MacOS搭建Caffe

青旅半醒 2022-07-13 10:27 134阅读 0赞

电脑有英伟达的显卡，查了一下型号是可以用cuda的，所以尝试了一下，大概的步骤是这样的：

**1、我的环境中装过了Anaconda,Python开发工具打包了一大包的一键式傻瓜安装的东西，没装想装的可以参考：**

http://blog.csdn.net/yimingsilence/article/details/52852309

http://blog.csdn.net/yimingsilence/article/details/52851919

**2、我的环境中装了homebrew，没有的参考这个**

http://blog.csdn.net/yimingsilence/article/details/52438228

**3、安装cuda，去官网上按照cuda driver -> cuda toolkit -> cuda sdk 的顺序来，具体看官网吧**

[https://developer.nvidia.com/cuda-downloads][https_developer.nvidia.com_cuda-downloads]

装完以后配一下环境变量，运行下面两句，并加到~/.bash\_profile里面

$ export PATH=/usr/local/cuda/bin:$PATH  
$ DYLD\_LIBRARY\_PATH=/usr/local/cuda/lib:$DYLD\_LIBRARY\_PATH

nvcc -V可以看一下装好之后的版本

**4、还要安装cuDNN**

1). 从https://developer.nvidia.com/cudnn页面下载并安装cuDNN库（别忘了在Makefile.config中取消USE\_CUDNN := 1的注释）  
2). tar -xzvf cudnn-6.5-osx-v2.tgz  
3). cd cudnn-6.5-osx-v2  
4). sudo cp lib\* /usr/local/cuda/lib（拷贝cuDNN的链接库文件）  
5). sudo cp cudnn.h /usr/local/cuda/include/（拷贝cuDNN的头文件）

**5、装一堆依赖的东西**

安装snappy,leveled,gflags,glog,szip,lmdb,hdf5,opencv..

**\[html\]** [view plain][] [copy][view plain]

[![在CODE上查看代码片][CODE]][CODE_CODE] [![派生到我的代码片][ico_fork.svg]][ico_fork.svg 1]

1.  brew install --fresh -vd snappy leveldb gflags glog szip lmdb hdf5 homebrew/science/opencv

安装protobuf以及boost

**\[html\]** [view plain][] [copy][view plain]

[![在CODE上查看代码片][CODE]][CODE_CODE] [![派生到我的代码片][ico_fork.svg]][ico_fork.svg 1]

1.  brew install --build-from-source --with-python --fresh -vd protobuf 
2.   brew install --build-from-source --fresh -vd boost boost-python

安装blas

brew install openblas

**6、安装xcode的时候默认没有安装command line用如下命令安装**

xcode-select --install

**7、编译caffe，这里是有gup用cuda的版本，没有gpu的话就选cuponly那个选项。在Makefile.config后半部分有一些路径，要配成自己的环境路径，在本文的最后贴出了我的版本，大家可以参考，细节不在解释。**

先下载

git clone https://github.com/BVLC/caffe.git

**\[html\]** [view plain][view plain 1] [copy][view plain 1]

[![在CODE上查看代码片][CODE]][CODE_CODE 1] [![派生到我的代码片][ico_fork.svg]][ico_fork.svg 2]

1.  cp Makefile.config.example Makefile.config 
2.  \# In your Makefile.config, make sure to have this line uncommented 
3.  WITH\_PYTHON\_LAYER := 1
4.  \# Unrelatedly, it's also recommended that you use CUDNN 
5.  USE\_CUDNN := 1
6.  make all -j8 && make pycaffe

make test

make runtest

make pytest

**8、如果没有出现错误，并全部通过后，此时make告一段落，但是并不能确定可以在python中import。**

在终端上：

vi etc/profile

或

sudo vi /etc/profile

进入修改系统的环境变量，将 `caffe/python/`添加到Python系统路径：

export PYTHONPATH=path to caffe/python:$PYTHONPATH
    例如：

export PYTHONPATH=/Users/cuiqi/Downloads/caffe-master/python:$PYTHONPATH

**9、其中出现过得问题及解决方法**

1）.

