in the use of massively parallel platforms. More in detail, each scheme is implemented in two parallel codes (C+CUDA and C+OpenMP), and in a sequential one (in C). We evaluated the performance of each implementation using several sizes of grid, and also, we measured the precision of these schemes for different discretization levels. The obtained results show that the GPU-based implementations offer significant speed-ups as compared to the CPU counterpart. Moreover, the explicit methods offer a much better performance scalability on this kind of massively parallel devices. This situation reveals that explicit methods can outperform implicit ones even if the accuracy is considered at least in contexts of non stiff problems.

Modeling of nonlinear interactions between guided waves and fatigue cracks using local interaction simulation approach

This article presents a parallel algorithm to model the nonlinear dynamic interactions between ultrasonic guided waves and fatigue cracks. The Local Interaction Simulation Approach (LISA) is further developed to capture the contact-impact clapping phenomena during the wave crack interactions based on the penalty method. Initial opening and closure distributions are considered to approximate the 3-D rough crack microscopic features. A Coulomb friction model is integrated to capture the stick-slip contact motions between the crack surfaces. The LISA procedure is parallelized via the Compute Unified Device Architecture (CUDA), which enables parallel computing on powerful graphic cards. The explicit contact formulation, the parallel algorithm, as well as the GPU-based implementation facilitate LISA’s high computational efficiency over the conventional finite element method (FEM). This article starts with the theoretical formulation and numerical implementation of the proposed algorithm, followed by the solution behavior study and numerical verification against a commercial finite element code. Numerical case studies are conducted on Lamb wave interactions with fatigue cracks. Several nonlinear ultrasonic phenomena are addressed. The classical nonlinear higher harmonic and DC response are successfully captured. The nonlinear mode conversion at a through-thickness and a half-thickness fatigue crack is investigated. Threshold behaviors, induced by initial openings and closures of rough crack surfaces, are depicted by the proposed contact LISA model.

P.2.b.030 - Peripapillary retinal nerve fiber layer thickness in major depressive patients treated with repetitive TMS

Vision and AHRS (attitude and heading reference system) sensors fusion strategy is prevalent in recent years for the legged robot's SLAM (Simultaneous Localization and Mapping), due to its low cost and effectiveness in the global positioning system. In this paper, a new adaptive estimation algorithm is proposed to achieve the robot SLAM by fusing binocular vision and AHRS sensors. A novel acceleration algorithm for SIFT implementation based on Compute Unified Device Architecture (CUDA) is presented to detect the matching feature points in 2D images. All the steps of SIFT were specifically distributed and implemented by CPU or GPU, according to the step's characteristics to make full use of computational resources. The registration of the 3D feature point cloud is performed by using the iterative closest point (ICP) algorithm. Our GPU-based SIFT implementation can run at the speed of 30 frames per second (fps) on most images with 900 × 750 resolution in the test. Compared to other methods, our algorithm is simple to implement and suitable for parallel processing. It can be easily integrated into mobile robot’s tasks like navigation or object tracking, which need the real-time localization information. Experiments results showed that in the unknown indoor environments, the proposed algorithm`s operation is stable and the positioning accuracy is high.

Real-time edge-aware weighted median filtering on the GPU

Edge-preserving weighted median filtering is a fundamental operator in a great variety of image processing and computer graphics applications. This paper presents a novel real-time weighted median filter which effectively smoothes out high-frequency details while preserving major edges. We propose an edge-aware data structure, the 4-dimensional bilateral grid, by extending the 3-dimensional bilateral grid with an additional range dimension. The edge-aware weights and the weighted median values are computed in the 4-dimensional space. The proposed algorithm is highly parallel, enabling a real-time GPU-based implementation. Experimental results show that our filter can be run in real-time on modern GPUs. A variety of edge-aware applications, including detail enhancement, inverse halftoning, JPEG artifact removal, image stylization, photographic look transfer and image colorization, are demonstrated to verify the feasibility of the proposed weighted median filter.

