Ubuntu18自带python版本为3.6.8，本文流程为
- 系统环境配置caffe（python3.6）
- miniconda3的3.6虚拟环境下运行caffe不报错
1.cuda+miniconda+cudnn+nvidia driver
首先，安装cuda等（之前写过，点这里）
2.系统环境配置caffe
2.1 安装caffe
根据caffe官网，版本高于17.04的ubuntu可以直接安装预编译的caffe，不用分别安装依赖了。
在命令行输入(不是conda里)
sudo apt install caffe-cuda #Gpu版本
sudo apt install caffe-cpu #Cpu版本
2.2 下载源码
cd到准备好的文件夹中，从github上下载caffe的源码
cd ***
git clone https://github.com/BVLC/caffe.git
2.3修改MakeFile.config
cd caffe
sudo cp Makefile.config.example Makefile.config
sudo vim Makefile.config #没有图形界面
sudo gedit Makefile.config #有图形界面
有图形界面直接修改，没有的话使用vim（按i之后修改，按esc退出修改模式，按：wq！保存退出）
修改位置如下
- 第4行
# cuDNN acceleration switch (uncomment to build with cuDNN).
USE_CUDNN := 1
- 第22行
# Uncomment if you're using OpenCV 3
OPENCV_VERSION := 3
- 第25行
# 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++
- 第35行
# 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_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
- 第47行
# BLAS choice:
# atlas for ATLAS (default)
# mkl for MKL
# open for OpenBlas
BLAS:=open
- 第68行
# 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
- 第79行
一定要检测是否有对应的文件夹。按照自己电脑python的路径来写
# Uncomment to use Python 3 (default is Python 2)
PYTHON_LIBRARIES := boost_python3 python3.6m
PYTHON_INCLUDE := /usr/include/python3.6m \
/usr/lib/python3/dist-packages/numpy/core/include
- 第95行
# Whatever else you find you need goes here.
INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial/
LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib /usr/lib/x86_64-linux-gnu/hdf5/serial/
- 第103行
# NCCL acceleration switch (uncomment to build with NCCL)
# https://github.com/NVIDIA/nccl (last tested version: v1.2.3-1+cuda8.0)
USE_NCCL := 1
2.4 修改Makefile
还是用vim或者gedit打开Makefile（sudo）,第425行
NVCCFLAGS += -D_FORCE_INLINES -ccbin=$(CXX) -Xcompiler -fPIC $(COMMON_FLAGS)
2.5 安装caffe需要的包
cd python
for req in $(cat requirements.txt); do pip install $req; done
2.6 编译caffe
cd ..
make all -j48 #j后面的数字代表处理器的核数
此时可能报错nccl.h
compilation terminated.
Makefile:592: recipe for target '.build_release/examples/cpp_classification/classification.o' failed
make: *** [.build_release/examples/cpp_classification/classification.o] Error 1
In file included from ./include/caffe/parallel.hpp:19:0,
from ./include/caffe/caffe.hpp:13,
from tools/upgrade_net_proto_text.cpp:10:
./include/caffe/util/nccl.hpp:5:18: fatal error: nccl.h: 没有那个文件或目录
#include
解决方法：
cd **** #准备好存放nccl的文件夹
git clone https://github.com/NVIDIA/nccl.git
sudo make install -j48
sudo ldconfig
最后一步创建动态链接非常重要
没有报错的话就继续，如果有报错需要重新make一定要先执行make clean
2.7 编译pycaffe
cd python
make pycaffe -j48
没有报错即成功
2.8 检查是否编译成功
***@*** : ****/caffe/python$ python3
Python 3.6.8 (default, Oct 7 2019, 12:59:55) [GCC 8.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import caffe
没有报错即成功
错误1：
ImportError: dynamic module does not define init function (init_caffe)
如果在caffe/python下直接进入python，python版本为2.7，就会出现这个错误，进入python3就解决了
错误2：
ImportError: libnccl.so.2: cannot open shared object file: No such file or directory
2.6安装nccl时没有建立动态链接，cd到nccl目录建立动态链接，cd回caffe目录make clean之后重新make all -j48 ，cd python ，make pycaffe -j48，然后重新检查
3. 在3.6虚拟环境中运行caffe工程
3.1 conda 中建立python==3.6.8的虚拟环境
conda create -n py36 python=3.6.8
y
source activate py36
3.2
cd ***/caffe/python
(py36) ****@**** : ****/caffe/python$ python
Python 3.6.8 |Anaconda, Inc.| (default, Dec 30 2018, 01:22:34) [GCC 7.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import caffe
错误1：
no mudule named...
pip 安装一下
错误2：
ModuleNotFoundError: No module named 'skimage'
pip install scikit-image
错误3：
from google.protobuf.internal import enum_type_wrapper
ModuleNotFoundError: No module named 'google'
pip install google
pip install protobuf