{"id":1403,"date":"2016-07-13T12:07:27","date_gmt":"2016-07-13T03:07:27","guid":{"rendered":"https:\/\/d-itlab.co.jp\/?p=1403"},"modified":"2026-03-10T12:07:37","modified_gmt":"2026-03-10T03:07:37","slug":"20160713","status":"publish","type":"post","link":"https:\/\/d-itlab.co.jp\/en\/blog\/20160713\/","title":{"rendered":"How to build matcaffe on OSX10.11.5 and MATLAB2016a"},"content":{"rendered":"

I guess that many engineers have some troubles when they try to build matcaffe on OSX.<\/p>\n

1. Link error \u2013 google::protobuf
\n2. Link error \u2013 mex functions \u2013 not found _mxArrayToString
\n3. Link error \u2013 dyld: Library not loaded: @rpath\/libmkl_intel_lp64.dylib<\/p>\n

I found a way to avoid these troubles, so I will write down about it.<\/p>\n

At first, my system is<\/p>\n

\u2013 Mac Pro (Late 2013)
\n\u2013 OSX 10.11.5
\n\u2013 MATLAB 2016a
\n\u2013 Intel Math Kernel Library(or OpenBLAS via homebrew)
\n\u2013 without Python interface
\n\u2013 without CUDA<\/p>\n

1. Install some packages according to \u201cOSX Installation\u201d.<\/h3>\n

http:\/\/caffe.berkeleyvision.org\/install_osx.html<\/a><\/p>\n

2. Install Intel Math Kernel Library.<\/h3>\n

https:\/\/software.intel.com\/en-us\/intel-mkl\/try-buy<\/a><\/p>\n

Or, you can use OpenBLAS or Accelerate.framework.<\/p>\n

3. Edit Makefile.config.<\/h3>\n

You have to edit Makefile.config<\/code> very carefully.<\/p>\n

Make it use only CPU.<\/p>\n

# CPU-only switch (uncomment to build without GPU support).\r\nCPU_ONLY := 1<\/code><\/pre>\n
Specify compiler.<\/p>\n
# To customize your choice of compiler, uncomment and set the following.\r\n# N.B. the default for Linux is g++ and the default for OSX is clang++\r\nCUSTOM_CXX := \/usr\/bin\/clang++<\/code><\/pre>\n
Set BLAS path according to your library.
\nThis sample is for Intel Math Kernel Library.<\/p>\n
# BLAS choice:\r\n# atlas for ATLAS (default)\r\n# mkl for MKL\r\n# open for OpenBlas\r\nBLAS := mkl\r\n# Custom (MKL\/ATLAS\/OpenBLAS) include and lib directories.\r\n# Leave commented to accept the defaults for your choice of BLAS\r\n# (which should work)!\r\nBLAS_INCLUDE := \/opt\/intel\/compilers_and_libraries_2016.3.170\/mac\/mkl\/include\r\nBLAS_LIB := \/opt\/intel\/compilers_and_libraries_2016.3.170\/mac\/mkl\/lib<\/code><\/pre>\n
Set MATLAB path to build matcaffe.<\/p>\n
# This is required only if you will compile the matlab interface.\r\n# MATLAB directory should contain the mex binary in \/bin.\r\n# MATLAB_DIR := \/usr\/local\r\nMATLAB_DIR := \/Applications\/MATLAB_R2016a.app<\/code><\/pre>\n
This is the most important point.
\nRemove paths for python and keep the under order of LIBRARY_DIRS\u2019s elements in order to avoid link error.<\/p>\n
# Whatever else you find you need goes here.\r\nINCLUDE_DIRS := \/usr\/local\/include\r\nLIBRARY_DIRS := \/usr\/local\/lib $(MATLAB_DIR)\/bin\/maci64 \/usr\/lib<\/code><\/pre>\n
Next, build caffe.<\/p>\n
4. Build caffe and matcaffe<\/h3>\n
If you want to use Intel Math Kernel Library, you have to execute this command before it.
\nI recommend you should copy this command into .bash_profile<\/code>.<\/p>\n
# build caffe\r\nsource \/opt\/intel\/compilers_and_libraries\/mac\/mkl\/bin\/mklvars.sh intel64 ilp64\r\ncd caffe\r\nmake all\r\nmake all test\r\nmake runtest\r\nmake all matcaffe<\/code><\/pre>\n
If failed to build caffe, check Makefile.config again.<\/p>\n
5. How to use matcaffe.<\/h3>\n
You have to run MATLAB from terminal.app because MATLAB does not load environment variables when it runs via Finder.app\u2026\u2026<\/p>\n
cd caffe\r\n# You have set this environment variable\r\n# in order to avoid caffe load MATLAB's tiff library....\r\nexport DYLD_INSERT_LIBRARIES=\/usr\/local\/opt\/libtiff\/lib\/libtiff.5.dylib\r\n\/Applications\/MATLAB_R2016a.app\/bin\/matlab &<\/code><\/pre>\n
You can use matcaffe in MATLAB console.<\/p>\n
cd caffe\r\naddpath('.\/matlab');\r\ncaffe.set_mode_cpu();<\/code><\/pre>\n
That\u2019s all! Have fun!<\/p>\n","protected":false},"excerpt":{"rendered":"
I guess that many engineers have some troubles when they try to build matcaffe on OSX. 1. Link error \u2013 google::protobuf 2. Link error \u2013 mex functions \u2013 not found _mxArrayToString 3. Link error \u2013 dyld: Library not loaded: @rpath\/libmkl_intel_lp64.dylib I found a way to avoid these troubles, so I will write down about it. At first, my system is \u2013 Mac Pro (Late 2013) \u2013 OSX 10.11.5 \u2013 MATLAB 2016a \u2013 Intel Math Kernel Library(or OpenBLAS via homebrew) \u2013 […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_locale":"en_US","_original_post":"https:\/\/d-itlab.co.jp\/?p=621","_jetpack_feature_clip_id":0,"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_post_was_ever_published":false},"categories":[51],"tags":[],"kind_blog":[],"member":[49],"keyword_tech":[],"class_list":["post-1403","post","type-post","status-publish","format-standard","hentry","category-programming","member-49","en-US"],"acf":[],"aioseo_notices":[],"aioseo_head":"\n\t\t\n\t\n\t\n\t\n\t\n\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t