1## 2.4 Caffe环境搭建 2 3### 注意：如果您的Ubuntu是20.04版本的 4 5* 请参考《[如何在Ubuntu20.02 上面搭建caffe环境](https://gitee.com/wgm2022/wu_guiming.gitee.io/blob/master/01%20%E7%8E%AF%E5%A2%83%E6%90%AD%E5%BB%BA%E7%9B%B8%E5%85%B3/%E5%A6%82%E4%BD%95%E5%9C%A8Ubuntu20.02%20%E4%B8%8A%E9%9D%A2%E6%90%AD%E5%BB%BAcaffe%E7%8E%AF%E5%A2%83.md)》 6 7### 如果您的Ubuntu是18.04版本的，请继续往下操作 8 9* 1、在Ubuntu系统中，分步执行下面的命令，对Ubuntu里面的软件进行更新。 10 11```sh 12sudo apt-get update 13sudo apt-get upgrade 14``` 15 16* 2、分步执行下面的命令，安装所需的依赖软件。 17 18```sh 19sudo apt-get install -y libopencv-dev -y 20sudo apt-get install -y build-essential cmake git pkg-config -y 21sudo apt-get install -y libprotobuf-dev libleveldb-dev libsnappy-dev libhdf5-serial-dev protobuf-compiler -y 22sudo apt-get install -y liblapack-dev -y 23sudo apt-get install -y libatlas-base-dev -y 24sudo apt-get install -y --no-install-recommends libboost-all-dev -y 25sudo apt-get install -y libgflags-dev libgoogle-glog-dev liblmdb-dev -y 26sudo apt-get install -y python-numpy python-scipy -y 27sudo apt-get install -y python3-pip -y 28sudo apt-get install -y python3-numpy python3-scipy -y 29``` 30 31* 3、执行下面的命令，下载caffe开源软件 32 33```sh 34git clone https://github.com/BVLC/caffe.git 35``` 36 37* 4、进入caffe/python/目录下，执行下面的命令，下载依赖的软件 38 39```sh 40cd caffe/python 41for req in $(cat requirements.txt); do pip3 install $req; done 42``` 43 44* 5、进入caffe目录下，执行下面的命令，将 Makefile.config.example 文件复制一份并更名为 Makefile.config 45 46```sh 47cp Makefile.config.example Makefile.config 48``` 49 50* 6、接下来是修改Makefile.config里面的配置,使用vim工具打开Makefile.config。 51 52```sh 53vim Makefile.config 54``` 55 56* ① 将CPU_ONLY前面的注释去掉。 57 58```sh 59# 将 60# CPU_ONLY := 1 61# 改为 62CPU_ONLY := 1 63``` 64 65* ② 将OPENCV_VERSION前面的注释去掉 66 67```sh 68# 将 69# OPENCV_VERSION := 3 70# 改为 71OPENCV_VERSION := 3 72``` 73 74* ③ 因为我们Ubuntu的环境是python3.6，所以请把PYTHON_INCLUDE = python2.7这个配置前面加上注释，且把PYTHON_INCLUDE=python3.5的注释打开，把所有的3.5都改成3.6,具体修改如下： 75* 注意：如果您的python3是其他的版本，就把python3.5改成您Ubuntu对应的python3版本即可 76 77 78 79* ④ 将WITH_PYTHON_LAYER := 1前面的注释去掉 80 81```sh 82# 将 83# WITH_PYTHON_LAYER := 1 84# 改为 85WITH_PYTHON_LAYER := 1 86``` 87 88* ⑤ 修改INCLUDE_DIRS和LIBRARY_DIRS 89 90```sh 91# 将 92INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include 93LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib 94改为 95INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial 96LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib usr/lib/x86_64-linux-gnu /usr/lib/x86_64-linux-gnu/hdf5/serial 97``` 98 99修改后的文件如下所示： 100 101```python 102## Refer to http://caffe.berkeleyvision.org/installation.html 103# Contributions simplifying and improving our build system are welcome! 104 105# cuDNN acceleration switch (uncomment to build with cuDNN). 106# USE_CUDNN := 1 107 108# CPU-only switch (uncomment to build without GPU support). 109CPU_ONLY := 1 110 111# uncomment to disable IO dependencies and corresponding data layers 112# USE_OPENCV := 0 113# USE_LEVELDB := 0 114# USE_LMDB := 0 115# USE_HDF5 := 0 116 117# uncomment to allow MDB_NOLOCK when reading LMDB files (only if necessary) 118# You should not set this flag if you will be reading LMDBs with any 119# possibility of simultaneous read and write 120# ALLOW_LMDB_NOLOCK := 1 121 122# Uncomment if you're using OpenCV 3 123OPENCV_VERSION := 3 124 125# To customize your choice of compiler, uncomment and set the following. 