①caffe从github下载：执行如下命令：git clone https://github/.../caffe.git

②进入caffe/python目录，安装caffe的python依赖库，执行如下命令：$for req in $(cat requirements.txt); do pip install $req;done

Successfully installed Cython-0.27.3
Requirement already satisfied: numpy>=1.7.1 in /users4/zs/.local/lib/python2.7/site-packages
Requirement already satisfied: numpy>=1.8.2 in /users4/zs/.local/lib/python2.7/site-packages (from scipy>=0.13.2)
Collecting scikit-image>=0.9.3
Collecting matplotlib>=1.3.1 (from scikit-image>=0.9.3)
Collecting networkx>=1.8 (from scikit-image>=0.9.3)
Collecting pillow>=2.1.0 (from scikit-image>=0.9.3)
Requirement already satisfied: six>=1.7.3 in /users4/zs/.local/lib/python2.7/site-packages (from scikit-image>=0.9.3)
Requirement already satisfied: scipy>=0.17.0 in /users4/zs/.local/lib/python2.7/site-packages (from scikit-image>=0.9.3)
Collecting PyWavelets>=0.4.0 (from scikit-image>=0.9.3)
Collecting subprocess32 (from matplotlib>=1.3.1->scikit-image>=0.9.3)
Collecting cycler>=0.10 (from matplotlib>=1.3.1->scikit-image>=0.9.3)
Collecting backports.functools-lru-cache (from matplotlib>=1.3.1->scikit-image>=0.9.3)
Collecting pytz (from matplotlib>=1.3.1->scikit-image>=0.9.3)
Collecting pyparsing!=2.0.4,!=2.1.2,!=2.1.6,>=2.0.1 (from matplotlib>=1.3.1->scikit-image>=0.9.3)
Requirement already satisfied: numpy>=1.7.1 in /users4/zsun/.local/lib/python2.7/site-packages (from matplotlib>=1.3.1->scikit-image>=0.9.3)
Collecting python-dateutil>=2.0 (from matplotlib>=1.3.1->scikit-image>=0.9.3)
Collecting decorator>=4.1.0 (from networkx>=1.8->scikit-image>=0.9.3)
Building wheels for collected packages: networkx, subprocess32
  Running setup.py bdist_wheel for networkx ... done
  Stored in directory: /users4/zsun/.cache/pip/wheels/c8/58/65/43f97d17a5ba1d88697063234c3ad92ad68885f5f702a20680
  Running setup.py bdist_wheel for subprocess32 ... done
  Stored in directory: /users4/zsun/.cache/pip/wheels/b3/98/13/6760c4255f07fec23c6b22640b1129e57d97b9e736dca5c08b
Successfully built networkx subprocess32
Installing collected packages: subprocess32, cycler, backports.functools-lru-cache, pytz, pyparsing, python-dateutil, matplotlib, decorator, networkx, pillow, PyWavelets, scikit-image
Successfully installed PyWavelets-0.5.2 backports.functools-lru-cache-1.4 cycler-0.10.0 decorator-4.2.1 matplotlib-2.1.1 networkx-2.0 pillow-5.0.0 pyparsing-2.2.0 python-dateutil-2.6.1 pytz-2017.3 scikit-image-0.13.1 subprocess32-3.2.7
Requirement already satisfied: matplotlib>=1.3.1 in /users4/zs/anaconda3/envs/py27/lib/python2.7/site-packages
Requirement already satisfied: subprocess32 in /users4/zs/anaconda3/envs/py27/lib/python2.7/site-packages (from matplotlib>=1.3.1)
Requirement already satisfied: cycler>=0.10 in /users4/zs/anaconda3/envs/py27/lib/python2.7/site-packages (from matplotlib>=1.3.1)
Requirement already satisfied: six>=1.10 in /users4/zs/.local/lib/python2.7/site-packages (from matplotlib>=1.3.1)
Requirement already satisfied: backports.functools-lru-cache in /users4/zs/anaconda3/envs/py27/lib/python2.7/site-packages (from matplotlib>=1.3.1)
Requirement already satisfied: pytz in /users4/zs/anaconda3/envs/py27/lib/python2.7/site-packages (from matplotlib>=1.3.1)
Requirement already satisfied: pyparsing!=2.0.4,!=2.1.2,!=2.1.6,>=2.0.1 in /zs/anaconda3/envs/py27/lib/python2.7/site-packages (from matplotlib>=1.3.1)
Requirement already satisfied: numpy>=1.7.1 in /zs/.local/lib/python2.7/site-packages (from matplotlib>=1.3.1)
Requirement already satisfied: python-dateutil>=2.0 in /zs/anaconda3/envs/py27/lib/python2.7/site-packages (from matplotlib>=1.3.1)
Collecting ipython>=3.0.0 
    Complete output from command python setup.py egg_info:
    IPython 6.0+ does not support Python 2.6, 2.7, 3.0, 3.1, or 3.2.
    When using Python 2.7, please install IPython 5.x LTS Long Term Support version.
    Beginning with IPython 6.0, Python 3.3 and above is required.
    See IPython `README.rst` file for more information:
        https://github.com/ipython/ipython/blob/master/README.rst
    Python sys.version_info(major=2, minor=7, micro=14, releaselevel='final', serial=0) detected.
    ----------------------------------------
Command "python setup.py egg_info" failed with error code 1 in /tmp/pip-build-HG651n/ipython/
Collecting h5py>=2.2.0
Requirement already satisfied: numpy>=1.7 in /users4/zs/.local/lib/python2.7/site-packages (from h5py>=2.2.0)
Requirement already satisfied: six in /users4/zs/.local/lib/python2.7/site-packages (from h5py>=2.2.0)
Successfully installed h5py-2.7.1
Successfully installed leveldb-0.194
Requirement already satisfied: networkx>=1.8.1 in /users4/zs/anaconda3/envs/py27/lib/python2.7/site-packages
Requirement already satisfied: decorator>=4.1.0 in /users4/zs/anaconda3/envs/py27/lib/python2.7/site-packages (from networkx>=1.8.1)
Successfully installed nose-1.3.7
Collecting pandas>=0.12.0
Requirement already satisfied: pytz>=2011k in /users4/zs/anaconda3/envs/py27/lib/python2.7/site-packages (from pandas>=0.12.0)
Requirement already satisfied: numpy>=1.9.0 in /users4/zs/.local/lib/python2.7/site-packages (from pandas>=0.12.0)
Requirement already satisfied: python-dateutil in /users4/zs/anaconda3/envs/py27/lib/python2.7/site-packages (from pandas>=0.12.0)
Requirement already satisfied: six>=1.5 in /users4/zs/.local/lib/python2.7/site-packages (from python-dateutil->pandas>=0.12.0)
Successfully installed pandas-0.22.0
Collecting python-dateutil<2,>=1.4
Building wheels for collected packages: python-dateutil
  Running setup.py bdist_wheel for python-dateutil ... done
  Stored in directory: /users4/zsun/.cache/pip/wheels/83/66/7c/0d6150f3c41e070516cb482e30597ee43e796529909ce23f45
Successfully built python-dateutil
Installing collected packages: python-dateutil
  Found existing installation: python-dateutil 2.6.1
    Uninstalling python-dateutil-2.6.1:
      Successfully uninstalled python-dateutil-2.6.1
Successfully installed python-dateutil-1.5
Requirement already satisfied: protobuf>=2.5.0 in /users4/zs/.local/lib/python2.7/site-packages
Requirement already satisfied: six>=1.9 in /users4/zs/.local/lib/python2.7/site-packages (from protobuf>=2.5.0)
Requirement already satisfied: setuptools in /users4/zs/.local/lib/python2.7/site-packages (from protobuf>=2.5.0)
Successfully installed python-gflags-3.1.2
Requirement already satisfied: pyyaml>=3.10 in /users4/zs/.local/lib/python2.7/site-packages/PyYAML-3.12-py2.7-linux-x86_64.egg
Requirement already satisfied: Pillow>=2.3.0 in /users4/zs/anaconda3/envs/py27/lib/python2.7/site-packages
Requirement already satisfied: six>=1.1.0 in /users4/zs/.local/lib/python2.7/site-packages

