Skip to content

Latest commit

 

History

History
174 lines (143 loc) · 5.7 KB

OpenCV_Installation.md

File metadata and controls

174 lines (143 loc) · 5.7 KB
Raw

NOTE: THIS INSTRUCTION TARGETS AT OPENCV < 4.5.0

FOR VER. >= 4.5.0, REFER TO HERE: https://docs.opencv.org/4.5.0/d7/d9f/tutorial_linux_install.html

REF #1: https://docs.opencv.org/4.4.0/d2/de6/tutorial_py_setup_in_ubuntu.html

REF #2: https://docs.opencv.org/4.4.0/d7/d9f/tutorial_linux_install.html

Steps for installing opencv

NOTE: Setup all pip2/3 & venv (depending on your choice of python environment for build & install opencv) beforehand

OPTION 1: Install using pre-built binary

sudo apt-get install libopencv-dev python(3)-opencv

OPTION 2: Build from source (python3 components included)

  1. Install OS libs & dependencies
sudo apt update
sudo apt-get dist-upgrade -> may not be necessary

sudo apt install cmake git gcc g++ build-essential unzip pkg-config

sudo apt install python(3)-dev python(3)-numpy
# NOTE: python(3)-numpy is unnecessary when module numpy was already installed using pip(3) or activated python(3) environment has it already

sudo apt install libavcodec-dev libavformat-dev libswscale-dev libv4l-dev
sudo apt install libgstreamer-plugins-base1.0-dev libgstreamer1.0-dev

# to support gtk2:
sudo apt install libgtk2.0-dev

# to support gtk3:
sudo apt install libgtk-3-dev

# Optional:
sudo apt install libtbb2 libtbb-dev \
                     libdc1394-22 libdc1394-22-dev \
                     libjpeg-dev libpng-dev libtiff-dev libopenexr-dev libwebp-dev

# Not exist in official tutorial but in various other pages:
sudo apt install libxvidcore-dev libx264-dev \
                     libatlas-base-dev gfortran

# For Ubuntu 16.04, add libjasper-dev to add a system level support for the JPEG2000 format:
sudo add-apt-repository "deb http://security.ubuntu.com/ubuntu xenial-security main"
sudo apt install libjasper-dev
  1. Get opencv and opencv_contrib git
git clone -b <version> --depth 1 https://github.com/opencv/opencv.git

or get a specific version of opencv at:

https://opencv.org/releases/
or
https://github.com/opencv/opencv/tags

git clone -b <version> --depth 1 https://github.com/opencv/opencv_contrib.git

NOTE: version of opencv & opencv_contrib MUST MATCH!!!

  1. Activate venv & install numpy using pip (only when user wish to install using venv python)
pip install numpy Matplotlib IPython
  1. Configure OpenCV with CMake
cd ~/opencv
# [optional] git checkout <version> (ex: 4.4.0)
mkdir build && cd build
cmake -D CMAKE_BUILD_TYPE=Release \
    -D CMAKE_INSTALL_PREFIX=/usr/local \
    -D WITH_CUDA=OFF \
    -D WITH_CUDNN=ON \
    -D OPENCV_DNN_CUDA=ON \
    -D ENABLE_FAST_MATH=1 \
    -D CUDA_FAST_MATH=1 \
    -D CUDA_ARCH_BIN=3.0 \
    -D WITH_CUBLAS=1 \
    -D INSTALL_PYTHON_EXAMPLES=ON \
    -D OPENCV_EXTRA_MODULES_PATH=<path to opencv_contrib dir>/modules \
    -D OPENCV_ENABLE_NONFREE=ON \
    -D BUILD_EXAMPLES=OFF \
    -D BUILD_NEW_PYTHON_SUPPORT=ON \
    -D BUILD_opencv_python3=ON \
    -D HAVE_opencv_python3=ON \
    -D PYTHON2(3)_EXECUTABLE=<path to python> \
    -D PYTHON_INCLUDE_DIR=/usr/include/python<version> \
    -D PYTHON_INCLUDE_DIR2=/usr/include/x86_64-linux-gnu/python<version> \
    -D PYTHON_LIBRARY=/usr/lib/x86_64-linux-gnu/libpython<version>.so \
    -D PYTHON2(3)_NUMPY_INCLUDE_DIRS=/usr/lib/python<version>/dist-packages/numpy/core/include/
    -D PYTHON_DEFAULT_EXECUTABLE=/usr/bin/python3 \
    -D PYTHON3_PACKAGES_PATH=<virtualenv dir>/lib/python3.6/site-packages/ \
    -D OPENCV_GENERATE_PKGCONFIG=ON \
    ..
NOTE:   Edit `PYTHON_EXECUTABLE` depend on venv setup
        `CMAKE_INSTALL_PREFIX` may be unnecessary (default: /usr -> system scope | /usr/local -> user scope)
        `PYTHON_DEFAULT_EXECUTABLE` may be unnecessary
        `PYTHON(3)_EXECUTABLE` could be <virtualenv dir>/bin/python(3) or /usr/bin/python(3) depend on whether you use venv or not
        `CUDA_ARCH_BIN` must match your nvidia gpu arch - find here: https://developer.nvidia.com/cuda-gpus
        Use "cmake -LA" or use "ccmake ." to view & modify full options list
        Or remove CMakeCache.txt & re-run cmake
        Or manually edit CMakeCache.txt & CMakeVars.txt

Example for using python3 - local scope & system python (not venv):

cmake -D CMAKE_BUILD_TYPE=Release \
    -D CMAKE_INSTALL_PREFIX=/usr/local \
    -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=3.0 \
    -D WITH_CUBLAS=1 \
    -D INSTALL_PYTHON_EXAMPLES=ON \
    -D OPENCV_EXTRA_MODULES_PATH=<path to opencv_contrib dir>/modules \
    -D OPENCV_ENABLE_NONFREE=ON \
    -D BUILD_EXAMPLES=ON \
    -D BUILD_NEW_PYTHON_SUPPORT=ON \
    -D BUILD_opencv_python3=ON \
    -D HAVE_opencv_python3=ON \
    -D PYTHON_INCLUDE_DIR=/usr/include/python3.6 \
    -D PYTHON_INCLUDE_DIR2=/usr/include/x86_64-linux-gnu/python3.6m \
    -D PYTHON_LIBRARY=/usr/lib/x86_64-linux-gnu/libpython3.6m.so \
    -D PYTHON3_NUMPY_INCLUDE_DIRS=/usr/local/lib/python3/dist-packages/numpy/core/include \
    -D PYTHON3_PACKAGES_PATH=$HOME/.local/lib/python3/site-packages \
    -D OPENCV_GENERATE_PKGCONFIG=ON \
    ..

For install cv2 to system scope python3, use PYTHON3_PACKAGES_PATH=/usr/local/lib/python3/dist-packages

  1. Build opencv
make -j6 # j: # of parallel threads
  1. Installing and verifying OpenCV
sudo make install
sudo ldconfig
  1. Done

Test import cv2 in python or run cmd:

opencv_version
  1. Uninstall
cd <path to opencv>/build
sudo make uninstall

Hotfix if import cv2 failed:

Create symlink to cv2.cpython-36m-x86_64-linux-gnu.so

cd <virtualenv dir>/lib/python3.6/site-packages
ln -s /usr/local/lib/python3.6/site-packages/cv2/python-3.6/cv2.cpython-36m-x86_64-linux-gnu.so cv2.so