For some reasons, I have to install OpenCV from source in my Xubuntu 20.04. To me, the tricky part was the options for CMake. Here I share my settings and the reason why I need them.

Install required packages

Source: How to Install OpenCV on Ubuntu 20.04

sudo apt install build-essential cmake git pkg-config libgtk-3-dev \
    libavcodec-dev libavformat-dev libswscale-dev libv4l-dev \
    libxvidcore-dev libx264-dev libjpeg-dev libpng-dev libtiff-dev \
    gfortran openexr libatlas-base-dev python3-dev python3-numpy \
    libtbb2 libtbb-dev libdc1394-22-dev libopenexr-dev \
    libgstreamer-plugins-base1.0-dev libgstreamer1.0-dev

Download the source

There are two sources I downloaded for installation. One is of course the opencv and another is the opencv_contrib. The later one is essential for some modules such as the DNN one for deep learning neural networks. Both of them could be found under OpenCV GitHub repo.

As the moment of this post, the latest version was 4.5.5. Download them ( opencv-4.5.5.tar.gz and opencv_contrib-4.5.5.tar.gz) and extract them in the same folder.

Installation

Find GPU information

If you need to run OpenCV with your CUDA-enabled GPU, follow the next two steps to get the necessary information.

  1. Use nvidia-smi -L to list your GPU name.
  2. Go to the page of GPU Compute Capability to find the GPU compute capability, which is the value of CUDA_ARCH_BIN in the cmake options.

(What if I have multiple GPUs and they have different CUDA_ARCH_BIN values? Sorry I have no such cool resources and don’t know how to set the option…)

Some CMake options

  • The options of CUDA or CUDNN should be obvious. If you just want to use OpenCV with CPU, then set them to OFF.
  • The OPENCV_GENERATE_PKGCONFIG is used to generate opencv4.pc file, which might be needed for some other packages. In my own case, I need it to install darknet for YOLO.
  • OPENCV_EXTRA_MODULES_PATH points to the opencv_contrib folder.
  • NOTE 1: For OpenCV 4.5.5, we need to set -D PYTHON3_PACKAGES_PATH to dist-packages as listed in the following section, or the Python packages will be installed to site-packages and we cannot import cv2 properly. (ref: Opencv issue #21359)
  • NOTE 2: You may need to install libcudnn8 and libcudnn8-dev if you want to build OpenCV with the dnn module (with -D OPENCV_DNN_CUDA=ON in cmake).

Install OpenCV

The steps are as follows.

  • cd to the downloaded and extracted opencv-4.5.5 folder.

  • make build && cd build

  • Run the following command

    cmake -D CMAKE_BUILD_TYPE=RELEASE \
    -D OPENCV_GENERATE_PKGCONFIG=ON \
    -D CMAKE_INSTALL_PREFIX=/usr/local \
    -D INSTALL_PYTHON_EXAMPLES=ON \
    -D INSTALL_C_EXAMPLES=OFF \
    -D OPENCV_ENABLE_NONFREE=ON \
    -D WITH_CUDA=ON \
    -D WITH_CUDNN=ON \
    -D OPENCV_DNN_CUDA=ON \
    -D ENABLE_FAST_MATH=1 \
    -D CUDA_FAST_MATH=1 \
    -D CUDA_ARCH_BIN=6.1 \
    -D WITH_CUBLAS=1 \
    -D OPENCV_EXTRA_MODULES_PATH=/home/user/opencv_contrib-4.5.1/modules/\
    -D HAVE_opencv_python3=ON \
    -D PYTHON3_PACKAGES_PATH=/usr/lib/python3/dist-packages \
    -D BUILD_EXAMPLES=ON ..

  • make -j$(nproc) (nproc returns the available number of your processors)

  • sudo make install

  • sudo ldconfig

Check the installation

  • We can check the installation by pkg-config --modversion opencv4. The terminal will show the version number (4.5.5 in this case).

  • In Python:

    $ python3
Python 3.8.10 (default, Nov 26 2021, 20:14:08)
[GCC 9.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import cv2
>>> cv2.__version__
'4.5.5

  • In C++, use the following code:

    #include <opencv2/opencv.hpp>
#include <iostream>
  
int main() {
    std::cout << "OpenCV version: " << cv::getBuildInformation().c_str() << "\n";
  
    return 0;
}

    and compile it using

    g++ cv_build_info.cpp -o cv_build_info `pkg-config --cflags --libs opencv4`

    Run ./cv_build_info to view all the build information including the installed version.