Skip to main content

OpenCV (with CUDA)

Dependencies#

apt install the following

  • libopenjpip7 libopenjp2-tools
  • libatlas-base-dev gfortran
  • liblapacke-dev libatlas-base-dev

Install OpenCV 4 from Source#

  • Clone repos of OpenCV and OpenCV Contrib, and
  • Checkout the matching release (e.g. 4.1.0)
cd ~/opencvmkdir buildcd build
cmake -D CMAKE_BUILD_TYPE=RELEASE \    -D CMAKE_INSTALL_PREFIX=/usr/local \    -D INSTALL_PYTHON_EXAMPLES=ON \    -D INSTALL_C_EXAMPLES=OFF \    -D OPENCV_ENABLE_NONFREE=ON \    -D OPENCV_EXTRA_MODULES_PATH=../../opencv_contrib/modules \    -D PYTHON3_EXECUTABLE=/usr/bin/python3 \    -D PYTHON2_EXECUTABLE=/usr/bin/python2 \    -D BUILD_EXAMPLES=ON \    -D WITH_CUDA=ON \    -D CUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda \    -D CUDA_ARCH_BIN="6.2" \    -D BUILD_opencv_cudacodec=OFF \    ..    
  • TX2 uses 6.2; For CUDA_ARCH_BIN versions, please refer to GPU Compute Capability
  • Note: cuda10以上没有dynlink_nvcuvid.h和nvcuvid.h,所以要将BUILD_opencv_cudacodec=OFF

Reference 1

Reference 2

Last updated on