7. OpenCV

7.1. OpenCV Installation

  1. clone both opencv and opencv-contrib into local directories

cd opencv_contrib
git checkout <version>
cd ../opencv
git checout <version>

mkdir build
cd build

  1. Setup cmake command:

cmake -DCMAKE_BUILD_TYPE=Release \
-DOPENCV_EXTRA_MODULES_PATH=../../opencv_contrib/modules \
-DCMAKE_BUILD_TYPE=RELEASE \
-DCMAKE_INSTALL_PREFIX=/usr/local \..

OpenCV with Cuda:

cmake -DCMAKE_BUILD_TYPE=Release \
-DOPENCV_EXTRA_MODULES_PATH=../../opencv_contrib/modules \
-DCUDA_ARCH_BIN=7.0 \
-DCUDA_ARCH_PTX=7.0 \
-DCMAKE_BUILD_TYPE=RELEASE \
-DCMAKE_INSTALL_PREFIX=/usr/local \
-DWITH_TBB=ON \
-DBUILD_NEW_PYTHON_SUPPORT=ON \
-DWITH_V4L=ON \
-DBUILD_TIFF=ON \
-DWITH_QT=ON \
-DWITH_OPENGL=ON \
-DENABLE_FAST_MATH=1 \
-DCUDA_FAST_MATH=1 \
-DWITH_CUBLAS=1 \
-DINSTALL_PYTHON_EXAMPLES=ON \
-DBUILD_EXAMPLES=ON \
-DCUDA_GENERATION=Volta \
-DCUDA_NVCC_FLAGS="-D_FORCE_INLINES" ..

The alternative is to install a docker contrainer with opencv with cuda/cudnn installed, and then use Jupyterlab to remote into it.

  1. Install

make install -j8
sudo make install