offline-pvanet-environment

因为甲方要求部署环境时不能联网，而且服务器没有计算显卡，所以需要摸索在纯净ubuntu14.04.5中部署纯CPU的pvanet环境

caffe依赖环境安装

安装build-essential

• 进入build-essential目录，依次输入以下命令：

  sudo dpkg -i libstdc++-4.8-dev_4.8.4-2ubuntu1~14.04.3_amd64.deb
  sudo dpkg -i g++-4.8_4.8.4-2ubuntu1~14.04.3_amd64.deb
  sudo dpkg -i dpkg-dev_1.17.5ubuntu5.7_all.deb
  sudo dpkg -i g++_4%3a4.8.2-1ubuntu6_amd64.deb
  sudo dpkg -i build-essential_11.6ubuntu6_amd64.deb

安装cmake

• 进入cmake目录，依次输入以下命令：

  sudo dpkg -i cmake-data_2.8.12.2-0ubuntu3_all.deb
  sudo dpkg -i cmake_2.8.12.2-0ubuntu3_amd64.deb

安装boost

• 进入boost目录，依次进入boost_basic boost_filesystem_regex boost_thread boost_python输入以下命令：

  sudo dpkg -i libboost1.54-dev_1.54.0-4ubuntu3.1_amd64.deb
  sudo dpkg -i libboost-dev_1.54.0.1ubuntu1_amd64.deb
  sudo dpkg -i libboost-system1.54-dev_1.54.0-4ubuntu3.1_amd64.deb
  sudo dpkg -i libboost-system-dev_1.54.0.1ubuntu1_amd64.deb
  
  sudo dpkg -i libboost-filesystem1.54.0_1.54.0-4ubuntu3.1_amd64.deb
  sudo dpkg -i libboost-filesystem1.54-dev_1.54.0-4ubuntu3.1_amd64.deb
  sudo dpkg -i libboost-filesystem-dev_1.54.0.1ubuntu1_amd64.deb
  sudo dpkg -i icu-devtools_52.1-3ubuntu0.7_amd64.deb
  sudo dpkg -i libicu52_52.1-3ubuntu0.7_amd64.deb
  sudo dpkg -i libicu-dev_52.1-3ubuntu0.7_amd64.deb
  sudo dpkg -i libboost-regex1.54.0_1.54.0-4ubuntu3.1_amd64.deb
  sudo dpkg -i libboost-regex1.54-dev_1.54.0-4ubuntu3.1_amd64.deb
  sudo dpkg -i libboost-regex-dev_1.54.0.1ubuntu1_amd64.deb
  
  sudo dpkg -i libboost-atomic1.54.0_1.54.0-4ubuntu3.1_amd64.deb
  sudo dpkg -i libboost-atomic1.54-dev_1.54.0-4ubuntu3.1_amd64.deb
  sudo dpkg -i libboost-chrono1.54.0_1.54.0-4ubuntu3.1_amd64.deb
  sudo dpkg -i libboost-chrono1.54-dev_1.54.0-4ubuntu3.1_amd64.deb
  sudo dpkg -i libboost-serialization1.54.0_1.54.0-4ubuntu3.1_amd64.deb
  sudo dpkg -i libboost-serialization1.54-dev_1.54.0-4ubuntu3.1_amd64.deb
  sudo dpkg -i libboost-date-time1.54-dev_1.54.0-4ubuntu3.1_amd64.deb
  sudo dpkg -i libboost-thread1.54.0_1.54.0-4ubuntu3.1_amd64.deb
  sudo dpkg -i libboost-thread1.54-dev_1.54.0-4ubuntu3.1_amd64.deb
  sudo dpkg -i libboost-thread-dev_1.54.0.1ubuntu1_amd64.deb
  
  sudo dpkg -i libboost-python1.54.0_1.54.0-4ubuntu3.1_amd64.deb
  sudo dpkg -i libexpat1_2.1.0-4ubuntu1.4_amd64.deb
  sudo dpkg -i libexpat1-dev_2.1.0-4ubuntu1.4_amd64.deb
  sudo dpkg -i libpython2.7-minimal_2.7.6-8ubuntu0.4_amd64.deb
  sudo dpkg -i libpython2.7-stdlib_2.7.6-8ubuntu0.4_amd64.deb
  sudo dpkg -i libpython2.7_2.7.6-8ubuntu0.4_amd64.deb
  sudo dpkg -i libpython2.7-dev_2.7.6-8ubuntu0.4_amd64.deb
  sudo dpkg -i libpython-dev_2.7.5-5ubuntu3_amd64.deb
  sudo dpkg -i libboost-python1.54-dev_1.54.0-4ubuntu3.1_amd64.deb
  sudo dpkg -i libboost-python-dev_1.54.0.1ubuntu1_amd64.deb

安装blas

BLAS是一个数学函数接口标准，有很多个实现。按照Caffe官方ubuntu的安装文档默认安装的是ATLAS。这个版本的BLAS不能利用多核CPU，我们将其换为OpenBLAS，可以利用多核CPU并行计算，加快Caffe的分类速度。

安装atlas

同上，找到atlas目录下，按照说明安装即可

安装openBLAS

进入到openblas目录，解压其中任何一个

1. make -j3

2. sudo make install，openblas即被安装在opt/OpenBLAS目录下

3. sudo ln -s /opt/OpenBLAS/lib/libopenblas_nehalemp-r0.2.20.so /usr/lib/x86_..../libopenblas.so.0

安装protobuf、hdf5、glog

• 同上

安装Python环境

安装python

1. 首先进入Python目录，./configure --prefix=/usr/local/python2.7

2. make -j

3. sudo make install

安装setuptools

1. 进入setuptools目录，sudo python setup.py build

2. sudo python setup.py install

安装Anaconda

1. sudo sh Anaconda.....sh

安装opencv

1. 解压opencv2.4.13

2. 编译OpenCV，使用如下命令：

   cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local -D PYTHON_EXECUTABLE=/home/cuizhou/anaconda2/bin/python -D BUILD_OPENCV_PYTHON2=ON -D PYTHON_INCLUDE_DIR=/home/cuizhou/anaconda2/include/python2.7 -D PYTHON_LIBRARY=/home/cuizhou/anaconda2/lib/libpython2.7.so -D PYTHON_NUMPY_PATH=/home/cuizhou/anaconda2/lib/python2.7/site-packages -D PYTHON_PACKAGES_PATH=/home/cuizhou/anaconda2/lib/python2.7/site-packages ..
   make -j3
   sudo make install

安装easydict

1. 进入easydict 目录

2. python setup.py build

3. python setup.py intall

4. 在运行的脚本里加入：import sys sys.path.insert(0, '$easydict 目录')

安装protobuf的python支持

1. 进入protobuf/python

2. python setup.py build

3. 在运行的脚本里加入：import sys sys.path.insert(0, 'protobuf/python')

编译caffe

• caffe-for-pvanet是配置好的用anaconda环境，纯cpu编译的caffe版本，相应更改可自行设置

1. 解压caffe-for-pvanet.tar.gz

2. 注意配置blas的选项和路径

3. make all -j

4. make pycaffe -j

5. 将python目录下编译好的_caffe.so拷贝至PvaNet/recognition/distribute/python/caffe目录下

6. 将build/lib目录下的文件拷贝至PvaNet/recognition/distribute/build/lib目录下

编译faster-rcnn相关lib

• 进入PvaNet/recognition/distribute/lib目录下，运行make -j即可

至此，ubuntu14.04.5下的纯cup pvanet已配置完成