126# N.B. the default for Linux is g++ and the default for OSX is clang++ 127# CUSTOM_CXX := g++ 128 129# CUDA directory contains bin/ and lib/ directories that we need. 130CUDA_DIR := /usr/local/cuda 131# On Ubuntu 14.04, if cuda tools are installed via 132# "sudo apt-get install nvidia-cuda-toolkit" then use this instead: 133# CUDA_DIR := /usr 134 135# CUDA architecture setting: going with all of them. 136# For CUDA < 6.0, comment the *_50 through *_61 lines for compatibility. 137# For CUDA < 8.0, comment the *_60 and *_61 lines for compatibility. 138# For CUDA >= 9.0, comment the *_20 and *_21 lines for compatibility. 139CUDA_ARCH := -gencode arch=compute_20,code=sm_20 \ 140 -gencode arch=compute_20,code=sm_21 \ 141 -gencode arch=compute_30,code=sm_30 \ 142 -gencode arch=compute_35,code=sm_35 \ 143 -gencode arch=compute_50,code=sm_50 \ 144 -gencode arch=compute_52,code=sm_52 \ 145 -gencode arch=compute_60,code=sm_60 \ 146 -gencode arch=compute_61,code=sm_61 \ 147 -gencode arch=compute_61,code=compute_61 148 149# BLAS choice: 150# atlas for ATLAS (default) 151# mkl for MKL 152# open for OpenBlas 153BLAS := atlas 154# Custom (MKL/ATLAS/OpenBLAS) include and lib directories. 155# Leave commented to accept the defaults for your choice of BLAS 156# (which should work)! 157# BLAS_INCLUDE := /path/to/your/blas 158# BLAS_LIB := /path/to/your/blas 159 160# Homebrew puts openblas in a directory that is not on the standard search path 161# BLAS_INCLUDE := $(shell brew --prefix openblas)/include 162# BLAS_LIB := $(shell brew --prefix openblas)/lib 163 164# This is required only if you will compile the matlab interface. 165# MATLAB directory should contain the mex binary in /bin. 166# MATLAB_DIR := /usr/local 167# MATLAB_DIR := /Applications/MATLAB_R2012b.app 168 169# NOTE: this is required only if you will compile the python interface. 170# We need to be able to find Python.h and numpy/arrayobject.h. 171# PYTHON_INCLUDE := /usr/include/python2.7 \ 172# /usr/lib/python2.7/dist-packages/numpy/core/include 173# Anaconda Python distribution is quite popular. Include path: 174# Verify anaconda location, sometimes it's in root. 175# ANACONDA_HOME := $(HOME)/anaconda 176# PYTHON_INCLUDE := $(ANACONDA_HOME)/include \ 177 # $(ANACONDA_HOME)/include/python2.7 \ 178 # $(ANACONDA_HOME)/lib/python2.7/site-packages/numpy/core/include 179 180# Uncomment to use Python 3 (default is Python 2) 181PYTHON_LIBRARIES := boost_python3 python3.6m 182PYTHON_INCLUDE := /usr/include/python3.6m \ 183 /usr/lib/python3.6/dist-packages/numpy/core/include 184 185# We need to be able to find libpythonX.X.so or .dylib. 186PYTHON_LIB := /usr/lib 187# PYTHON_LIB := $(ANACONDA_HOME)/lib 188 189# Homebrew installs numpy in a non standard path (keg only) 190# PYTHON_INCLUDE += $(dir $(shell python -c 'import numpy.core; print(numpy.core.