面向异构计算机平台的HPL方案

HPL(high performance Linpack)是一套被广泛用于测评计算机性能的测试程序,几十年来学术界及产业界十分关注对HPL测试程序的定制化优化工作,以充分反应同时代新兴计算机平台的性能.面向当今主流多设备异构计算平台,尝试为HPL的优化工作提供一种解决方案:Hetero-HPL.在Hetero-HPL中...     

Distraction pattern classification and comparisons under different conditions in the full-touch HMI mode

Understanding driver distraction patterns is an important part of human–machine interaction (HMI), which is beneficial for the development of control strategies in human–machine co-driving systems. However, comparatively few studies have focused on driver distraction patterns. To address this issue, this study proposes a framework to characterize distraction patterns using glance behavior and manual behavior, and classifies distraction patterns into: aggressive, moderate, and conservative patterns based on real road experiments. Subsequently, differences in distraction behavior and effects on lateral vehicle control ability across distraction pattern groups, as well as distraction behavior differences exhibited by drivers in the same distraction pattern group under different conditions, are analyzed. Firstly, the results show that the aggressive distraction patterns have a smaller number of eyes-off-road (NoEOR) incidences but longer mean single eyes-off-road time (MSEORT), maximum single eyes-off-road time (MaxEORT) and a higher percentage of long eyes-off-road (PoLEOR) incidences than the other patterns. There are slight differences in the single eyes-off-road times (EORTs) between the conservative and moderate patterns and in the manual behavior for the aggressive and moderate distraction patterns. Secondly, the same distraction pattern exhibited by drivers for different road and secondary task conditions has differences in terms of the behavioral performance. Finally, there is few differences in the lateral motion of a vehicle with different distraction patterns. Surprisingly, the standard deviation of the steering wheel angle (SDSWA) is the smallest in the aggressive distraction pattern.     

Spatiotemporal directional coherence-based scrolling-text detection for frame rate up-conversion

This paper proposes a novel scrolling-text detection method that uses spatiotemporal directional coherence for frame rate up-conversion. Spatiotemporal directional coherence is defined as the coherence level of the spatiotemporal gradient distribution in consecutive frames. Existing scrolling-text detection methods usually detect text using the motion vectors of an image or the distributions of gradient components. The number of motion vectors to determine the scrolling-text decreases at the start and end points of the frame boundary. Therefore, they have difficulty accurately detecting the scrolling text from the start or end points of a frame boundary. The distributions of gradient components can be generated by non-scrolling-text components and they cannot consider the temporal information of consecutive frames, so they may erroneously detect non-scrolling-text regions as the scrolling-text. Unlike the previous methods, which are vulnerable to the presence of texture and noise, the core idea of the proposed method is to use the spatiotemporal directional coherence of the gradient components in the consecutive frames to detect the regions with a dominant edge orientation component for generation of the scrolling-text map and use bit codes from the luminance values to analyze the diversity of luminance patterns for refinement process. With these, the proposed method can further improve the detection accuracy compared to the benchmark methods. The experimental results showed that the proposed method enhanced the average F₁ score by up to 0.5195 (a 137.69 % improvement) compared to the benchmark methods. The average computation time per pixel of the proposed method was also reduced by up to 16.205 μs (a 70.57 % reduction) compared to the benchmark methods.     

基于IR-UWB雷达的多视角融合动态目标追踪

近年来,对运动目标的定位和追踪被广泛地应用于室内导航、智能家居、安防监控和智慧医疗等场景.基于无线射频信号的非接触式定位追踪受到了研究人员的广泛关注,其中基于商用IR-UWB的技术能够以较低的成本和功耗实现目标定位和追踪的功能,具有较强的发展潜力.然而,现有工作大多存在以下问题:1)追踪场景受限,只针对理想情况下室外或者相对空旷的室内场景进行建模和处理; 2)目标的运动状态受限且建模过于理想; 3)虚假动态目标引起的追踪精度不足.为了解决这些问题,在理解多径场景下接收信号谱组成的基础上,提出一个基于IR-UWB的动态目标追踪方法.首先提取原始信号谱中动态成分,并利用基于高斯模糊的多径消除和距离提取算法,消除了多径干扰,仅保留与运动目标直接相关的一次反射信息,从而准确地获取了目标的距离变化曲线.随后,提出多视角融合算法,将不同视角上的设备距离信息进行融合,实现对自由活动目标的准确定位和追踪.此外,还搭建一个基于低成本商用IR-UWB雷达的实时动态目标追踪系统.真实室内家居场景中的实验结果表明,系统估计的人体中心的位置与真实运动轨迹的误差始终小于20 cm.在改变实验环境、实验者、活动速度...     