③安装依赖

$ yumdownloader protobuf-devel leveldb-devel snappy-devel opencv-devel boost-devel hdf5-devel gflags-devel glog-devel lmdb-devel

$ rpm2cpio  以上.rpm |cpio -idvm

$ cp  -rf   ./usr/* ~/usr/

④手动安装其他

1.glog

$ pip install glog
Requirement already satisfied: glog in /users4/zsun/anaconda3/lib/python3.6/site-packages
Requirement already satisfied: six in /users4/zsun/anaconda3/lib/python3.6/site-packages (from glog)
Requirement already satisfied: python-gflags>=3.1 in /users4/zsun/anaconda3/lib/python3.6/site-packages (from glog)

2.gflags

$ wget https://github.com/schuhschuh/gflags/archive/master.zip 
unzip master.zip 
cd gflags-master 
mkdir build && cd build 
export CXXFLAGS=”-fPIC” && cmake .. && make VERBOSE=1 
make && make install

3.lmdb

git clone https://github.com/LMDB/lmdb 
cd lmdb/libraries/liblmdb 
make && make install

4.protobuf重要

root用户安装步骤请参考 http://blog.csdn.net/tdmyl/article/details/31811317

非root用户参考 http://blog.csdn.net/ccbrid/article/details/79169440

5.hdf5安装

hdf5安装请查看官方文档 
http://www.hdfgroup.org/HDF5/release/obtainsrc.html#src

6.leveldb安装 

leveldb下载请见https://github.com/google/leveldb 
安装请见http://blog.sina.com.cn/s/blog_560e310001015jfx.html 
缺失的步骤： 
leveldb文件夹下依次执行 
make 
cd out-shared 
cp lib* /usr/local/lib

7.OpenBLAS安装OpenBLAS下载链接

下载后，解压，然后安装。  
然后执行下面的指令 
tar -xzvf OpenBLAS.tar.gz 
make 
make PREFIX=~/usr/local/ install 
上述步骤只做参考，实际操作过程可能会存在部分差异，安装的时候注意一下就行 
若在caffe编译的时候出现/usr/bin/ld: cannot find -lopenblas 
请在openblas编译文件夹下执行cp lib* ~/usr/local/lib即可

⑤最终的caffe配置。caffe的github地址为https://github.com/BVLC/caffe 

将工程文件拷贝到本地后，执行

cp Makefile.config.example Makefile.config 
vim ./Makefile.config 
修改CUDA_DIR :=/usr/local/cuda-7.5 (按照个人之前安装cuda的路径进行配置) 
修改BLAS := open 
修改BLAS_INCLUDE,BLAS_LIB为安装路径 
保存(如果用户显卡不够用，可以去除前面的CPU_ONLY :=1前面的注释符号”#”，表示只使用CPU做运算)

接下来就可以正常的执行

make all 如果编译速度慢，可以使用make all -jN (N为核心处理器的个数) 
make test 对编译结果进行测试

下面需要对用户根目录下的.bashrc文件进行编辑，加载一些动态库

vim ~/.bashrc 
文本最后加入 
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-7.5/lib64 
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/glog/lib 
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/lib

再次进入caffe目录，执行make runtest

参考http://blog.csdn.net/s2392735818/article/details/49796017