WIN10 VS2015 Cmake编译 opencv3.3 cuda9.0

最新推荐文章于 2024-08-08 10:25:00 发布

Tsang-hx

最新推荐文章于 2024-08-08 10:25:00 发布

阅读量3.6k

点赞数 3

文章标签： cuda cmake win10 opencv

本文链接：https://blog.csdn.net/weixin_38151188/article/details/78700800

版权

WIN10 VS2015 Cmake编译 opencv3.3 cuda9.0

由于opencv 和cuda版本不匹配在Windows端经常出现安装失败，或者with_cuda_on 但在实际代码中缺报错 The library is compiled without CUDA support 。这其实都是版本不同导致的结果。

目前cuda9.0和opencv3.3可以匹配 cuda8.0和opencv3.2可以匹配；

本篇讲讲述在win10系统中利用cmake3.9.6 编译opencv3.3 with cuda9.0 测试有效

确保自己的win10 系统可以正常安装CUDA9.0 ，测试是否正常安装可以在cmd下
这里写图片描述

之后安装cmake 3.9.6
这里写图片描述

下载opencv3.3.0 exe版本并且安装到一个目录下
这里写图片描述

在sources -> cmake 文件夹中找到FindCUDA.cmake文件
找到行
find_cuda_helper_libs(nppi)
更改为
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)

找到行
set(CUDA_npp_LIBRARY “ ${CUDA_nppc_LIBRARY};$ {CUDA_nppi_LIBRARY}; ${CUDA_npps_LIBRARY}”) 更改为 set(CUDA_npp_LIBRARY “$ {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}”)

找到行
unset(CUDA_nppi_LIBRARY CACHE)
更改为
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)

之后找到文件OpenCVDetectCUDA.cmake
在文件中找到
set(__cuda_arch_ptx “”)
if(CUDA_GENERATION STREQUAL “Fermi”)
set(__cuda_arch_bin “2.0”)
elseif(CUDA_GENERATION STREQUAL “Kepler”)
set(__cuda_arch_bin “3.0 3.5 3.7”)
更改为
set(__cuda_arch_ptx “”)
if(CUDA_GENERATION STREQUAL “Kepler”)
set(__cuda_arch_bin “3.0 3.5 3.7”)
elseif(CUDA_GENERATION STREQUAL “Maxwell”)
set(__cuda_arch_bin “5.0 5.2”)

最后一步在头文件中加入 #include “cuda_fp16.h”
添加至 opencv\modules\cudev\include\opencv2\cudev\common.hpp

其次由于在编译过程中会下载几个文件但是文件的服务器在国外所以下载大概率会失败所以下面给出这几个文件的下载办法
前往https://github.com/opencv/opencv_3rdparty/branches/all
下什么版本可以在
D:\opencv3.3.0\opencv\sources\3rdparty\ffmpeg
D:\opencv3.3.0\opencv\sources\3rdparty\ippicv 里面的 ffmpeg或ippicv.cmake文件中查看到
比如说
这里写图片描述
里面第一行为对应的版本

下面开始拷贝下载文件
将下载目录中的 ffmpeg_version.cmake、opencv_ffmpeg.dll、opencv_ffmpeg_64.dll 三项复制到 “path to your OpenCV\opencv\sources\3rdparty\ffmpeg” 目录下。

编辑 ffmpeg.cmake
在 “path to your OpenCV\opencv\sources\3rdparty\ffmpeg” 中打开文件 ffmpeg.cmake ，将其内容修改为：
message(STATUS “FFMPEG: Package successfully downloaded”)
include(${CMAKE_CURRENT_LIST_DIR}/ffmpeg_version.cmake)

保存完毕

在前面下载解压的目录“..\opencv_3rdparty-ippicv-master_20170822\ippicv” 中将ippicv_windows_20170822.zip 解压到当前目录。
在 “path to your OpenCV \opencv\sources\3rdparty\ippicv” 中创建文件夹unpack，将解压得到的ippicv_win 复制到新创建的unpack 中。
编辑 downloader.cmake
将downloader.cmake 的内容修改为：

The script downloads ICV package

On return this will define:

OPENCV_ICV_PATH - path to unpacked downloaded package

function(_icv_downloader)
# Commit SHA in the opencv_3rdparty repo
set(IPPICV_BINARIES_COMMIT “81a676001ca8075ada498583e4166079e5744668”)
# Define actual ICV versions
if(APPLE)
set(OPENCV_ICV_PACKAGE_NAME “ippicv_macosx_20151201.tgz”)
set(OPENCV_ICV_PACKAGE_HASH “4ff1fde9a7cfdfe7250bfcd8334e0f2f”)
set(OPENCV_ICV_PLATFORM “macosx”)
set(OPENCV_ICV_PACKAGE_SUBDIR “/ippicv_osx”)
elseif(UNIX)
if(ANDROID AND NOT (ANDROID_ABI STREQUAL x86 OR ANDROID_ABI STREQUAL x86_64))
return()
endif()
set(OPENCV_ICV_PACKAGE_NAME “ippicv_linux_20151201.tgz”)
set(OPENCV_ICV_PACKAGE_HASH “808b791a6eac9ed78d32a7666804320e”)
set(OPENCV_ICV_PLATFORM “linux”)
set(OPENCV_ICV_PACKAGE_SUBDIR “/ippicv_lnx”)
elseif(WIN32 AND NOT ARM)
set(OPENCV_ICV_PACKAGE_NAME “ippicv_windows_20151201.zip”)
set(OPENCV_ICV_PACKAGE_HASH “04e81ce5d0e329c3fbc606ae32cad44d”)
set(OPENCV_ICV_PLATFORM “windows”)
set(OPENCV_ICV_PACKAGE_SUBDIR “/ippicv_win”)
else()
return() # Not supported
endif()