环境要求

Windows 11 22H2
cuda 12.2.0
cuDNN 8.9.5
vs 2022 以及 c++桌面开发环境
python 3.9
opencv 4.8.0

安装conda环境

# 安装包
https://repo.anaconda.com/archive/Anaconda3-2023.09-0-Windows-x86_64.exe
安装地址（可选）#
D:\Program\anaconda3
安装配置（勾选）#
1.Create start menu shortcuts (supported packages only).
2.Register Anaconda3 as my default Python 3.X
3.Clear the package cache upon completion
注册conda（可选）#
conda init powershell

常用conda命令

# 创建名为opencv的Python版本为3.9的环境
conda create -n op python=3.9
查看Anaconda下管理的所有Python环境#
conda info -e
激活pytorch#
conda activate op
退出虚拟环境#
conda deactivate
删除opencv环境：#
conda remove —name op —all

配置Cmake环境

# https://cmake.org/
下载gui版&安装#
https://github.com/Kitware/CMake/releases/download/v3.28.0-rc3/cmake-3.28.0-rc3-windows-x86_64.msi
新建文件夹#
mkdir opencv
cd opencv
mkdir build
下载opencv#
https://github.com/opencv/opencv/archive/refs/tags/4.8.0.zip
解压至当前目录后clone opencv-contrib#
git clone https://github.com/opencv/opencv_contrib.git
完成后的目录结构#
|—opencv

|—opencv\build
|—opencv\opencv-4.8.0
|—opencv\opencv_contrib-4.8.0
打开camke#
source_path = ..\opencv\opencv-4.8.0
build_path = ..\opencv\opencv_contrib-4.8.0
Add Entry 添加Python环境#
PYTHON3_EXECUTABLE

FILEPATH

”PATH TO YOUR PYTHON.EXE WHERE IN ANACONDA ENVIREMENT”
第一次点击configure后，如果没有error报错，则
0.弹出的VS配置选项-VS17-x64
1.勾选BUILD_opencv_world、WITH_CUDA、OPENCV_DNN_CUDA
2.设置OPENCV_EXTRA_MODULES_PATH为\opencv\opencv_contrib-4.8.0\modules
第二次点击configure，如果没有报错，则
1.勾选FAST_MATH字样选项，去除所有TEST、Java字样选项
2.按需求勾选ENABLE-NOFREE、DNN-PERF、BUILD-XX等
3.检查是否勾选opencv-python3
第三次点击configure，如果没有报错，则点击Generate
若没有报错，进入编译阶段

VS编译

1.点击Open Project
2.设置编译平台（Release/Debug均可）和 x64, 右键ALL_BUILD-生成，生成时间较长
3.生成成功后，右键INSTALL-仅用于项目-仅生成INSTALL

配置modules

在编译输出文件夹lib\python3\Release下的
\cv2.cp39-win_amd64.pyd复制到anaconda3\envs\op\Lib\site-packages\cv2\python-3.9
\bin\Release目录下所有文件复制到anaconda3\envs\op\Lib\site-packages

测试程序

import cv2
cv2.cuda.printCudaDeviceInfo(0)