In file included from ./include/caffe/util/device\_alternate.hpp:40:

./include/caffe/util/cudnn.hpp:8:10: fatal error: 'caffe/proto/caffe.pb.h' file not found

\#include "caffe/proto/caffe.pb.h"

1 error generated.

make: \*\*\* \[.build\_release/src/caffe/test/test\_accuracy\_layer.o\] Error 1

解决方法

$ protoc src/caffe/proto/caffe.proto --cpp_out=.
    $ mkdir include/caffe/proto
    $ mv src/caffe/proto/caffe.pb.h include/caffe/proto

2)Check failed: error == cudaSuccess (2 vs. 0) out of memory

显存溢出，GPU的显存不够用了，kill掉显存占用高的其他进程或者把batch\_size调小点或者升级硬件，或者用cpu模式

**10、我的Makefile.config:**

\#\# Refer to http://caffe.berkeleyvision.org/installation.html  
\# Contributions simplifying and improving our build system are welcome!  
  
  
\# cuDNN acceleration switch (uncomment to build with cuDNN).  
USE\_CUDNN := 1  
  
  
\# CPU-only switch (uncomment to build without GPU support).  
\# CPU\_ONLY := 1  
  
  
\# uncomment to disable IO dependencies and corresponding data layers  
\# USE\_OPENCV := 0  
\# USE\_LEVELDB := 0  
\# USE\_LMDB := 0  
  
  
\# uncomment to allow MDB\_NOLOCK when reading LMDB files (only if necessary)  
\# You should not set this flag if you will be reading LMDBs with any  
\# possibility of simultaneous read and write  
ALLOW\_LMDB\_NOLOCK := 1  
  
  
\# Uncomment if you're using OpenCV 3  
\# OPENCV\_VERSION := 3  
  
  
\# To customize your choice of compiler, uncomment and set the following.  
\# N.B. the default for Linux is g++ and the default for OSX is clang++  
CUSTOM\_CXX := clang++  
  
  
\# CUDA directory contains bin/ and lib/ directories that we need.  
CUDA\_DIR := /usr/local/cuda  
\# On Ubuntu 14.04, if cuda tools are installed via  
\# "sudo apt-get install nvidia-cuda-toolkit" then use this instead:  
\# CUDA\_DIR := /usr  
  
  
\# CUDA architecture setting: going with all of them.  
\# For CUDA < 6.0, comment the \*\_50 lines for compatibility.  
CUDA\_ARCH := -gencode arch=compute\_20,code=sm\_20 \\  
\-gencode arch=compute\_20,code=sm\_21 \\  
\-gencode arch=compute\_30,code=sm\_30 \\  
\-gencode arch=compute\_35,code=sm\_35 \\  
\-gencode arch=compute\_50,code=sm\_50 \\  
\-gencode arch=compute\_50,code=compute\_50  
  
  
\# BLAS choice:  
\# atlas for ATLAS (default)  
\# mkl for MKL  
\# open for OpenBlas  
BLAS := open  
\# Custom (MKL/ATLAS/OpenBLAS) include and lib directories.  
\# Leave commented to accept the defaults for your choice of BLAS  
\# (which should work)!  
\# BLAS\_INCLUDE := /path/to/your/blas  
\# BLAS\_LIB := /path/to/your/blas  
  
  
\# Homebrew puts openblas in a directory that is not on the standard search path  
BLAS\_INCLUDE := $(shell brew --prefix openblas)/include  
BLAS\_LIB := $(shell brew --prefix openblas)/lib  
  
  
\# This is required only if you will compile the matlab interface.  
\# MATLAB directory should contain the mex binary in /bin.  
\# MATLAB\_DIR := /usr/local  
MATLAB\_DIR := /Applications/MATLAB\_R2014b.app  
  
