在MacOS安装Caffe环境以及相关的问题

#: 安装流程参考自《深度学习 21天实战Caffe》——赵永科著

（1）安装homebrew包管理工具，作用相当于yum或apt-get:

$ ruby -e "(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install)"

（2）等待安装成功，然后利用该工具安装caffe依赖包:

$ brew install -vd snappy leveldb gflags glog szip lmdb
$ brew tap homebrew/science
$ brew install hdf5 opencv
$ brew install protobuf boost wget

如果python是3.x版本，需要安装boost-python3：

$ brew install boost-python3

（3）下载Caffe源码:

$ git clone https://github.com/bvlc/caffe.git
$ cd caffe/
$ mv Makefile.config.example Makefile.config

（4）修改Makefile.config，打开CPU_ONLY选项，修改BLAS选项，保存。

#: 仅CPU模式开关，打开该选项（去掉“#”）表示Caffe编译时仅支持CPU，不支持GPU

CPU_ONLY := 1

BLAS := atlas

如果python版本是3.x，需要修改PYTHON_LIBRARIES，因为我的电脑是python3.6，所以这里是boost_python36 python3.6:

PYTHON_LIBRARIES := boost_python36 python3.6

（5）执行make进行编译:

#: -j 选项表示使用多线程编译，利用所有可用的CPU，加快编译速度

#: 也可指定数字，如-j8表示开启8个线程编译

$ make -j

等待编译成功。

遇到的问题：
1、

Error: Your Xcode (8.3.3) is too outdated.
Please update to Xcode 9.1 (or delete it).
Xcode can be updated from the App Store.

Error: Kernel.exit
/usr/local/Homebrew/Library/Homebrew/cmd/install.rb:305:in `exit'
/usr/local/Homebrew/Library/Homebrew/cmd/install.rb:305:in `check_development_tools'
/usr/local/Homebrew/Library/Homebrew/cmd/install.rb:320:in `perform_preinstall_checks'
/usr/local/Homebrew/Library/Homebrew/cmd/install.rb:223:in `install'
/usr/local/Homebrew/Library/Homebrew/brew.rb:100:in `<main>'

XCode版本过低，需要升级。只需要从App Store中升级XCode到最新版本就好。
2、

Undefined symbols for architecture x86_64:
  "cv::imread(cv::String const&, int)", referenced from:
      caffe::WindowDataLayer<float>::load_batch(caffe::Batch<float>*) in window_data_layer.o
      caffe::WindowDataLayer<double>::load_batch(caffe::Batch<double>*) in window_data_layer.o
      caffe::ReadImageToCVMat(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&, int, int, bool) in io.o
  "cv::imdecode(cv::_InputArray const&, int)", referenced from:
      caffe::DecodeDatumToCVMatNative(caffe::Datum const&) in io.o
      caffe::DecodeDatumToCVMat(caffe::Datum const&, bool) in io.o
  "cv::imencode(cv::String const&, cv::_InputArray const&, std::__1::vector<unsigned char, std::__1::allocator<unsigned char> >&, std::__1::vector<int, std::__1::allocator<int> > const&)", referenced from:
      caffe::ReadImageToDatum(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&, int, int, int, bool, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&, caffe::Datum*) in io.o
ld: symbol(s) not found for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)
make: *** [.build_release/lib/libcaffe.so.1.0.0] Error 1

编译时没有添加相关依赖库。需要添加opencv_highgui opencv_imgproc opencv_core opencv_imgcodecs opencv_videoio这几个依赖库到Makefile的LIBRARIES字段后面，make clean后重新编译即可。
3、
make成功之后，调用caffe：import caffe
报错：segmentation fault :11
是因为make时候使用的Python的lib和include地址不正确。
注意Makefile.config中此处代码：

# NOTE: this is required only if you will compile the python interface.
# We need to be able to find Python.h and numpy/arrayobject.h.
PYTHON_INCLUDE := /usr/include/python2.7 \
        /usr/lib/python2.7/dist-packages/numpy/core/include

这个配置需要修改为你自己的可以找到Python.h和numpy/arrayobject.h两个文件的路径。
然后注意Makefile.config中此处代码：

# We need to be able to find libpythonX.X.so or .dylib.
PYTHON_LIB := /usr/lib

这个配置需要修改为你自己的可以找到libpythonX.X.so或者libpythonX.X.dylib的路径，此处注意，这个路径要和上面PYTHON_INCLUDE的路径保持版本统一，要么都是自己装的目录，要么都是系统自带的。
比如：我的系统自带的libpython.dylib在系统自带路径/usr/lib和后来自己通过brew安装的路径/usr/local/Cellar/python/2.7.14/Frameworks/Python.framework/Versions/2.7/lib/下都有，但是上面PYTHON_INCLUDE我配置的是

PYTHON_INCLUDE := /usr/local/Cellar/python/2.7.14/Frameworks/Python.framework/Versions/2.7/include/python2.7 \
/usr/local/Cellar/numpy/1.13.3/lib/python2.7/site-packages/numpy/core/include

所以PYTHON_LIB我要配置为/usr/local/Cellar/python/2.7.14/Frameworks/Python.framework/Versions/2.7/lib/而不是/usr/lib。
修改完之后，make clean，然后重新make -j。编译完成，再import caffe发现就不报segmentation fault :11错了。