ABBNKHS03-15◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆【厦门莫格电气自动化有限公司】【XiamenMoggetElectricAutomationCo.,Ltd】【当天顺丰发货，欢迎***验货，不要犹豫，不要徘徊，错失良机，后悔晚矣】【来电咨询：雷(女士）】【销售***请点上面↑↑↑↑↑↑↑↑↑↑】【传真：0592-6514751（请备注：雷琳收）】【邮箱：1982497648@***.com】公司主营AB、本特利、黑马、施耐德、GE、ABB【DSQC系列】ICS英维思西门子yokogawa横河霍尼韦尔福克斯波Rosemount（罗斯蒙特）德国EPRO（飞利浦）ENTEK（恩泰克）VIBRO-METER（韦伯）Yaskawa（安川）Motorola（摩托罗拉）BoschRexroth（博世力士乐）Woodward（伍德沃德）等品牌◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆全新原装现货：NKHS03-15NKHS03-15NKHS03-15TensorFlow在Google的推动下，加之其设计***网络结构的代码的简洁度、分布式深度学习算法的执行效率，还有部署的便利性，在诸多的深度框架中脱颖而出。Caffe则是主要用于计算机视觉领域的深度学习框架，其全称为ConvolutionalArchitectureforFastFeatureEmbedding，目前由伯克利视觉学中心affe的代码成熟度较高，可以算是一个主流工业级单机运行的深度学习框架。本文采用Toradex基于nVdiaTegraK1芯片的ARM核心板ApalisTK1模块进行测试，TegraK1片上集成了192个支持CUDA运算的GPU核心。该GPU除了能够完成视频编***外，还可以借助CUDA、OpenCL用于并行计算。由于TegraK1采用了和桌面显卡一样的GPU架构，ApalisTK1也能够实现利用cuDNN对Caffe进行加速运算。2).具体操作下面我们将介绍如何在ApalisTK1上安装Caffe、OpenCV3，并演示物体识别算法。首先，使用ToradexEasyInstaller在ApalisTK1模块上安装L4TUbuntu系统。具体操作请参考ToradexEasyInstaller使用说明。由于***文件较大，整个安装过程需要10分钟左右的时间。Caffe所需的cuDNN需要单独从JetPack中安装，由于仅需要CUDA工具以及cuDNN，因此在安装的时间可以不选择其他组件，减少安装时间。安装的组件如下：通过下面脚本***并编译OpenCV3，以sudo权限执行脚本。---------------------------------------#!/bin/shsudoapt-add-repositoryuniversesudoapt-getupdatesudoapt-getinstallbuild-essentialmakecmakecmake-curses-guig++pkg-config-ysudoapt-getinstalllib***format-devlib***util-devlibswscale-dev-ysudoapt-getinstalllibv4l-dev-ysudoapt-getinstalllibeigen3-dev-ysudoapt-getinstalllibglew1.6-dev-ysudoapt-getinstalllibgtk2.0-dev-ysudoapt-get-yinstallcheckinstallya***sudoapt-get-yinstalllibgstreamer1.0-devlibgstreamer-plugins-base1.0-devlibxine-devlibgstreamer0.10-devlibgstreamer-plugins-base0.10-devsudoapt-get-yinstallpython-devpython-numpy-ysudoapt-get-yinstalllibfaac-devlibjack-jackd2-devlibmp3lame-devlibopencore-amrnb-devlibopencore-amrwb-devlibsdl1.2-devlibva-devlibvdpau-devlibxvidcore-devtexi2htmlgitNUM_THREADS=4ver=3.4.0gitclonegit:///opencv/opencv.gitopencv-$vercdopencv-$vergitcheckout$vermkdirbuildcdbuildcmake-DWITH_CUDA=ON-DCUDA_ARCH_BIN="3.2"-DCUDA_ARCH_PTX=""-DBUILD_TESTS=OFF-DBUILD_PERF_TESTS=OFF-DENABLE_NEON=ON-DBUILD_EXAMPLES=ON-DBUILD_opencv_python2=ON-DWITH_OPENMP=ON-DENABLE_NEON=ON-DWITH_GSTREAMER_0_10=ON..make-j$NUM_THREADSsudomake-j$NUM_THREADSinstall/bin/echo-e"\e[1;32mOpenCVsimplebuildinstallationcomplete.\e[0m"---------------------------------------https:///tgHSYzw3完成安装后运行sudoldconfig命令，更新OpenCV库文件。***Caffe，***新版本的Caffe需要更高版本的cuDNN支持，TegraK1目前只支持cuDNNv2。---------------------------------------#!/bin/shsudoapt-getinstalllibprotobuf-devprotobuf-compilergfortran\libboost-devcmakelibleveldb-devlibsnappy-dev\libboost-thread-devlibboost-system-devlibboost-python-dev\libatlas-base-devlibhdf5-serial-devlibgflags-dev\libgoogle-glog-devliblmdb-dev-ygitclonegit:///platotek/caffetk1.gitcdcaffetk1nfig.examplenfig---------------------------------------https:///SvkfBDy0修改nfig，添加cuDNN、OpenCV3支持，以及CUDA库文件---------------------------------------#cuDNNaccelerati***witch(uncommenttobuildwithcuDNN).USE_CUDNN:=1#Whateverelseyoufindyouneedgoeshere.INCLUDE_DIRS:=$(PYTHON_INCLUDE)/usr/local/include\/usr/local/cuda-6.5/includeLIBRARY_DIRS:=$(PYTHON_LIB)/usr/local/lib/usr/lib\/usr/local/cuda-6.5/lib#Uncommentifyou'reusingOpenCV3OPENCV_VERSION:=3#Uncommenttouse`pkg-config`tospecifyOpenCVlibrarypaths.#(Usuallynotnecessary--OpenCVlibrariesarenormallyinstalledinoneoftheabove$LIBRARY_DIRS.)USE_PKG_CONFIG:=1---------------------------------------修改Makefile文件，添加imgcodecs---------------------------------------LIBRARIES+=gloggflagsprotobufleveldbsnappy\lmdbboost_systemhdf5_hlhdf5m\opencv_coreopencv_highguiopencv_imgprocopencv_imgcodecs---------------------------------------***后运行---------------------------------------make-j4allmakepycaffe---------------------------------------***物体识别python应用。解压后进入应用目录，配置caffepython目录---------------------------------------exportPYTHONPATH=/home/ubuntu/caffetk1/python:$PYTHONPATHpythondeep_learning_object_detection.py--totxt.txt--ffemodel--imageimages/g)