GPU-based parallel optimization of immune convolutional neural network and embedded system

Up to now, the image recognition system has been utilized more and more widely in the security monitoring, the industrial intelligent monitoring, the unmanned vehicle, and even the space exploration. In designing the image recognition system, the traditional convolutional neural network has some defects such as long training time, easy over-fitting and high misclassification rate. In order to overcome these defects, we firstly used the immune mechanism to improve the convolutional neural network and put forward a novel immune convolutional neural network algorithm, after we analyzed the network structure and parameters of the convolutional neural network. Our algorithm not only integrated the location data of the network nodes and the adjustable parameters, but also dynamically adjusted the smoothing factor of the basis function. In addition, we utilized the NVIDIA GPU (Graphics Processing Unit) to accelerate the new immune convolutional neural network (ICNN) in parallel computing and built a real-time embedded image recognition system for this ICNN. The immune convolutional neural network algorithm was improved with CUDA programming and was tested with the sample data in the GPU-based environment. The GPU-based implementation of the novel immune convolutional neural network algorithm was made with the cuDNN, which was designed by NVIDIA for GPU-based accelerating of DNNs in machine learning. Experimental results show that our new immune convolutional neural network has higher recognition rate, more stable performance and faster computing speed than the traditional convolutional neural network.

GPU-Based Parallel Design of the Hyperspectral Signal Subspace Identification by Minimum Error (HySime)

Signal subspace identification provides a performance improvement in hyperspectral applications, such as target detection, spectral unmixing, and classification. The HySime method is a well-known unsupervised approach for hyperspectral signal subspace identification. It computes the estimated noise and signal correlation matrices from which a subset of eigenvectors is selected to best represent the signal subspace in the least square sense. Depending on the complexity and dimensionality of the hyperspectral scene, the HySime algorithm may be computationally expensive. In this paper, we propose a massively parallel design of the HySime method for acceleration on NVIDIA's graphics processing units (GPUs). Our pure GPU-based implementation includes the optimal use of the page-locked host memory, block size, and the number of registers per thread. The proposed implementation was validated in terms of accuracy and performance using the NASA AVIRIS hyperspectral data. The benchmark with the NVIDIA GeForce GTX 580 and Tesla K20 GPUs shows significant speedups with regards to the optimized CPU-based serial counterpart. This new fast implementation of the HySime method demonstrates good potential for real-time hyperspectral applications.

Rapid real-time generation of super-resolution hyperspectral images through compressive sensing and GPU

ABSTRACTRecently, compressive sensing (CS) has offered a new framework whereby a signal can be recovered from a small number of noisy non-adaptive samples. This is now an active area of research in many image-processing applications, especially super-resolution. CS algorithms are widely known to be computationally expensive. This paper studies a real time super-resolution reconstruction method based on the compressive sampling matching pursuit (CoSaMP) algorithm for hyperspectral images. CoSaMP is an iterative compressive sensing method based on the orthogonal matching pursuit (OMP). Multi-spectral images record enormous volumes of data that are required in practical modern remote-sensing applications. A proposed implementation based on the graphical processing unit (GPU) has been developed for CoSaMP using computed unified device architecture (CUDA) and the cuBLAS library. The CoSaMP algorithm is divided into interdependent parts with respect to complexity and potential for parallelization. The proposed implementation is evaluated in terms of reconstruction error for different state-of-the-art super-resolution methods. Various experiments were conducted using real hyperspectral images collected by Earth Observing-1 (EO-1), and experimental results demonstrate the speeding up of the proposed GPU implementation and compare it to the sequential CPU implementation and state-of-the-art techniques. The speeding up of the GPU-based implementation is up to approximately 70 times faster than the corresponding optimized CPU.

