ubuntu14.04 编译安装CPU版caffe

 

本文，试图中一个干净的ubuntu14.04机器上安装caffe的cpu版本。

http://blog.csdn.net/sinat_35188997/article/details/73530434

http://caffe.berkeleyvision.org/installation.html

http://blog.csdn.net/zly_kemgine/article/details/78749764

首先卸载掉一些包：

卸载所有cuda有关的包：

apt-get remove --purge cuda-*

卸载掉所有nvidia有关的包：

apt-get remove --purge nvidia-*

卸载掉所有cudnn包：

apt-get remove --purge libcudnn*

卸载掉所有caffe包：

apt-get remove --purge caffe-*

卸载掉conda：

直接删除conda安装目录，然后在环境变量配置文件中删除旧路径即可http://blog.csdn.net/qq_22474567/article/details/54984257

然后使用dpkg -l|grep 或 dpkg -S 软件名 或 find 查看是否卸载干净

因为我是用的是什么都没有的干净的ubuntu系统，所以免去上面的步骤。

然后安装conda：

上官网下载.sh文件，运行，运行过程中配置一下安装目录，确认一下是否要加入环境变量即可，安装完成后重启terminal即可使用

对于caffe的依赖包有哪些，请参考Prerequisites：http://caffe.berkeleyvision.org/installation.html

然后安装openblas：

源码：https://github.com/xianyi/OpenBLAS

安装教程：http://www.tuicool.com/articles/vimi6v

下载openblas源码，解压，进入目录，

make

make完成后，根据输出中的提示，修改Makefile.rule文件中的prefix字段，设置安装目录，然后执行，

make install

（或者也可以直接执行 make PREFIX=安装路径 install 命令）

然后安装protobuf：

注意，caffe暂不支持protobuf3，我们一般选择安装protobuf 2.5.0

源码：https://github.com/google/protobuf

安装教程：http://blog.csdn.net/10km/article/details/53023527

http://blog.csdn.net/nyist327/article/details/49908891

http://blog.csdn.net/missdaddio/article/details/67631777

官方教程上说，直接

$ ./configure --prefix=安装路径
$ make -j16
$make check -j16
$ make install

但是，解压后发现没有configure文件，上网搜索后发现，需要先运行autogen.sh生成configure文件

运行autogen.sh，发现程序试图从一个被墙的网站上下载gtest-1.5.0.tar.bz2，因为机器不能FQ，所以无法下载。

查看autogen.sh的代码，发现下面这段代码：

于是手动下载googletest-release-1.5.0.zip,解压缩后，改名为gtest放在protobuf-2.5.0目录下

再次运行autogen.sh，再次出错

/autogen.sh: 40: autoreconf: not found

搜了一下，发现是因为缺包

解决办法： sudo apt-get install autoconf automake libtool

再次运行autogen.sh，成功生成configure，然后

$ ./configure --prefix=安装路径
$ make -j16
$make check -j16
$ make install

最后运行protoc命令，查看是否安装成功

然后，安装到anaconda中

export PATH=/HOME_ANACONDA_PATH/bin:PATH

#cd ./python
#python setup.py build
#python setup.py test
#python setup.py install

正确安装如图：

然后安装boost：

源码：http://www.boost.org/users/history/version_1_66_0.html（这里注意，不要用github上下载的zip，github上下载下来的zip不全，会报can't cd into ...错误）

教程：

./bootstrap.sh --prefix=/usr --with-python=/opt/anaconda2/bin/ --with-python-root=/opt/anaconda2/

./b2
./b2 install

然后安装glog

源码：https://github.com/google/glog/tree/v0.3.3

教程：

tar zxvf glog-0.3.3.tar.gz
cd glog-0.3.3
./configure ¨C-prefix=/usr
make && make install

然后安装gflags

源码：https://github.com/gflags/gflags

教程：

要先安装cmake：apt-get update & apt-get install cmake

cd gflags-master

mkdir build

cd build
export CXXFLAGS="-fPIC"
cmake -D CMAKE_INSTALL_PREFIX=/usr .. （这里注意最后的两个点，别漏了）
make VERBOSE=1
make
make install

然后安装hdf5

源码：https://support.hdfgroup.org/HDF5/release/obtainsrc518.html

教程：

tar -zxvf hdf5-1.8.3.tar.gz
$ ./configure --prefix=/usr
make
make check # run test suite.
make install
make check-install # verify installation.

