Intro.

本篇blog紀錄在macOS10.12上，部署Caffe

build with Anaconda python2.7 , opencv3 and Matlab 2016b .

Install Dependency

至官網安裝anaconda
使用Homebrew 安裝library

brew install -vd snappy leveldb gflags glog szip lmdb
brew tap homebrew/science
brew install opencv3 hdf5
brew link opencv3 --force

可以使用anaconda的 hdf5，不用brew install hdf5
更新：使用brew install hdf5 比較不會有link不到的問題

「坑」opencv3 是keg only，記得要link

brew search boost 
    boost                              boost@1.55
    boost-bcp                          boost@1.57
    boost-build                        boost@1.59 ✔
    boost-mpi                          boost@1.60
    boost-python                       homebrew/science/boost-compute
    boost-python@1.59 ✔
 
brew install --build-from-source -vd boost@1.59 boost-python@1.59 
brew link --force boost@1.59

「坑」建議安裝1.59版本的boost，才能正常編譯

Caffe中把某些依賴庫所在的文件夾名字限定為boost，可以將/usr/local/opt/boost@1.59，改名為/usr/local/opt/boost。

brew install --build-from-source --with-python -vd protobuf

Install Caffe

cd (安裝目的地)
git clone https://github.com/BVLC/caffe.git
cd caffe
cp Makefile.config.example Makefile.config

修改Makefile.config

此步驟很重要，要很細心地去依照自己安裝的依賴，去更改內容！
注意這邊要修改的是Makefile.config，不是Makefile.config.example
打開 Makefile.config 客製化編譯環境

由於macbook沒有GPU，只用CPU跑，故要去掉cpu only的註解

# CPU-only switch (uncomment to build without GPU support).
CPU_ONLY := 1

「坑」註解掉USE LEVELDB := 0 才不會在最後編譯matlab接口的時候出錯

# uncomment to disable IO dependencies and corresponding data layers
# USE_OPENCV := 0
USE_LEVELDB := 0
# USE_LMDB := 0

使用opencv3，故要註解掉此行

# Uncomment if you're using OpenCV 3
OPENCV_VERSION := 3

有些教學會推薦使用openblas，或是用intel的MKL，但我這邊就用預設的，若有改變記得將BLAS:=改為自己使用的BLAS

# BLAS choice:
# atlas for ATLAS (default)
# mkl for MKL
# open for OpenBlas
BLAS := atlas

去掉matlab註解，並修改matlab路徑，我是2016b

MATLAB_DIR := /Applications/MATLAB_R2016b.app

因為要使用anaconda的python，所以這邊要去掉anaconda註解，並修改anaconda的正確安裝位置，並把原本的PYTHON INCLUDE註解掉，避免路徑衝突。

# 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
ANACONDA_HOME := $(HOME)/anaconda
PYTHON_INCLUDE := $(ANACONDA_HOME)/include \
        $(ANACONDA_HOME)/include/python2.7 \
        $(ANACONDA_HOME)/lib/python2.7/site-packages/numpy/core/include

使用 anaconda python 的library

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

如果要編譯Matlab的接口，要在LIBRARY DIR這邊加入maci64的路徑，
尤其各個路徑排列順序要對，如下所示。

由於編譯會有找不到opencv的問題，故在INCLUDE DIR & LIBRARY DIR 後面加上opencv3的路徑

# Whatever else you find you need goes here.
INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/local/Cellar/opencv3/3.2.0/include
LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /Applications/MATLAB_R2016b.app/bin/maci64 /usr/lib /usr/local/Cellar/opencv3/3.2.0/lib

編譯caffe

修改完成後

make all -j4
make test -j4
make runtest -j4
make matcaffe
make mattest

FAQ

由於caffe 的dependency很多，建議找一個月內的教程來安裝，避免某一個dependency版本與caffe編譯衝突，小編依照半年前自己紀錄的教程來安裝，發現一堆坑，後來將boost版本改為1.59，用opencv2，才順利解決。
以下就稍微紀錄一些我遇到的問題。

編譯期間會有很多問題，只能慢慢google，並解決。

Runtest 可能會報錯：

error：make runtest
.build_release/tools/caffe
dyld: Library not loaded: @rpath/libhdf5_hl.10.dylib
  Referenced from: /Users/work/gitclone/caffe/.build_release/tools/caffe
  Reason: image not found
make: *** [runtest] Abort trap: 6

原因是Library not loaded，需要手動加入，如下(需改為自己的路徑)

install_name_tool -add_rpath '/Users/junhao/anaconda/lib'  /Users/junhao/gitclone/caffe/.build_release/tools/caffe

繼續出錯：

dyld: Library not loaded: @rpath/libhdf5_hl.10.dylib
  Referenced from: /Users/work/gitclone/caffe/.build_release/test/test_all.testbin
  Reason: image not found
make: *** [runtest] Abort trap: 6

一樣的問題，再度手動加入

install_name_tool -add_rpath '/Users/junhao/anaconda/lib'  /Users/work/gitclone/caffe/.build_release/test/test_all.testbin

出錯：

MATLAB is exiting because of fatal error
/bin/sh: line 1: 14220 Killed: 9 /Applications/MATLAB_R2016b.app/bin/matlab -nodisplay -r 'caffe.run_tests(), exit()'
make: *** [mattest] Error 137

把Makefile.conig中的leveldb=0註解掉即可mattest成功

最後再次
make runtest
make matcaffe
make mattest
就成功了。