Skip to content

Build opencv from source with cuda support

Using docker to build opencv from source with cuda support for python and cpp. pack the result as debs file to install on other machines.

Check machine CUDA version

1
nvidia-smi

The docker base on the host CUDA version, I use pre-built docker image from NVIDIA, FROM nvidia/cuda:${CUDA}-cudnn-devel-ubuntu${UBUNTU},

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
ARG CUDA="12.6.0"
ARG UBUNTU="22.04"

# docker pull nvidia/cuda:12.6.0-cudnn-devel-ubuntu22.04
FROM nvidia/cuda:${CUDA}-cudnn-devel-ubuntu${UBUNTU}

RUN apt update && apt install -y --no-install-recommends \
    build-essential \
    cmake \
    cmake-data \
    gcc \
    g++ \
    ninja-build \
    gdb \
    git \
    wget \
    curl \
    unzip \
    yasm \
    doxygen \
    pkg-config \
    checkinstall \
    build-essential cmake git libgtk2.0-dev pkg-config libavcodec-dev libavformat-dev \
    python3-dev python3-numpy \
    libtbb2 libtbb-dev libjpeg-dev libpng-dev libtiff-dev \
    libswscale-dev libgtk-3-dev libv4l-dev v4l-utils\
    libopenblas-dev libopenblas-base libatlas-base-dev liblapacke-dev \
    protobuf-compiler python3-dev python3-venv python3-numpy python3-wheel python3-setuptools \
&& rm -rf /var/lib/apt/lists/*

Build process

  • Download and extract opencv and opencv_contrib
1
2
3
4
5
6
7
8
9
version="4.10.0"
folder="opencv"
mkdir $folder
cd ${folder}
curl -L https://github.com/opencv/opencv/archive/${version}.zip -o opencv-${version}.zip
curl -L https://github.com/opencv/opencv_contrib/archive/${version}.zip -o opencv_contrib-${version}.zip
unzip opencv-${version}.zip
unzip opencv_contrib-${version}.zip
rm opencv-${version}.zip opencv_contrib-${version}.zip
  • Create build folder
  • Run cmake
  • Build
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
mkdir opencv-${version}/build
cd opencv-${version}/build

cmake \
-D WITH_CUDA=ON \
-D WITH_CUDNN=ON \
-D OPENCV_DNN_CUDA=ON \
-D CUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-12.6 \
-D CUDA_ARCH_BIN="5.2 5.3 6.0 6.1 6.2 7.0 7.2 7.5 8.0 8.6" \
-D CUDA_ARCH_PTX="8.6rm -rf" \
-D ENABLE_FAST_MATH=1 \
-D CUDA_FAST_MATH=1 \
-D OPENCV_GENERATE_PKGCONFIG=ON \
-D OPENCV_EXTRA_MODULES_PATH=../../opencv_contrib-4.10.0/modules \
-D WITH_GSTREAMER=OFF \
-D WITH_LIBV4L=ON \
-D BUILD_opencv_python3=ON \
-D BUILD_TESTS=OFF \
-D BUILD_PERF_TESTS=OFF \
-D BUILD_EXAMPLES=OFF \
-D CMAKE_BUILD_TYPE=RELEASE \
-D CMAKE_INSTALL_PREFIX=/usr/local \
-D WITH_QT=ON \
-D WITH_GTK=ON \
-D WITH_VTK=OFF \
-D WITH_FFMPEG=OFF \
-D WITH_1394=OFF \
-D CPACK_BINARY_DEB=ON \
-D BUILD_JAVA=OFF \
-D PYTHON3_PACKAGES_PATH=/usr/lib/python3/dist-packages \
-D CPACK_PACKAGE_VERSION=4.10.0 \
-D CPACK_DEBIAN_PACKAGE_VERSION=4.10.0-1 \
..

# Build
make -j8

pack to debs

pack
1
make package ..

Downloads

Install and check

install
1
dpkg -i *.deb

check

check cuda support
1
2
3
4
5
>>> import cv2
>>> cv2.__version__
'4.10.0'
>>> cv2.cuda.getCudaEnabledDeviceCount()
1
simple.py
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
import cv2

# Load an image from file
image = cv2.imread("/workspaces/opencv_builder/src/lena.png")

# Check if the image was loaded successfully
if image is None:
    print("Failed to load image")
else:
    # Display the image in a window
    cv2.imshow("My Image", image)

    # Wait for any key press
    cv2.waitKey(0)

    # Close all OpenCV windows
    cv2.destroyAllWindows()

Download lena img

simple.cpp
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
// g++ simple.cpp -o simple `pkg-config --cflags --libs opencv4` -I/usr/include/opencv4/

#include <opencv2/opencv.hpp>
#include <iostream>

int main()
{
    // Load the image
    cv::Mat image = cv::imread("/workspaces/opencv_builder/src/lena.png");

    // Check if image loaded successfully
    if (image.empty()) {
        std::cerr << "Failed to load image" << std::endl;
        return 1;
    }

    // Show the image in a window
    cv::imshow("My Image", image);

    // Wait for a key press
    cv::waitKey(0);

    // Close all windows
    cv::destroyAllWindows();

    return 0;
}
build
1
g++ simple.cpp -o simple `pkg-config --cflags --libs opencv4` -I/usr/include/opencv4/

include search path

  • /usr/include/opencv4/ is the default include path for opencv4, if you use other version, please check the path.
  • pkg-config --cflags --libs opencv4 will return the include path and lib path for opencv4, you can use it in your build command.
  • using pkg-config the include path point -I/usr/local/include/opencv4
  • After the installation from debs the include path is /usr/include/opencv4/