然后安装snappy

源码：http://google.github.io/snappy/

教程：https://github.com/google/snappy/blob/master/README.md

cd google-snappy-4f7bd2d

mkdir build

cd build

（export CXXFLAGS='-fPIC'）——这句后来被证明应该加上，以便在编译时产生share文件夹

cmake -D CMAKE_INSTALL_PREFIX=/usr ..

make

make install

然后安装opencv（这里注意一下，尽量编译安装opencv3。我尝试过直接用conda install opencv而不使用编译安装，结果在安装caffe时存在一些问题）

源码：https://github.com/opencv/opencv/tree/3.0.0

教程：https://www.cnblogs.com/kanuore/p/7587434.html（这个是给python3安装opencv的，2同理）

下载源码，解压，进入目录，发现是cmake的项目，故创建build文件夹并在其中cmake。

mkdir build

cd build

执行cmake命令，如下图：

发现报错，如下图，一开始进入CMakeError.log文件中查看，发现有错误提示提到ffmpeg等。以为是缺包，因此安装缺少的包：http://blog.csdn.net/zhang_shuai12/article/details/52289825

sudo apt-get install cmake git libgtk2.0-dev pkg-config libavcodec-dev libavformat-dev libswscale-dev

再次执行camke命令，还是报一样的错。

然后仔细阅读了错误提示中的内容，发现程序试图从一个被墙的网站下载ippicv_linux_20141027.tgz，下载不下来因此报错！

一定要仔细阅读错误提示！如果错误提示看不出什么的话再去看CMakeError.log文件。

根据上面的错误提示，查看3rdparty/ippicv/downloader.cmake文件，或者根据网上搜到的结果，发现需要手动下载ippicv_linux_20141027.tgz并放到指定位置（如下图）

于是从github上下载ippicv_linux_20141027.tgz，https://github.com/pkpp1233/ippicv，下载后解压得到所需压缩包，拷贝到指定文件夹中。

再次执行cmake命令，cmake通过，然后make，makeinstall，都成功了。

但是，因为我在cmake命令中指定了将opencv安装到anaconda中的python中，因此想要在python中看一下opencv是否安装成功

import cv2

结果import失败。这个包并没有被安装上。

回过头来思考，可能是cmake命令的选项出错了，因此仔细阅读了cmakelist，发现cmake命令中的的选项都在CMakeLists.txt中有解释，之前的cmake命令中好多选项写错了：

1、用ctrl查找，发现cublas是cuda的配置，应该不需要，去掉

2、查找之前输入的选项，发现有几个选项查找不出来，则说明是选项名字错了，比如PYTHON_EXECUTEABLE应该为PYTHON2_EXECUTEABLE，PYTHON_LIBRARY应为PYTHON2_LIBRARIES……逐个选项修复

错误的cmake命令：

得到正确的cmake命令：

删掉生成的/opt/opencv，删掉编译目录build，重新mkdir build，cd build，执行正确的cmake命令，make，make install。都通过了。

然后测试import cv2，成功。

总结：

cmake命令安装的时候一定要仔细阅读cmakelist.txt文件

然后安装lmdb

源码：https://github.com/LMDB/lmdb

教程：

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

打开makefile文件，修改prefix字段

make && make install

（或使用make install  DESTDIR=/root/_install_caffe_cpu/lmdb命令）

然后安装leveldb

源码：https://github.com/google/leveldb/tree/v1.10

教程：

解压，进入文件夹

make

make命令发现报错：

在github上发现了关于相似问题的讨论，只不过我们是snappy库报的错，它是gflags库报的错：https://github.com/BVLC/caffe/issues/2171

