ubuntu16.04 在cuda9.0环境下编译安装opencv2.4.13.7

た入场券 2024-04-17 14:08 13阅读 0赞

ubuntu16.04 在cuda9.0环境下编译安装opencv2.4.13.7  
安装步骤:  
1.安装cuda9.0：https://blog.csdn.net/zhuangwu116/article/details/81063234  
2.安装依赖包:

sudo apt-get install build-essential  
sudo apt-get install python-dev python-numpy libtbb2 libtbb-dev libjpeg-dev libpng-dev libtiff-dev libjasper-dev libdc1394-22-dev  
sudo apt-get install cmake git libgtk2.0-dev pkg-config

这里还需要安装解码音视频库，安装解码音视频库有两种方法:  
（1）一种是编译安装：  
（2）另一种是命令安装：

sudo apt-get install libavcodec-dev libavformat-dev libswscale-dev

如何已经编译安装解码视频库的可以跳过，不需要命令安装，以免冲突。  
3.下载代码:

git clone https://github.com/opencv/opencv.git  
cd opencv  
git checkout -b 2.4.13.7

或者你也可以直接下载压缩包 地址  
4.修改opencv编译脚本

修改opencv编译的脚本是因为两点:

1.在cuda9里面，NVIDIA把 libnppi.so换成libnppc.so libnppial.so libnppicc.so libnppicom.so libnppidei.so libnppif.so libnppig.so libnppim.so libnppist.so libnppisu.so libnppitc.so libnpps.so

2.cuda9里面不支持‘ compute-20 ’

1.解决第一个问题：  
修改opencv/cmake/FindCUDA.cmake 文件，将524行的

unset(CUDA\_nppi\_LIBRARY CACHE)

修改为

unset(CUDA\_nppial\_LIBRARY CACHE)  
unset(CUDA\_nppicc\_LIBRARY CACHE)  
unset(CUDA\_nppicom\_LIBRARY CACHE)  
unset(CUDA\_nppidei\_LIBRARY CACHE)  
unset(CUDA\_nppif\_LIBRARY CACHE)  
unset(CUDA\_nppig\_LIBRARY CACHE)  
unset(CUDA\_nppim\_LIBRARY CACHE)  
unset(CUDA\_nppist\_LIBRARY CACHE)  
unset(CUDA\_nppisu\_LIBRARY CACHE)  
unset(CUDA\_nppitc\_LIBRARY CACHE)

修改opencv/cmake/FindCUDA.cmake 文件，将799行和801行的

find\_cuda\_helper\_libs(nppi)  
set(CUDA\_npp\_LIBRARY"$\{CUDA\_nppc\_LIBRARY\};$\{CUDA\_nppi\_LIBRARY\};$\{CUDA\_npps\_LIBRARY\}")

修改为

find\_cuda\_helper\_libs(nppial)  
find\_cuda\_helper\_libs(nppicc)  
find\_cuda\_helper\_libs(nppicom)  
find\_cuda\_helper\_libs(nppidei)  
find\_cuda\_helper\_libs(nppif)  
find\_cuda\_helper\_libs(nppig)  
find\_cuda\_helper\_libs(nppim)  
find\_cuda\_helper\_libs(nppist)  
find\_cuda\_helper\_libs(nppisu)  
find\_cuda\_helper\_libs(nppitc)  
set(CUDA\_npp\_LIBRARY "$\{CUDA\_nppc\_LIBRARY\};$\{CUDA\_nppial\_LIBRARY\};$\{CUDA\_nppicc\_LIBRARY\};$\{CUDA\_nppicom\_LIBRARY\};$\{CUDA\_nppidei\_LIBRARY\};$\{CUDA\_nppif\_LIBRARY\};$\{CUDA\_nppig\_LIBRARY\};$\{CUDA\_nppim\_LIBRARY\};$\{CUDA\_nppist\_LIBRARY\};$\{CUDA\_nppisu\_LIBRARY\};$\{CUDA\_nppitc\_LIBRARY\};$\{CUDA\_npps\_LIBRARY\}")

2.解决第二个问题方案  
更改 OpenCVDetectCUDA.cmake 文件，把有关 ‘ compute-20 ’ 的修改掉。  
将74行中的

73 if(CUDA\_GENERATION STREQUAL "Fermi")  
74 set(\_\_cuda\_arch\_bin "2.0 2.1(2.0)")  
75 elseif(CUDA\_GENERATION STREQUAL "Kepler")  
76 if($\{CUDA\_VERSION\} VERSION\_LESS "5.0")  
77 set(\_\_cuda\_arch\_bin "3.0")  
78 else()  
79 set(\_\_cuda\_arch\_bin "3.0 3.5")  
80 endif()

修改成

73 if(CUDA\_GENERATION STREQUAL "Fermi")  
74 set(\_\_cuda\_arch\_bin "3.0 3.5")  
75 elseif(CUDA\_GENERATION STREQUAL "Kepler")  
76 if($\{CUDA\_VERSION\} VERSION\_LESS "5.0")  
77 set(\_\_cuda\_arch\_bin "3.0")  
78 else()  
79 set(\_\_cuda\_arch\_bin "3.0 3.5")  
80 endif()

将107行

104 if($\{CUDA\_VERSION\} VERSION\_LESS "5.0")  
105 set(\_\_cuda\_arch\_bin "1.1 1.2 1.3 2.0 2.1(2.0) 3.0")  
106 elseif($\{CUDA\_VERSION\} VERSION\_GREATER "6.5")  
107 set(\_\_cuda\_arch\_bin "2.0 2.1(2.0) 3.0 3.5")  
108 else()  
109 set(\_\_cuda\_arch\_bin "1.1 1.2 1.3 2.0 2.1(2.0) 3.0 3.5")  
110 endif()  
111 set(\_\_cuda\_arch\_ptx "3.0")  
112 endif()

修改成

104 if($\{CUDA\_VERSION\} VERSION\_LESS "5.0")  
105 set(\_\_cuda\_arch\_bin "1.1 1.2 1.3 2.0 2.1(2.0) 3.0")  
106 elseif($\{CUDA\_VERSION\} VERSION\_GREATER "6.5")  
107 set(\_\_cuda\_arch\_bin "3.0 3.5")  
108 else()  
109 set(\_\_cuda\_arch\_bin "1.1 1.2 1.3 2.0 2.1(2.0) 3.0 3.5")  
110 endif()  
111 set(\_\_cuda\_arch\_ptx "3.0")  
112 endif()

5.编译安装OpenCV

mkdir release  
cd release  
cmake -D CMAKE\_BUILD\_TYPE=RELEASE -D CMAKE\_INSTALL\_PREFIX=/usr/local ..  
make -j7 \#此处看您的cpu内核数，本人机器8核  
sudo make install

6.测试OpenCV

$ python  
>>import cv2  
>>print cv2.\_\_version\_\_

参考：https://docs.opencv.org/2.4.13.7/doc/tutorials/introduction/linux\_install/linux\_install.html\#linux-installation  
参考：http://wgshun.top/2018/06/ubuntu-install-opencv/  
————————————————  
版权声明：本文为CSDN博主「径向基」的原创文章，遵循CC 4.0 by-sa版权协议，转载请附上原文出处链接及本声明。  
原文链接：https://blog.csdn.net/zhuangwu116/article/details/81136117