Compiling and installing OpenCV DNN with CUDA (With Python Interface)

build_with_cuda.sh

cmake -D CMAKE_BUILD_TYPE=RELEASE \
    -D CMAKE_INSTALL_PREFIX=/home/tau/software/opencv-4.5.1 \
    -D BUILD_opencv_python2=OFF \
    -D BUILD_opencv_python3=ON \
    -D PYTHON3_LIBRARY=/home/tau/anaconda3/envs/opencv_dnn_cuda/lib \
    -D PYTHON3_INCLUDE_DIR=/home/tau/anaconda3/envs/opencv_dnn_cuda/include/python3.8 \
    -D PYTHON3_EXECUTABLE=/home/tau/anaconda3/envs/opencv_dnn_cuda/bin/python3.8 \
    -D INSTALL_PYTHON_EXAMPLES=OFF \
    -D INSTALL_C_EXAMPLES=OFF \
    -D BUILD_EXAMPLES=OFF \
    -D WITH_CUDA=ON \
    -D WITH_CUDNN=ON \
    -D OPENCV_DNN_CUDA=ON \
    -D ENABLE_FAST_MATH=1 \
    -D CUDA_ARCH_BIN=7.5 \
    -D WITH_CUBLAS=1 \
    -D CUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-10.2 \
    -D OPENCV_EXTRA_MODULES_PATH=/home/tau/software/src/opencv_contrib-4.5.1/modules \
    -D OPENCV_GENERATE_PKGCONFIG=YES ..

opencv_dnn_cuda.md

Compiling and installing OpenCV DNN with CUDA

In this tutorial, we are compiling and install OpenCV 4.5.1 and OpenCV Contrib 4.5.1 with CUDA 10.2 and cuDNN 7.6.5. Other versions should aslo work.

Preparation

• Download and extract the source code of OpenCV 4.5.1 and OpenCV Contrib 4.5.1. Assume that the root directory for each is $opencv-4.5.1 and $opencv_contrib-4.5.1.
• Install CUDA 10.2 and cuDNN 7.6.5 following NVIDIA's guide. Assume that the root directory for CUDA is /usr/local/cuda-10.2.

Configuring using CMake

See build_with_cuda.sh and modify accordingly.

Compiling and Installing

make -j
make install

Installing Python Interface

After compiling and installing, there should be a file named cv2.cpython-38-x86_64-linux-gnu.so in $CMAKE_INSTALL_PREFIX/lib/python3.8/site-packages/cv2/python-3.8. Create a symlink to this file from your Python environment:

cd $PYTHON3_LIBRARY/python3.8/site-packages
ln -s $CMAKE_INSTALL_PREFIX/lib/python3.8/site-packages/cv2/python-3.8/cv2.cpython-38-x86_64-linux-gnu.so cv2.so

Test if Python Can Load OpenCV

Switch to the Python env you just installed OpenCV, then:

python
>>> import cv2
>>> cv2.__version__
`4.5.1`

@SixtyTrees I am still using the same script (but with a little modifications) to build the latest OpenCV with CUDA on my Jetson Nano board:

TIMESTAMP="release-4.8.0"

cmake -B build/${TIMESTAMP} \
    -D CMAKE_BUILD_TYPE=RELEASE \
    -D CMAKE_INSTALL_PREFIX=build/${TIMESTAMP}/install \
    -D BUILD_opencv_python2=OFF \
    -D BUILD_opencv_python3=ON \
    -D PYTHON3_LIBRARY=/usr/lib/aarch64-linux-gnu/libpython3.6m.so \
    -D PYTHON3_INCLUDE_DIR=/usr/include/python3.6m \
    -D PYTHON3_EXECUTABLE=/usr/bin/python3.6 \
    -D INSTALL_PYTHON_EXAMPLES=OFF \
    -D INSTALL_C_EXAMPLES=OFF \
    -D BUILD_DOCS=OFF \
    -D BUILD_PERF_TESTS=OFF \
    -D BUILD_TESTS=OFF \
    -D BUILD_EXAMPLES=OFF \
    -D WITH_CUDA=ON \
    -D WITH_CUDNN=ON \
    -D OPENCV_DNN_CUDA=ON \
    -D CUDA_FAST_MATH=ON \
    -D ENABLE_NEON=ON \
    -D OPENCV_DNN_CUDA=ON \
    -D ENABLE_FAST_MATH=1 \
    -D CUDA_ARCH_BIN=5.3 \
    -D WITH_CUBLAS=1 \
    -D CUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-10.2 \
    -D OPENCV_EXTRA_MODULES_PATH=/home/opencv-cn/Workspace/opencv/opencv_contrib/modules opencv

check : https://gist.github.com/madtunebk/5f20437725eb0e0cfc2a4934153b0ab4