原来是当时安装snappy的时候应该先执行一句export CXXFLAGS='-fPIC'，这样编译后才会生成share文件夹。

所以重新安装snappy：删除build文件夹，mkdir build，cd build，export CXXFLAGS='-fPIC'，cmake -D CMAKE_INSTALL_PREFIX=/usr ..，make，make install

make install，发现报错，原来leveldb的工具包不支持make install（查看makefile文件，可以发现找不到install命令的定义）

所以需要手动将make出的.a和.so文件拷贝到/usr/lib下，将include文件夹拷贝到/usr/include下

最后，安装caffe：

解压，进入caffe-master，官方推荐使用make方式编译caffe，也可以使用cmake，这里我用make

复制一份Makefile.config.example，改名为Makefile.config

根据自己的需要，修改makefile.config，以下是我的配置：

## 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不使用cudnn

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

# uncomment to disable IO dependencies and corresponding data layers
USE_OPENCV := 1使用opencv
USE_LEVELDB := 1使用leveldb
USE_LMDB := 1使用lmdb

# 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 := 3opencv 3.0.0

# 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 := g++

# 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不使用cuda

# CUDA architecture setting: going with all of them.
# For CUDA < 6.0, comment the *_50 through *_61 lines for compatibility.
# For CUDA < 8.0, comment the *_60 and *_61 lines for compatibility.
# For CUDA >= 9.0, comment the *_20 and *_21 lines for compatibility.
#CUDA_ARCH := -gencode arch=compute_20,code=sm_20 \
# -gencode arch=compute_20,code=sm_21 \不使用cuda，全都注释掉
# -gencode arch=compute_30,code=sm_30 \
# -gencode arch=compute_35,code=sm_35 \
# -gencode arch=compute_50,code=sm_50 \
# -gencode arch=compute_52,code=sm_52 \
# -gencode arch=compute_60,code=sm_60 \
# -gencode arch=compute_61,code=sm_61 \
# -gencode arch=compute_61,code=compute_61

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

BLAS := open使用openblas
# Custom (MKL/ATLAS/OpenBLAS) include and lib directories.
# Leave commented to accept the defaults for your choice of BLAS
# (which should work)!
BLAS_INCLUDE := /usr/includeopenblas安装路径为/usr，则配置其include和lib为/usr/include和/usr/lib
BLAS_LIB := /usr/lib

# 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_R2012b.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 := /opt/anaconda2关键！！使用自己安装的anaconda的python，配置自己安装的anaconda的位置，下同
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

# Uncomment to support layers written in Python (will link against Python libs)
# WITH_PYTHON_LAYER := 1这个不知道开不开？先没开？

# Whatever else you find you need goes here.
INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/include /usr/local/include编译所用的库路径
LIBRARY_DIRS := $(PYTHON_LIB) /usr/lib /usr/local/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

# NCCL acceleration switch (uncomment to build with NCCL)
# https://github.com/NVIDIA/nccl (last tested version: v1.2.3-1+cuda8.0)
# USE_NCCL := 1

# 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 ?= @

修改完成后，

make all

make pycaffe

make test

然后将caffe源码目录中的python文件夹配置到PYTHONPATH这个环境变量里面去（如下图）←注意别漏了这一步

make方式编译的caffe不需要install，直接使用解压的这个文件夹里的东西

测试一下caffe：

测试一下pycaffe：

成功！

参考：

opencv3.0 + cuda 8.0 error
http://blog.csdn.net/yuehanliushuang/article/details/52116926

/*

#include "precomp.hpp"

-#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER)
+// GraphCut has been removed in NPP 8.0
+#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER) || (CUDART_VERSION >= 8000)

void cv::cuda::graphcut(GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }
void cv::cuda::graphcut(GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }

*/

OpenBLAS install
http://www.tuicool.com/articles/vimi6v

Protobuf

Install protobuf 2.4.1
$ tar ¨Cxvf protobuf-2.4.1.tar.gz
$ cd protobuf-2.4.1
$ ./configure --prefix=/root/_install_caffe_cpu/protobuf-2.4.1
$ make -j16
$make check -j16
$ make install

export PATH=/HOME_ANACONDA_PATH/bin:PATH
#cd ./python
#python setup.py build
#python setup.py test
#python setup.py install