申威1621处理器上矩阵乘法优化研究

稠密矩阵乘法(GEMM)是很多科学与工程计算应用中大量使用的函数,也是很多代数函数库中的基础函数,其性能高低对整个应用往往有决定性的影响.另外,因其计算密集的特点,矩阵乘法效率往往也是体现硬件平台性能的重要指标.针对国产申威1621处理器,对稠密矩阵乘法进行了系统性地优化.基于对各部分开销的分析,以及对体系结构特点与指令集的充分利用,对DGEMM函数从循环与分块方案,打包方式,核心计算函数实现,数据预取等方面进行了深入优化.此外,开发了代码生成器,为不同的输入参数生成不同版本的汇编代码和C语言代码,配合自动调优脚本,选取最佳参数.经过优化和调优,单线程DGEMM性能达到了单核浮点峰值性能的85%,16线程DGEMM性能达到16核浮点峰值性能的80%.对DGEMM函数的优化不仅提高了申威1621平台BLAS函数库性能,也为国产申威系列多核处理器上稠密数据计算优化提供了重要参考.     

第三方库依赖冲突问题研究综述

软件开发过程中, 开发人员通过大量使用第三方库来实现代码复用. 不同第三方库之间存在依赖关系, 第三方库间的不兼容会导致第三方库的安装、加载、调用时出现错误, 进而导致系统异常, 这类问题称之为第三方库依赖冲突问题. 依赖冲突的根本原因是加载的第三方库无法覆盖软件引用的必需特性(例如: 方法). 依赖冲突问题会在第三方库的下载安装, 项目编译和运行时中出现, 且定位困难. 依赖冲突问题的修复要求开发人员对使用的第三方库版本间差别具有准确的理解, 并且第三方库之间复杂的依赖关系增加了修复难度. 为了能够在软件运行前, 发现软件中存在的依赖冲突, 并且能够响应和处理运行过程中由依赖冲突引发的系统异常, 国内外学者展开了各种针对依赖冲突问题的研究. 从依赖冲突问题的4个方面, 对当前已有研究工作进行了梳理, 包括: 第三方库的使用实证分析、依赖冲突原因分析、依赖冲突检测方法以及依赖冲突常用修复方式. 最后对该领域未来值得关注的研究问题进行了展望.     

Person re-identification based on improved attention mechanism and global pooling method

Deep network has become a new favorite for person re-identification (Re-ID), whose research focus is how to effectively extract the discriminative feature representation for pedestrians. In the paper, we propose a novel Re-ID network named as improved ReIDNet (iReIDNet), which can effectively extract the local and global multi-granular feature representations of pedestrians by a well-designed spatial feature transform and coordinate attention (SFTCA) mechanism together with improved global pooling (IGP) method. SFTCA utilizes channel adaptability and spatial location to infer a 2D attention map and can help iReIDNet to focus on the salient information contained in pedestrian images. IGP makes iReIDNet capture more effectively the global information of the whole human body. Besides, to boost the recognition accuracy, we develop a weighted joint loss to guide the training of iReIDNet. Comprehensive experiments demonstrate the availability and superiority of iReIDNet over other Re-ID methods. The code is available at https://github.com/XuRuyu66/ iReIDNet.     

Screen-shooting resistant image watermarking based on lightweight neural network in frequency domain

Currently, digital mobile devices, especially smartphones, can be used to acquire information conveniently through photograph taking. To protect information security in this case, we propose an efficient screen-shooting resistant watermarking scheme via deep neural network (DNN) in the frequency domain to achieve additional information embedding and source tracing. Specifically, we enhance the imperceptibility of watermarked images and the robustness against various attacks in real scene by computing the residual watermark message and encoding it with the original image using a lightweight neural network in the DCT domain. In addition, a noise layer is designed to simulate the photometric and radiometric effects of screen-shooting transfer. During the training process, the enhancing network is used to highlight the coding features of distorted images and improve the accuracy of extracted watermark message. Experimental results demonstrate that our scheme not only effectively ensures the balance between the imperceptibility of watermark embedding and the robustness of watermark extraction, but also significantly improves computational efficiency compared with some state-of-the-art schemes.