如何配置一个实用又装x的Ubuntu(三) CUDA+caffe+OpenCV+PCL+OpenNI+ROS+Matlab

description: Ubuntu setting tutorial,首发于reasonw.github.io,同步

---

如何配置一个实用又装x的Ubuntu(三) CV相关配置

CUDA

前提是已经装好的nvidia的显卡驱动,我装的是nvidia-367

去官网下载最新的cuda ,这里要注意如果你的系统,opencv都是最新的,cuda也要选最新的,不然会有一些包的版本不匹配

• 安装

    sh cuda_8.0.61_375.26_linux.run --override

then put 's' still to 100% or 'PgDn','q'to quit description

输入accept接受条款

输入n不安装nvidia图像驱动，之前已经安装过了//此处一定要选择n

输入y安装cuda 8.0工具

回车确认cuda默认安装路径：/usr/local/cuda-8.0

输入y用sudo权限运行安装，输入密码

输入y或者n安装或者不安装指向/usr/local/cuda的符号链接

输入y安装CUDA 8.0 Samples，以便后面测试

回车确认CUDA 8.0 Samples默认安装路径：

• 添加路径

    echo "export CUDA_HOME=/usr/local/cuda-8.0 " >> ~/.zshrc
    echo "export LD_LIBRARY_PATH=/usr/local/cuda-8.0/lib64:$LD_LIBRARY_PATH" >> ~/.zshrc
    echo "export PATH=/usr/local/cuda-8.0/bin:$PATH" >> ~/.zshrc
    source ~/.zshrc

• Test CUDA

    cd /usr/local/cuda-8.0/samples/1_Utilities/deviceQuery  
    sudo make -j4 
    sudo ./deviceQuery  

Matlab

• 安装linux版本的Matlab及破解文件link

    sudo mkdir /mnt/tmp
    sudo mount -o loop R2015b_glnxa64.iso /mnt/tmp
    cd /mnt/tmp
    sudo ./install
    sudo mkdir /usr/local/MATLAB/R2015b/licenses
    cd /media/reasonw/661A1FD01A1F9C5D/Ubuntu/
    sudo cp Matlab_Linux/Matlab\ 2015b\ Linux64\ Crack/license_standalone.lic /usr/local/MATLAB/R2015b/licenses
    sudo cp -r Matlab_Linux/Matlab\ 2015b\ Linux64\ Crack/R2015b /usr/local/MATLAB/ 
    sudo umount /mnt/tmp
    sudo rm -r /mnt/tmp/

• 桌面图标

    sudo cp -r Matlab_Linux/ShortCut /usr/local/MATLAB/R2015b 
    sudo cp Matlab_Linux/ShortCut/Matlab_2015b.desktop /usr/share/applications/

• 整理一下

    sudo apt-get update  
    sudo apt-get upgrade  
    sudo apt-get install build-essential  
    sudo apt-get autoremove 

ROS

    sudo sh -c '. /etc/lsb-release && echo "deb http://mirror.sysu.edu.cn/ros/ubuntu/ $DISTRIB_CODENAME main" > /etc/apt/sources.list.d/ros-latest.list'
    sudo apt-key adv --keyserver hkp://ha.pool.sks-keyservers.net:80 --recv-key 421C365BD9FF1F717815A3895523BAEEB01FA116
    sudo apt-get update

有些包的版本不对，要降级

    sudo apt-get install libssl-dev=1.0.2g-1ubuntu4.5 libssl1.0.0=1.0.2g-1ubuntu4.5
    sudo apt-get install libkrb5support0=1.13.2+dfsg-5 libkrb5-3=1.13.2+dfsg-5 libk5crypto3=1.13.2+dfsg-5 libgssapi-krb5-2=1.13.2+dfsg-5

正式安装

    sudo apt install ros-kinetic-desktop-full 
    apt-cache search ros-kinetic
    sudo rosdep init
    rosdep update

如果都没报错,否则重复 rosdep update

    echo "source /opt/ros/kinetic/setup.zsh" >> ~/.zshrc
    source ~/.zshrc
    sudo apt-get install python-rosinstall
    mkdir catkin_ws
    cd catkin_ws 
    mkdir src
    cd src
    catkin_init_workspace
    cd ..
    catkin_make
    sudo apt install ros-kinetic-openni2-launch
    sudo apt install ros-kinetic-openni-launch

cudnn

自己下载 cudnn

    tar xvf cudnn*.tgz
    cd cuda
    sudo cp */*.h /usr/local/cuda/include/
    sudo cp */libcudnn* /usr/local/cuda/lib64/
    sudo chmod a+r /usr/local/cuda/lib64/libcudnn*