Boost

Install boost
./bootstrap.sh --prefix=/home/inf/web_demo/tools/_install_boost_1.58.0_anaconda/ --with-python=/home/inf/anaconda2/bin/ --with-python-root=/home/inf/anaconda2/
./b2
./b2 install

Install LMDB
git clone https://github.com/LMDB/lmdb
cd lmdb/libraries/liblmdb
make && make install DESTDIR=/root/_install_caffe_cpu/lmdb

Install glog-0.3.3
tar zxvf glog-0.3.3.tar.gz
cd glog-0.3.3
./configure ¨C-prefix=/root/_install_caffe_cpu/glog-0.3.3
make && make install

Install gflag
use master.zip

cd gflags-master
export CXXFLAGS="-fPIC"
cmake -D CMAKE_INSTALL_PREFIX=/root/_install_caffe_cpu/gflag .
make VERBOSE=1
make
make install

hdf5

tar -zxvf hdf5-1.8.3.tar.gz
$ ./configure --prefix=/usr/local/hdf5-1.8.3/
make
make check # run test suite.
make install
make check-install # verify installation.

snappy

http://blog.csdn.net/hairetz/article/details/18049257
1.去官网http://code.google.com/p/snappy/下载snappy-1.1.1.tar.gz安装包。

2.加压到指定目录
cd ~/snappy
tar -zxvf snappy-1.1.1.tar.gz
cd snappy-1.1.1
./configure -prefix=/home/tools/_install_snappy
make
make install

opencv 3
error:
cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/opt/opencv -D PYTHON_INCLUDE_DIR=/opt/anaconda2/include/python2.7/ -D PYTHON_INCLUDE_DIR2=/opt/anaconda2/include/python2.7 -D PYTHON_EXECUTABLE=/opt/anaconda2/bin/python -D PYTHON_LIBRARY=/opt/anaconda2/lib/libpython2.7.so -D PYTHON_PACKAGES_PATH=/opt/anaconda2/lib/python2.7/site-packages/ -D BUILD_EXAMPLES=ON -D BUILD_NEW_PYTHON_SUPPORT=ON -D PYTHON2_LIBRARY=/opt/anaconda2/lib/libpython2.7.so -D BUILD_opencv_python3=OFF -D BUILD_opencv_python2=ON -D WITH_CUBLAS=1 ..
true:
cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/opt/opencv -D PYTHON2_INCLUDE_DIR=/opt/anaconda2/include/python2.7/ -D PYTHON2_EXECUTABLE=/opt/anaconda2/bin/python2.7 -D PYTHON2_LIBRARIES=/opt/anaconda2/lib/libpython2.7.so -D PYTHON2_PACKAGES_PATH=/opt/anaconda2/lib/python2.7/site-packages/ -D BUILD_EXAMPLES=ON -D BUILD_NEW_PYTHON_SUPPORT=ON -D BUILD_opencv_python3=OFF -D BUILD_opencv_python2=ON ..

cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/home/tools/_install_opencv_3.0_anaconda/ -D PYTHON_INCLUDE_DIR=/home/tools/anaconda2/include/python2.7/ -D PYTHON_INCLUDE_DIR2=/home/tools/anaconda2/include/python2.7 -D PYTHON_EXECUTABLE=/home/tools/anaconda2/bin/python -D PYTHON_LIBRARY=/home/tools/anaconda2/lib/libpython2.7.so -D PYTHON_PACKAGES_PATH=/home/tools/anaconda2/lib/python2.7/site-packages/ -D BUILD_EXAMPLES=ON -dBUILD_NEW_PYTHON_SUPPORT=ON -D PYTHON2_LIBRARY=/home/tools/anaconda2/lib/libpython2.7.so -D BUILD_opencv_python3=OFF -D BUILD_opencv_python2=ON -D WITH_CUBLAS=1 ..

sudo route add -net 172.17.0.0/17 gw 192.168.43.1
sudo route del -net 172.17.0.0/23 gw 192.168.43.1