drsfting...

Laygond January 07, 2023

Installation Steps

I could not find anything about installing OpenCV without using the Visual Studio compiler, so I decided to share the way I did it using MinGW gcc compiler.

  1. Create a directory at C:\ , e.g., C:\opencv
    (this is where the opencv-[version_number] and opencv_contrib-[version_number] source files will be extracted).

  2. Go to opencv github and download from the master branch the latest tag. (It will be downloaded as opencv-[version_number])
      

  3. Go to opencv contrib and download the same master and tag version as in the previous step

  4. Extract the two source files at the directory created in step 1 (C:\opencv in my case) and create an empty folder named build**.** This is how it should look so far

  5. Download the MinGW→ gcc compiler by following step 3 to 6 of the following link: Get Started with C++ and Mingw-w64 in Visual Studio Code. If you would like to use visual studio code as your editor, then also follow step 1 and 2 from that same link.

  6. Download the cmake binary distribution for windows at **Download | CMake

  7. That’s all we need! It is time to generate those build files, compile it, and execute it.

    IMPORTANT: Make sure in your "edit environmental variables” that only `C:\msys64\mingw64\bin` is set in path **and not** `C:\msys64\usr\bin` (as in image from step 10). There are several make commands and that other path contains the ‘msys make’ and you dont want to combine make commands during compilation( we only need the ‘mingw make’).

    To     generate those build files open Cmake GUI and set your source and build directories and then click `Configure` with Mingw. This will scan the source code for a `CmakeList.txt` and set the parameters to later create a MakeFile (next step) inside the build folder. This MakeFile file contains all the instructions for compilation.  



  8. Before we click generate. This is the part where we can set everything: CUDA, Cudnn, math libraries, GPU, etc. We are only setting opencv contrib but if you want to set GPU configurartion the info is specified below. Now search for “extra” and set the path to the contrib modules like in image. Finally click Configure to update and then Generate. 

GPU CONFIG 
Search and set the following:
WITH_CUDA = ON 
OPENCV_DNN_CUDA=ON
CUDA_STUBS=ON
ENABLE_FAST_MATH=1
BUILD_opencv_dnn =ON
BUILD_opencv_world =ON
Click Configure to update/unlock new config parameters (Then continue)
CUDA_FAST_MATH=ON
CUDA_ARCH_BIN=7.5 # Check yours according to your GPU Here!!!! 
CMAKE_BUILD_TYPE=RELEASE
CMAKE_CONFIGURATION_TYPES=RELEASE
Click Configure to update/unlock new config parameters (Then continue)
BUILD_EXAMPLES = OFF  # OPTIONAL
INSTALL_PYTHON_EXAMPLES=OFF # OPTIONAL
INSTALL_C_EXAMPLES=OFF # OPTIONAL
WITH_CUBLAS=1
WITH_CUDNN = ON
OPENCV_ENABLE_NONFREE=ON 
WITH_TBB=ON     # For Thread blocking
CMAKE_INSTALL_PREFIX=C:\opencv\build\install
All configs can be found here

  1. Time to compile and run! Open a terminal, cd inside your build file where now is full of files and most importantly lies your MakeFile. 

     > RULE of THUMB: Always match cmake makefile generation with make command. 
 > cmake mingw makefiles → mingw32-make
 > cmake msys makefile → msys make
 > cmake visual studio makefiles → vc make

    At this directory level type in your terminal wingw32-make and enjoy! (It will take a while: maybe hours). Finally run the executable.

  2. Final step is to set the opencv environmental variables so that you can later use it in your code. It should like this:
      

OpenCV Test Example! Read an Image

Create a directory say opencvtest and under your directory place the following:

img.png
main.cpp
CMakeLists.txt

main.cpp

NOTE: Use double backslash & single forward for path.

#include <stdio.h>
#include <opencv2/opencv.hpp>
using namespace cv;
int main(int argc, char** argv )
{
    Mat image;
    image = imread("C:/path_to_image/img.png");
    if ( !image.data )
    {
        printf("No image data \n");
        return -1;
    }
    namedWindow("Display Image", WINDOW_AUTOSIZE );
    imshow("Display Image", image);
    waitKey(0);
    return 0;
}

CMakeLists.txt

NOTE: Use as project name same as that of directory you created. In my case opencvtest

cmake_minimum_required(VERSION 3.0.0)
project(opencvtest VERSION 0.1.0)
include(CTest)
enable_testing()
find_package( OpenCV REQUIRED )
include_directories( ${OpenCV_INCLUDE_DIRS} )
add_executable(opencvtest main.cpp)
target_link_libraries( opencvtest ${OpenCV_LIBS} )
set(CPACK_PROJECT_NAME ${PROJECT_NAME})
set(CPACK_PROJECT_VERSION ${PROJECT_VERSION})
include(CPack)

Compile & Run

NOTE: opencvtest is this project’s main directory

cd opencvtest
mkdir build
cd build
cmake -G "MinGW Makefiles" ..   #Create makefiles using MinGW with instructions at ..
mingw32-make     # create executable
opencvtest.exe   # RUN


REFERENCES