使用源码编译 搭建AI 运行环境在SUSE12 Intereasen cuda pylibfreenect2 libfreenect2

AI Program Enviroment  Installtion doc

1 :compile tools

cmake           version>=3.8

g++             version>=5.5

/usr/bin/g++ -> /usr/bin/g++-5(make sure it's exits)

cXX             version>=5.5

/usr/bin/cc -> /usr/bin/gcc-5 (make sure it's exits)

 

ATTENTION

---------------------------------

gcc5-c++-5.5.0+r253576-1.1.x86_64 contains gcc>5.5 version and other dependience you could zyyper from suse remote repo

http://download.opensuse.org/repositories/devel:/gcc/SLE-12/

2：NEEDED DEPENDIENCE FOR ALL AI SERVER

-------------------------------------NEEDED DEPENDIENCE BELOW---------------------------------------------------

You can find 90% dependience in  SLE-12-SP2-SDK-DVD-x86_64-GM-DVD1.iso SLE-12-SP2-SDK-DVD-x86_64-GM-DVD2.iso

You'd better set these iso as suse local repo;

--------- FOR EXAMPLE

mkdir or

chmod -R 777 or

mount -o loop SLE-12-SP2-SDK-DVD-x86_64-GM-DVD2.iso /or

zypper ar file:///or/ local-sles

zypper lr

--------END

-------------------------------------RPM DEPENDIENCE---------------------------------------------------------------------

'zypper' 'in' 'fontconfig-devel'

'zypper' 'in' 'Mesa-libEGL-devel'

'zypper' 'in' 'libpng16-compat-devel'

'zypper' 'in' 'cairo-devel'

'zypper' 'in' 'harfbuzz-devel'

'zypper' 'in' 'gdk-pixbuf-devel'

'zypper' 'in' 'libXinerama-devel'

'zypper' 'in' 'libXi-devel'

'zypper' 'in' 'libXcursor-devel'

'zypper' 'in' 'libjpeg62-devel'

'zypper' 'in' 'libva-devel'

'zypper' 'in' 'libXrandr-devel-1.5.0-6.2.x86_64'

'zypper' 'in' 'libudev-devel-228-117.12.x86_64'

'zypper' 'in' 'glu-devel-9.0.0-18.1.x86_64'

'zypper' 'in' 'libusb-1_0-devel-1.0.20-5.3.x86_64'

'zypper' 'in' 'libjpeg8-devel-8.0.2-30.3.x86_64'

'zypper' 'in' 'atk-devel-2.20.0-4.5.x86_64'

'zypper' 'in' 'python-devel'

 

3: AI SERVER

 

CUDA 9.0            https://developer.nvidia.com/cuda-90-download-archive

 

CUDNN 7.05          https://developer.nvidia.com

 

miniconda           https://conda.io/miniconda.html (64-bit python2.7 installer/ use IE browser )

 

librealsense   https://github.com/IntelRealSense/librealsense

 

libfreenect2   https://github.com/OpenKinect/libfreenect2

 

pylibfreenect2 https://github.com/r9y9/pylibfreenect2

 

glfw                 https://github.com/glfw/glfw

 

OpenNi2         https://github.com/OpenNI/OpenNI2

 

ffmpeg         https://github.com/FFmpeg/FFmpeg

step 1 : set-up cuda cudnn and miniconda

1.        Install CUDA 9.0

i.        Download CUDA rpm file from nvidia official website. I’ve downloaded it and placed it in /usr/g/ctuser/wangwei. The website address https://developer.nvidia.com/cuda-90-download-archive

ii.        sudo rpm -i cuda-repo-sles122-9-0-local-9.0.176-1.x86_64.rpm

iii.        sudo zypper refresh

iv.        sudo zypper install cuda

v.        after installation, verfity that /usr/local/cuda directory exist

 

2.        Install CUDNN 7.05

i.        Download cudnn tgz file from nvidia website. It requires nvidia membership. I’ll also download and save it in /usr/g/ctuser/wangwei

ii.        Unzip tgz file

iii.        Copy cuda folder in unzipped folder to /usr/loca/cuda, you may need switch to root

iv.        cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2, verify that it’s cudnn 7.05

3.        Install miniconda

i.        Download miniconda from https://conda.io/miniconda.html. Please download 64-bit python2.7 installer

ii.        Install it. It’s not difficult

4.        Install virtual environment

i.        Before following steps, verify that computer can get connected to Internet.

ii.        conda create -n py2.7 python=2.7, wait for installation finish

iii.        conda env list. Verify that py2.7 exist

 

iv.        activate py2.7. All the following steps are executed inside the virtual environment

v.        Install tensorflow, keras, numpy, opency……….etc from requirement.txt. pip install -r requirement.txt. requirement.txt is also in /usr/g/ctuser/wangwei.

vi.        conda install -c menpo dlib=18.18

 

 

step2: set up OpenNi2  glfw  libfreenect2 pylibfreenect2

set up OpenNi2 first,next libfreenect2

-------------------------------------------------------------------------------------------------

if you set up glfw earlier than libfreenect2 something will goes wrong when you build libfreenect2

 

--------------------------ATTENTION-----------------------------------

/user/lib/libglfw3.a will get conflict with libfreenect2

----------------------------------------------------------------------

 

step2.2:

set up libfreenect2

    cd ..

    mkdir build && cd build

    cmake .. -DCMAKE_INSTALL_PREFIX=$HOME/freenect2

    make

    make install

set up pylibfreenect2

    requirements:

    libfreenect2 (0.2.0 or later)

python (2.7, 3.4 or 3.5)

numpy

cython

C++ compiler (clang, gcc or MSVC)

--------------

cd pylibfreenect2

python setup.py install

 

set up         OpenNi2  glfw

  it's really easy just follow the README.md

 

 step3:

  librealsense

     cd librealsense&mkdir build

     cd build

     cmake ../ -DBUILD_PYTHON_BINDINGS=bool:true -DPYTHON_EXECUTABLE=/usr/bin/python2.7 -DBUILD_WITH_OPENMP=false

     make

     make install