python 35 使用 3D densecrf （DenseInferenceWrapper）

If you want to use this in python3 then you can follow the following steps

Requirements

make, g++, boost-python3.5

Installation of Boost for Python3

Make sure you have the libboost-python-dev libraries installed: sudo apt-get install libboost-python-dev

#####Install python 3.5 :

Make sure you have the libboost-python-dev libraries installed:

#####Download the 1.58.0 不要最新的会有问题！！ of Boost from http://www.boost.org/

#####Run bootstrap with correct flags :

#####Compile Boost in directory :

#####Update the Makefile to take the new boost locations (Already did)

Boost.python的编译安装

• 说明：因为caffe需要boost的支持，在卸载以后所有boost文件都需要通过安装(对于非boost.python来说，安装就是将头文件拷贝到PREFIX路径[默认为/usr/local])，不只boost.python(boost.python不是headers-only文件，需要单独build)

step1：源码文件的下载和解压

首先，按照boost官网下载指引描述，到所示网站下载boost源码的压缩文件，当前时间的文件是boost_1_66_0.tar.bz2，放在$HOME目录下。

~$ tar --bzip2 -xvf boost_1_66_0.tar.bz2
~$ cd boost_1_66_0
~/boost_1_66_0$ ./bootstrap.sh

 

#返回信息如下
Building Boost.Build engine with toolset gcc... tools/build/src/engine/bin.linuxx86_64/b2
Detecting Python version... 2.7
Detecting Python root... /usr
Unicode/ICU support for Boost.Regex?... not found.
Backing up existing Boost.Build configuration in project-config.jam.1
Generating Boost.Build configuration in project-config.jam... #配置文件名为project-config.jam

Bootstrapping is done. To build, run:

    ./b2 #通常的下一步操作

To adjust configuration, edit 'project-config.jam'.
Further information:

   - Command line help:
     ./b2 --help #如果需要更详细的说明，运行左边的命令后再根据需要修改自己的命令

   - Getting started guide: 
     http://www.boost.org/more/getting_started/unix-variants.html

   - Boost.Build documentation:
     http://www.boost.org/build/doc/html/index.html

我们应关注上述返回信息，包括配置文件名和一些使用提示。

step2:重新编译boost

这一步是为了保险起见，将所有boost相关文件都重新编译，运行一下命令：

~/boost_1_66_0$ ./b2 -a

关注shell返回信息中，建议将路径1加入#include path，将路径2加入#compiler path，以下根据指引设置环境变量。

step3:设置环境变量

~$ sudo gedit ~/.profile
#根据上一步shell中返回的信息，在打开的文件中加入如下两行信息
export BOOST_INCLUDE=$HOME/boost_1_66_0
export BOOST_LIB=$HOME/boost_1_66_0/stage/lib
#保存后关闭文件
~$source ~/.profile

 

step4:验证安装boost成功

对于我来说，因为caffe是需要boost支持的，验证方式是在caffe文件夹下重新编译Makefile.config，依次make all,make test,make runtest直至最后测试完全通过。 
当然简单的方式，应该是调用库，运行一小段代码呗。

~$ touch sample.cpp
~$ sudo gedit sample.cpp

向打开的文件中填入以下代码

#include <string>
#include <iostream>
#include <boost/version.hpp>
#include <boost/timer.hpp>
using namespace std;
int main()
{
    boost::timer t;
    cout << "max timespan: " << t.elapsed_max() / 3600 << "h" << endl;
    cout << "min timespan: " << t.elapsed_min() << "s" << endl;
    cout << "now time elapsed: " << t.elapsed() << "s" << endl;
    cout << "boost version" << BOOST_VERSION <<endl;
    cout << "boost lib version" << BOOST_LIB_VERSION <<endl;
    return 0;
}

~$ g++ sample.cpp -o sample.out #编译生成sample.out文件
~$ ./sample.out #查看输出

得到以下输出

max timespan: 2.56205e+09h



min timespan: 1e-06s
now time elapsed: 0.000135s
boost version105800#好吧，真的是没有卸载干净....
boost lib version1_58

附上makefile

附上makefile
# update the path variables

CC	= g++
CFLAGS	= -W -Wall -O2 -DNDEBUG
#CFLAGS	= -W -Wall -g
LIB_ROOT = denseinference/lib/

all:
	make -C clean
	make -C python

clean:
	rm -f *.a
	rm -f *.o
	rm -f *.so

libDenseCRF.a: libDenseCRF/bipartitedensecrf.cpp libDenseCRF/densecrf.cpp libDenseCRF/filter.cpp libDenseCRF/permutohedral.cpp libDenseCRF/util.cpp libDenseCRF/densecrf.h libDenseCRF/fastmath.h libDenseCRF/permutohedral.h libDenseCRF/sse_defs.h libDenseCRF/util.h
	$(CC) -fPIC libDenseCRF/bipartitedensecrf.cpp libDenseCRF/densecrf.cpp libDenseCRF/filter.cpp libDenseCRF/permutohedral.cpp libDenseCRF/util.cpp -c $(CFLAGS) -DNDEBUG
	ar rcs libDenseCRF.a bipartitedensecrf.o densecrf.o filter.o permutohedral.o util.o

dense_inference.so: dense_inference.o libDenseCRF.a
	$(CC) -shared -I/usr/include/python2.7 -Wl,-no-undefined -o dense_inference.so dense_inference.o -lpython2.7 -lboost_python -L. -lDenseCRF $(CFLAGS)

dense_inference.o: refine_3d/dense_inference.cpp refine_3d/dense_inference.h libDenseCRF.a
	$(CC) -c -fPIC -I/usr/include/python2.7 refine_3d/dense_inference.cpp -o dense_inference.o $(CFLAGS) -lDenseCRF $(CFLAGS)


# For python3 use the following \
dense_inference.so: dense_inference.o libDenseCRF.a \
	$(CC) -shared -I/usr/include/python3.5m -Wl,-no-undefined -o dense_inference.so dense_inference.o -lboost_python-py35 -lpython3.5m -L. -lDenseCRF $(CFLAGS) \
dense_inference.o: refine_3d/dense_inference.cpp refine_3d/dense_inference.h libDenseCRF.a \
	$(CC) -c -fPIC -I/usr/include/python3.5m refine_3d/dense_inference.cpp -o dense_inference.o $(CFLAGS) -lDenseCRF $(CFLAGS)

python: dense_inference.o dense_inference.so libDenseCRF.a

.PHONY: default clean

中间还遇到过一个问题，就是numpy.hpp不存在，这个问题是因为虽然pip 安装了numpy，但是我是anaconda的python，所以系统里没有numpy，需要用apt-get install  python3-numpy 安装