  
\# 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 Python distribution is quite popular. Include path:  
\# Verify anaconda location, sometimes it's in root.  
ANACONDA\_HOME := $(HOME)/anaconda  
PYTHON\_INCLUDE := $(ANACONDA\_HOME)/include \\  
$(ANACONDA\_HOME)/include/python2.7 \\  
$(ANACONDA\_HOME)/lib/python2.7/site-packages/numpy/core/include \\  
  
  
\# Uncomment to use Python 3 (default is Python 2)  
\# PYTHON\_LIBRARIES := boost\_python3 python3.5m  
\# PYTHON\_INCLUDE := /usr/include/python3.5m \\  
\# /usr/lib/python3.5/dist-packages/numpy/core/include  
  
  
\# We need to be able to find libpythonX.X.so or .dylib.  
\# PYTHON\_LIB := /usr/lib  
PYTHON\_LIB := $(ANACONDA\_HOME)/lib  
  
  
\# Homebrew installs numpy in a non standard path (keg only)  
PYTHON\_INCLUDE += $(dir $(shell python -c 'import numpy.core; print(numpy.core.\_\_file\_\_)'))/include  
\# PYTHON\_LIB += $(shell brew --prefix numpy)/lib  
PYTHON\_LIB +=$(ANACONDA\_HOME)/lib/python2.7/site-packages/numpy/core/lib  
  
  
\# Uncomment to support layers written in Python (will link against Python libs)  
WITH\_PYTHON\_LAYER := 1  
  
  
\# Whatever else you find you need goes here.  
INCLUDE\_DIRS := $(PYTHON\_INCLUDE) /usr/local/include /usr/local/Cellar/boost/1.63.0/include  
LIBRARY\_DIRS := $(PYTHON\_LIB) /usr/local/lib /usr/lib /usr/local/Cellar/boost/1.63.0/lib /usr/local/opt/boost/lib  
  
  
\# If Homebrew is installed at a non standard location (for example your home directory) and you use it for general dependencies  
\# INCLUDE\_DIRS += $(shell brew --prefix)/include  
\# LIBRARY\_DIRS += $(shell brew --prefix)/lib  
  
  
\# Uncomment to use \`pkg-config\` to specify OpenCV library paths.  
\# (Usually not necessary -- OpenCV libraries are normally installed in one of the above $LIBRARY\_DIRS.)  
\# USE\_PKG\_CONFIG := 1  
  
  
\# N.B. both build and distribute dirs are cleared on \`make clean\`  
BUILD\_DIR := build  
DISTRIBUTE\_DIR := distribute  
  
  
\# Uncomment for debugging. Does not work on OSX due to https://github.com/BVLC/caffe/issues/171  
\# DEBUG := 1  
  
  
\# The ID of the GPU that 'make runtest' will use to run unit tests.  
TEST\_GPUID := 0  
  
  
\# enable pretty build (comment to see full commands)  
Q ?= @

**11、其他参考：**

macos安装cuda：

http://blog.sina.com.cn/s/blog\_6a507d440100l4z7.html

macos搭建caffe：

http://cs2.swfc.edu.cn/~zyl/?p=2172  
  
  
http://www.tuicool.com/articles/FjAnqin  
  
  
http://www.cnblogs.com/keyky/p/6209456.html  
  
  
http://www.jianshu.com/p/0a5e5af2627f  
  
  
http://www.linuxidc.com/Linux/2016-06/132548.htm  
  
  
http://blog.csdn.net/brightming/article/details/51106629

[https_developer.nvidia.com_cuda-downloads]: https://developer.nvidia.com/cuda-downloads
[view plain]: http://blog.csdn.net/ddreaming/article/details/52739893#
[CODE]: https://code.csdn.net/assets/CODE_ico.png
[CODE_CODE]: https://code.csdn.net/snippets/1914013
[ico_fork.svg]: https://code.csdn.net/assets/ico_fork.svg
[ico_fork.svg 1]: https://code.csdn.net/snippets/1914013/fork
[view plain 1]: http://blog.csdn.net/u010678153/article/details/52639267#
[CODE_CODE 1]: https://code.csdn.net/snippets/1896851
[ico_fork.svg 2]: https://code.csdn.net/snippets/1896851/fork