Real-time Graphics Processing Units Research Articles

A miniature temperature sensor based on upconversion luminescence for real-time GPU temperature monitoring

This paper introduces a miniaturized optical fiber temperature sensor based on Fluorescence Intensity Ratio (FIR) technology for real-time monitoring of graphics processing unit (GPU) temperature. Utilizing a femtosecond micromachining system, miniature rectangular holes are etched on a standard multimode fiber. These holes are then filled with a mixture of Er3+/Yb3+ co-doped TeO2-Na2CO3-ZnO powder and polydimethylsiloxane (PDMS) through capillary action, forming a sandwich structure. When illuminated by a 980 nm light source, upconversion (UC) fluorescence is generated, and a mathematical model correlating the fluorescence intensity ratio of two adjacent energy levels with temperature is established. The fundamental temperature sensing characteristics of the sensor are tested in a temperature-controlled chamber, with a maximum error of ± 1 K. The sensor is also applied to a GPU with real-time temperature variations, demonstrating a maximum error of 0.9 K and a response time of 1.96 s. The sensor not only possesses the advantages of a simple structure, micro size, and convenient fabrication but also exhibits immunity to electromagnetic interference, rapid response, good stability, and excellent repeatability in real-time monitoring of GPU temperature, showing potential for large-scale application.

Volume reconstruction based on the six-direction cubic box-spline

We propose a new volume reconstruction technique based on the six-direction cubic box-spline M6. M6 is C1 continuous and possesses an approximation order of three, the same as that of the tri-quadratic B-spline but with much lower degree. In fact, M6 has the lowest degree among the symmetric box-splines on Z3 with at least C1 continuity. We analyze the polynomial structure induced by the shifts of M6 and propose an efficient analytic evaluation algorithm for splines and their derivatives (gradient and Hessian) based on the high symmetry of M6. To verify the evaluation algorithm, we implement a real-time GPU (graphics processing unit) isosurface raycaster which exhibits interactive performance (54.5 fps (frames per second) with 2413 dataset on 5122 framebuffer) on a modern graphics hardware. Moreover, we analyze M6 as a reconstruction filter and state that it is comparable to the tri-cubic B-spline, which possesses a higher approximation order.

Wide-field sensorless adaptive optics swept-source optical coherence tomographic angiography in rodents.

In this study, we present a sensorless adaptive optics swept-source optical coherence tomographic angiography (sAO-SS-OCTA) imaging system for mice. Real-time graphics processing unit (GPU)-based OCTA image acquisition and processing software were applied to guide wavefront correction using a deformable mirror based on signal strength index (SSI) from both OCT and OCTA images. High-resolution OCTA images with aberrations corrected and contrast enhanced were successfully acquired. Fifty-degree field of view high-resolution montaged OCTA images were also acquired.

Image deconvolution for optical small satellite with deep learning and real-time GPU acceleration

In-orbit optical-imaging instruments may suffer from degradations due to space environment impacts or long-time operation. The degradation causes blurring on the image received from the ground. Degradations come from defocus and spherical aberrations cause blurring on the received image. Image deblurring should be done in pre-processing step to compensate the sensor bad impacts. The aberrations are modeled by Zernike polynomials and treated by deep learning in deblurring method. This paper presents a method to deconvolve the acquired data to improve the image quality. A convolution neural network is trained to estimate the point spread function (PSF) parameters using acquired images over satellite calibration site with specific pattern. Image deconvolution is performed to obtain image signal-to-noise (SNR) and modulation transfer function (MTF) improvement. Technical and image data used for modeling and experiment are used from VNREDSat-1 satellite (the first operational Vietnam Earth observation optical small satellite). The experiment is performed on computers accelerated by graphics processing units (GPU) to ensure fast computation.