Tensorflow

    sudo apt-get install python-pip python-dev 
    sudo apt install libcudart7.5
    sudo pip install six --upgrade --target="/usr/lib/python2.7/dist-packages"
    sudo pip install --upgrade https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow-0.8.0-cp27-none-linux_x86_64.whl

OpenBlas

    mkdir ~/git
    cd ~/git
    git clone https://github.com/xianyi/OpenBLAS.git
    cd OpenBLAS 
    sudo apt-get install gfortran
    make FC=gfortran -j $(($(nproc) + 1))
    sudo make PREFIX=/usr/local install
    echo 'export LD_LIBRARY_PATH=/usr/local/lib:$LD_LIBRARY_PATH' >> ~/.bashrc

caffe

    sudo apt-get install libprotobuf-dev libleveldb-dev libsnappy-dev libopencv-dev libhdf5-serial-dev protobuf-compiler
    sudo apt-get install --no-install-recommends libboost-all-dev
    sudo apt-get install python-skimage ipython python-pil python-h5py ipython python-gflags python-yaml
    sudo apt-get install libgflags-dev libgoogle-glog-dev liblmdb-dev

    cd ~/git
    git clone https://github.com/BVLC/caffe.git
    cd caffe
    cp Makefile.config.example Makefile.config
    sed -i 's/# USE_CUDNN := 1/USE_CUDNN := 1/' Makefile.config
    sed -i 's/BLAS := atlas/BLAS := open/' Makefile.config

    sudo pip install  Cython
    sudo pip install  leveldb 
    sudo pip install  networkx 
    sudo pip install  pandas
    sudo pip install -r python/requirements.txt

    make all -j $(($(nproc) + 1))

if Caffe didn't see hdf5.h when compiling

--- INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include

+++ INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial/

and rename hdf5_hl and hdf5 to hdf5_serial_hl and hdf5_serial in the Makefile:
--- LIBRARIES += glog gflags protobuf boost_system boost_filesystem m hdf5_hl hdf5

+++ LIBRARIES += glog gflags protobuf boost_system boost_filesystem m hdf5_serial_hl hdf5_serial

More about the bug fix here.

    make test -j $(($(nproc) + 1))
    make runtest -j $(($(nproc) + 1))
    make pycaffe -j $(($(nproc) + 1))

OpenNI+OpenNI2

• OpenNI
  
  自己点开下OpenNI SensorKinect NITE1.5

    sudo apt-get install g++ python libusb-1.0-0-dev freeglut3-dev openjdk-8-jdk  doxygen graphviz
    sudo apt-get install mono-complete
    cp -r openni ~/Documents/
    cd ~/Documents/openni/OpenNI
    cd Platform/Linux/CreateRedist
    chmod +x RedistMaker
    ./RedistMaker
    cd ../Redist/OpenNI-Bin-Dev-Linux-x64-v1.5.4.0 
    sudo  ./install.sh
    # sudo chmod +x /usr/include/ni/Linux-x86/
    # sudo chmod +x /usr/include/ni/Linux-Arm/
    # sudo chmod +x /usr/include/ni/MacOSX/
    cd ~/Documents/openni/SensorKinect
    cd Platform/Linux/CreateRedist
    chmod +x RedistMaker
    ./RedistMaker
    cd ../Redist/Sensor-Bin-Linux-x64-v5.1.2.1
    chmod +x install.sh
    sudo ./install.sh
    cd ~/Documents/openni/NITE-Bin-Dev-Linux-x64-v1.5.2.23
    sudo sh ./install.sh

• OpenNI2
  
  自己点开下OpenNI2 lifreenet NITE2.0

    sudo apt-get install -y g++ python libusb-1.0-0-dev freeglut3-dev doxygen graphviz
    sudo apt-get install libudev-dev