__file__)'))/include 191# PYTHON_LIB += $(shell brew --prefix numpy)/lib 192 193# Uncomment to support layers written in Python (will link against Python libs) 194WITH_PYTHON_LAYER := 1 195 196# Whatever else you find you need goes here. 197INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial 198LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib usr/lib/x86_64-linux-gnu /usr/lib/x86_64-linux-gnu/hdf5/serial 199 200# If Homebrew is installed at a non standard location (for example your home directory) and you use it for general dependencies 201# INCLUDE_DIRS += $(shell brew --prefix)/include 202# LIBRARY_DIRS += $(shell brew --prefix)/lib 203 204# NCCL acceleration switch (uncomment to build with NCCL) 205# https://github.com/NVIDIA/nccl (last tested version: v1.2.3-1+cuda8.0) 206# USE_NCCL := 1 207 208# Uncomment to use `pkg-config` to specify OpenCV library paths. 209# (Usually not necessary -- OpenCV libraries are normally installed in one of the above $LIBRARY_DIRS.) 210# USE_PKG_CONFIG := 1 211 212# N.B. both build and distribute dirs are cleared on `make clean` 213BUILD_DIR := build 214DISTRIBUTE_DIR := distribute 215 216# Uncomment for debugging. Does not work on OSX due to https://github.com/BVLC/caffe/issues/171 217# DEBUG := 1 218 219# The ID of the GPU that 'make runtest' will use to run unit tests. 220TEST_GPUID := 0 221 222# enable pretty build (comment to see full commands) 223Q ?= @ 224``` 225 226* 7、修改Makefile文件里面的一些配置，使用vim 打开Makefile，进行修改。 227 228``` 229vim Makefile 230``` 231 232* ① 修改DYNAMIC_VERSION_REVISION的值 233 234```sh 235# 将 236DYNAMIC_VERSION_REVISION := 0 237# 改为 238DYNAMIC_VERSION_REVISION := 0-rc3 239``` 240 241* ② 修改LIBRARIES的值 242 243```sh 244# 将 245LIBRARIES += glog gflags protobuf boost_system boost_filesystem m 246# 改为 247LIBRARIES += glog gflags protobuf boost_system boost_filesystem boost_regex m hdf5_hl hdf5 248``` 249 250 251 252```sh 253# 将 254LIBRARIES += opencv_imgcodecs 255# 改为 256LIBRARIES += opencv_imgcodecs opencv_videoio 257``` 258 259 260 261* ③ 将# NCCL acceleration configuration下面的四行注释掉 262 263```sh 264# 将 265# NCCL acceleration configuration 266ifeq ($(USE_NCCL), 1) 267 268 LIBRARIES += nccl 269 COMMON_FLAGS += -DUSE_NCCL 270endif 271# 改为 272# NCCL acceleration configuration 273# ifeq ($(USE_NCCL), 1) 274# LIBRARIES += nccl 275# COMMON_FLAGS += -DUSE_NCCL 276# endif 277``` 278 279 280 281* 8、在caffe目录下，分步执行下面的命令，来编译caffe。 282 283```sh 284make -j4 285make pycaffe 286``` 287 288* 9、执行下面的命令，将caffe的python路径设置为环境变量，并更新环境变量。 289 290 执行下面的命令，打开.bashrc 291 292```sh 293sudo vim ~/.bashrc 294``` 295 296把下面的命令，添加到.bashrc文件的末尾，主要/home/hispark/code/caffe/python是我自己Ubuntu的caffe/python路径，这里请填写为您自己Ubuntu中的caffe/python路径。 297 298```shell 299在文件的末尾加上下面的语句 300export PYTHONPATH=/home/hispark/code/caffe/python:$PYTHONPATH 301``` 302 303 304 305 306 307 在执行下面的命令，更新环境变量 308 309```sh 310source ~/.bashrc 311``` 312 313* 10、测试caffe环境是否OK，在Ubuntu的任意目录下，执行 python3，当出现”>>>”的提示符后，再输入import caffe，如果没有任何报错信息，说明caffe环境已经搭建成功了。 314 315```sh 316python3 317import caffe 318``` 319 320