Parallel degree computation for binomial systems

Solution sets of systems of binomial equations are of great interest in applied mathematics. For both theoretic and applied purposes, the degree of a solution set (its maximum number of isolated intersections with an affine space of complementary dimension) often plays an important role in understanding its geometric structure. This paper proposes a specialized parallel algorithm for computing the degree on GPUs that takes advantage of the massively parallel nature of GPU devices. The preliminary implementation shows remarkable efficiency and scalability when compared to the closest CPU-based counterpart. As a case study, the algorithm is applied to the master space problem of N=1 gauge theories. The GPU-based implementation achieves nearly 30 fold speedup over its CPU-only counterpart enabling the discovery of previously unknown results.

Small-Signal Stability Analysis of Large Power Systems With Inclusion of Multiple Delays

The paper focuses on the small-signal stability analysis of large power systems with inclusion of multiple delayed signals. The following four techniques are compared: 1) a Chebyshev discretization scheme of an equivalent partial differential equations that resembles the original delay differential-algebraic equations (DDAEs); 2) an approximation of the time integration operator; 3) a linear multistep discretization of the DDAEs based on an high-order implicit time-integration scheme; and 4) the well-known Pade approximants. These techniques are compared using a GPU-based parallel implementation of the Shur method and QR factorization and tested through a real-world transmission system.

Redundant Network Traffic Elimination with GPU Accelerated Rabin Fingerprinting

Recently, redundant network traffic elimination has attracted a lot of attention from both the academia and the industry. A core challenge and enabling technique in implementing redundancy elimination is to perform content-based chunking, which typically involves the computationally heavy Rabin fingerprinting algorithm. In this paper, we propose a GPU-based implementation of Rabin fingerprinting to address this issue. To maximize performance gains, a diverse set of optimization strategies, such as efficient buffer management, GPU memory hierarchy optimization, and balanced load distribution, is proposed by either exploiting the intrinsic hardware features or addressing domain-specific challenges. Extensive evaluations on both the overall and microscopic performance reveal the effectiveness of the GPU-accelerated Rabin fingerprinting algorithm, and we can achieve up to 40 Gpbs throughput on a GTX 780 card. The throughput shows 1.87 $\times$ speedup against the state-of-the-art using comparable hardware. In addition, although some optimization designs are specific for the problem, techniques proposed in this work including the indexed compact buffer scheme and approximate sorting would also be beneficial and applicable to other network applications leveraging GPU acceleration.

GPU-accelerated compressive holography.

In this paper, we show fast signal reconstruction for compressive holography using a graphics processing unit (GPU). We implemented a fast iterative shrinkage-thresholding algorithm on a GPU to solve the ℓ1 and total variation (TV) regularized problems that are typically used in compressive holography. Since the algorithm is highly parallel, GPUs can compute it efficiently by data-parallel computing. For better performance, our implementation exploits the structure of the measurement matrix to compute the matrix multiplications. The results show that GPU-based implementation is about 20 times faster than CPU-based implementation.

Massively parallel GPU computing for fast stereo correspondence algorithms

A GPU-based implementation of the fixed window partially demosaiced CFA stereo matching application is presented.Accelerations up to 20 times are obtained for large size image pairs.A GPU-based implementation of the adaptive window color stereo matching application is presented.It can handle four pairs of standard images from Middlebury database within roughly 100 ms.A comparative study among different GPU-based local dense stereo matching implementations is provided. Current accurate stereo matching algorithms employ some key techniques that are not suitable for parallel GPU architecture. It will be tricky and cumbersome to directly take these techniques into GPU applications. Trying to tackle this difficulty, we design two GPU-based stereo matching algorithms, one using a local fixed aggregation window whose size is configurable, and the other using an adaptive aggregation window which only includes necessary pixels. We use the winner-takes-all (WTA) principle for optimization and a plain voting refinement for post-processing; both do not need complex data structures. We aim to implement on GPU platforms fast stereo matching algorithms that produce results with same-level quality as other WTA local dense methods that use window-based cost aggregation. In our GPU-based implementation of the fixed window partially demosaiced CFA stereo matching application, accelerations up to 20 times are obtained for large size images. In our GPU-based implementation of the adaptive window color stereo matching application, experiment results show that it can handle four pairs of standard images from Middlebury database within roughly 100 ms.