这里如果libudev-dev安装出错，换源，用aliyun,update,upgrade,再安装就行

    sudo apt-get install cmake pkg-config build-essential libxmu-dev libxi-dev
    cp -r openni2 ~/Documents/
    cd ~/Documents/openni2/OpenNI2
    make
    cd Bin/x64-Release
    sudo cp libOpenNI2.jni.so libOpenNI2.so /usr/lib
    sudo cp -r OpenNI2 /usr/lib
    cd ../../../libfreenect
    mkdir build
    cd build
    cmake ..
    make
    sudo make install
    cmake .. -DBUILD_OPENNI2_DRIVER=ON
    make
    sudo make install
    sudo ldconfig /usr/local/lib/
    cd ../../NiTE-2.0.0
    sh ./install.sh
    cd Redist
    sudo cp libNiTE2.so NiTE.ini /usr/local/lib
    sudo cp -r NiTE2 /usr/local/lib
    echo "export NITE2_INCLUDE=$HOME/Documents/openni2/NiTE-2.0.0/Include" >> ~/.zshrc
    echo "export NITE2_REDIST64=$HOME/Documents/openni2/NiTE-2.0.0/Redist" >> ~/.zshrc

然后就行了

3_1.png

OpenCV

sudo apt install cmake-gui

    sudo apt install build-essential
    sudo apt install cmake git pkg-config libavcodec-dev libavformat-dev libswscale-dev
    sudo apt install python-dev python-numpy libtbb2 libtbb-dev libjpeg-dev libpng12-dev libtiff5-dev libjasper-dev libdc1394-22-dev
    sudo apt install libgtk2.0-dev 
    cp -r opencv-3.2.0 ~/Documents
    cd ~/Documents/opencv-3.2.0
    mkdir build && cd build
    sudo apt install libprotobuf-c-dev 
    sudo apt install libprotobuf-c1

    git clone https://github.com/opencv/opencv.git
    cd opencv
    git checkout 3.1.0
    git clone https://github.com/opencv/opencv_contrib.git
    cd opencv_contrib
    git checkout 3.1.0

or

    git clone https://github.com/opencv/opencv.git
    cd opencv
    git checkout 3.1.0
    git clone https://github.com/opencv/opencv_contrib.git
    cd opencv_contrib
    git checkout 3.2.0

    cmake -DOPENCV_EXTRA_MODULES_PATH=../opencv_contrib/modules  -DBUILD_opencv_dnn=ON -DBUILD_EXAMPLES=ON -DBUILD_PNG=ON -DINSTALL_C_EXAMPLES=ON  -DWITH_OPENNI=ON -DWITH_OPENNI2=ON ..

如果openni2没有编译上,就过一会儿再编译一次
这里cuda的版本一定要支持系统gcc版本，只能高不能低，版本只能新不能老,如果编译报错

sudo subl /usr/local/cuda/include/host_config.h

line: 115 comment out error

//#error -- unsupported GNU version! gcc versions later than 4.9 are not supported!

    sudo make -j $(($(nproc) + 1))
    sudo make install 

PCL

    sudo apt install g++ cmake cmake-qt-gui  doxygen mpi-default-dev openmpi-bin openmpi-common  libflann1.8 libflann-dev libeigen3-dev libboost-all-dev libvtk6.2-qt libvtk6.2  libvtk5-dev  libqhull*  libusb-dev libgtest-dev build-essential libxmu-dev libxi-dev libusb-1.0-0-dev graphviz mono-complete phonon-backend-gstreamer  phonon-backend-vlc  libvtk-java python-vtk6
    sudo apt install  git-core freeglut3-dev pkg-config
    cp -r pcl ~/Documents
    cd ~/Documents/pcl 
    mkdir build
    cd build
    cmake -DBUILD_CUDA=ON -DBUILD_GPU=ON -DBUILD_apps=ON -DBUILD_examples=ON -DWITH_OPENNI=ON -DWITH_OPENNI2=ON  -DBUILD_cuda_io=ON -DBUILD_cuda_apps=ON -DBUILD_gpu_tracking=ON -DBUILD_gpu_surface=ON ..
    sudo make -j $(($(nproc) + 1))
    sudo make install

如果openni2没有编译上,就先安装,过一会儿重启再重启再编译安装一次,一般就有了

NOTES

我从14版本升到16版本以后,编译以前的视觉包一直报一个libmpi的错,后来才知道是系统的动态链接库管理ld因升级有了变化，主要表现为依赖的包再次依赖不会自动展开，报一个很烦人的错误

    undefined reference to symbol '_ZN3MPI8Datatype4FreeEv'
    usr/lib/libmpi_cxx.so.1: error adding symbols: DSO missing from command line

知道了原因就好办了,编译的时候加上mpi链接就行,CMakeLists.txt里面加上

SET(CMAKE_C_COMPILER mpicc)

SET(CMAKE_CXX_COMPILER mpicxx)

include_directories(MPI_INCLUDE_PATH)

target_link_libraries(mytest ${MPI_LIBRARIES})