Fast and stable evaluation of splines and their derivatives generated by the seven-direction quartic box-spline

In this paper, we propose a fast and stable evaluation method for the analytic derivatives of splines generated by the 7-direction quartic box-spline. We can maintain the spline structure by determining the derivative functions that can be represented as finite differences of box-splines defined by the sub-directions. Thus, the evaluation overhead can be reduced. We demonstrate that the first and second derivative functions are composed of only three cubic and six quadratic polynomial formulas, respectively, owing to their symmetries. Moreover, for each derivative order, all of the required functions possess change-of-variables relation with each other. Therefore, additional formulas are not required. As a result, for a given point, we only need to evaluate one quartic, three cubic, and six quadratic polynomial formulas to evaluate its spline value, gradient, and Hessian, respectively. This reduction in cases is especially advantageous for graphics processing unit (GPU) kernels, where conditional statements significantly degrade performance. To verify our technique, we implemented a real-time curvature-based GPU isosurface raycaster. Compared with other implementations, our method (i) achieves superior accuracy, (ii) is more than four times faster, and (iii) requires less than 15% of memory.

GPU Sparse Ray-Traced Segmentation

This paper introduces a real-time region growing segmentation algorithm, designed for graphics processing units (GPUs), which labels only a fraction of the input elements. Instead of searching locally around each element for strong similarity, like state-of-the-art segmentation and pre-segmentation methods do, the proposed algorithm searches both locally and remotely, using a unique ray tracing-based search strategy, which quickly covers the segmentation search space. The presented algorithm fully exploits the parallelism of the GPUs, sparsely segmenting high-resolution images (4K) in real-time on low range laptops and other mobile devices, approximately 5× times faster than the state-of-the-art simple linear iterative clustering (SLIC). While this paper demonstrates the results with images, the algorithm is trivially modifiable to work with input sets of any dimension. In contrast to the state-of-the-art real-time GPU methods, this algorithm doesn't require additional merging steps, as pre-segmentation methods do, and it produces complete segmentation. Additionally, post-segmentation optional stages for complete labeling and region merging on the GPU are also provided, although they are not always necessary.

Real-Time GPU Resource Management with Loadable Kernel Modules

Graphics processing unit (GPU) programming environments have matured for general-purpose computing on GPUs. Significant challenges for GPUs include system software support for bounded response times and guaranteed throughput. In recent years, GPU technologies have been applied to real-time systems by extending the operating system modules to support real-time GPU resource management. Unfortunately, such a system extension makes it difficult to maintain the system with version updates because the OS kernel and device drivers must be modified at the source-code level, thereby preventing continuous research and development of GPU technologies for real-time systems. A loadable kernel module (LKM) framework, called Linux Real-Time eXtention with GPUs (Linux-RTXG), for managing real-time GPU resources with Linux without modifying the OS kernel and device drivers is proposed and evaluated experimentally. Linux-RTXG provides mechanisms for interrupt interception and independent synchronization to achieve real-time scheduling and resource reservation capabilities for GPU applications on top of existing device drivers and runtime libraries. Experimental results demonstrate that the overhead incurred by introducing the proposed Linux-RTXG is comparable to that of introducing existing kernel-dependent approaches. In addition, the results demonstrate that multiple GPU applications can be scheduled successfully by Linux-RTXG to meet their priority and quality-of-service requirements in real time.

Enhancement and Edge-Preserving Denoising: An OpenCL-Based Approach for Remote Sensing Imagery

Image enhancement and edge-preserving denoising are relevant steps before classification or other postprocessing techniques for remote sensing images. However, multisensor array systems are able to simultaneously capture several low-resolution images from the same area on different wavelengths, forming a high spatial/spectral resolution image and raising a series of new challenges. In this paper, an open computing language based parallel implementation approach is presented for near real-time enhancement based on Bayesian maximum entropy (BME), as well as an edge-preserving denoising algorithm for remote sensing imagery, which uses the local linear Stein's unbiased risk estimate (LLSURE). BME was selected for its results on synthetic aperture radar image enhancement, whereas LLSURE has shown better noise removal properties than other commonly used methods. Within this context, image processing methods are algorithmically adapted via parallel computing techniques and efficiently implemented using CPUs and commodity graphics processing units (GPUs). Experimental results demonstrate the reduction of computational load of real-world image processing for near real-time GPU adapted implementation.