Skip to content

OpenCV

orange requires OpenCV with the opencv_sfm module. If you also use the Red labeling app, build OpenCV with cuDNN support; otherwise the simpler CPU-only build is sufficient.

Install dependencies first

Download and unzip opencv-4.8.0.zip and opencv_contrib-4.8.0.zip to ~/build/. (For CUDA 12.2, use OpenCV 4.10 instead.)

To build OpenCV with opencv_sfm, follow OpenCV's SFM installation guide to install the SFM dependency.

Ceres solver is optional. If you want it, see Ceres' installation guide. For simplicity, disable CUDA in Ceres:

git clone --recurse-submodules https://github.com/ceres-solver/ceres-solver
cd ceres-solver
mkdir build && cd build
cmake -DUSE_CUDA=OFF ..
make -j $(nproc)
make test
sudo make install

Install cuDNN

Requires CUDA. We use cuDNN 8.9.3 with driver 525.105.17 and CUDA 12.0.

Download cudnn-linux-x86_64-8.9.3.28_cuda12-archive.tar.xz from the cuDNN archives. Extract and copy:

sudo cp cudnn-*-archive/include/cudnn*.h /usr/local/cuda/include
sudo cp -P cudnn-*-archive/lib/libcudnn* /usr/local/cuda/lib64
sudo chmod a+r /usr/local/cuda/include/cudnn*.h /usr/local/cuda/lib64/libcudnn*

Verify:

source ~/.bashrc
cat /usr/local/cuda/include/cudnn_version.h | grep CUDNN_MAJOR -A 2

Expected output:

#define CUDNN_MAJOR 8
#define CUDNN_MINOR 9
#define CUDNN_PATCHLEVEL 3
--
#define CUDNN_VERSION (CUDNN_MAJOR * 1000 + CUDNN_MINOR * 100 + CUDNN_PATCHLEVEL)

Build OpenCV with CUDA + cuDNN

cd opencv-4.8.0/
mkdir build
cd build

cmake -D CMAKE_BUILD_TYPE=RELEASE \
-D CMAKE_INSTALL_PREFIX=/usr/local \
-D WITH_TBB=ON \
-D ENABLE_FAST_MATH=1 \
-D CUDA_FAST_MATH=1 \
-D WITH_CUBLAS=1 \
-D WITH_CUDA=ON \
-D BUILD_opencv_cudacodec=OFF \
-D WITH_CUDNN=ON \
-D OPENCV_DNN_CUDA=ON \
-D CUDA_ARCH_BIN=7.5 \
-D WITH_V4L=ON \
-D WITH_QT=ON \
-D WITH_OPENGL=ON \
-D WITH_GSTREAMER=ON \
-D OPENCV_GENERATE_PKGCONFIG=ON \
-D OPENCV_PC_FILE_NAME=opencv.pc \
-D OPENCV_ENABLE_NONFREE=ON \
-D OPENCV_EXTRA_MODULES_PATH=~/build/opencv_contrib-4.8.0/modules \
-D INSTALL_PYTHON_EXAMPLES=OFF \
-D INSTALL_C_EXAMPLES=ON \
-D BUILD_EXAMPLES=ON ..

make -j $(nproc)
sudo make install

Build option B — without CUDA (simpler)

If you only need orange (no Red), this is enough:

cd opencv-4.8.0/
mkdir build
cd build

cmake -D CMAKE_BUILD_TYPE=RELEASE \
-D CMAKE_INSTALL_PREFIX=/usr/local \
-D WITH_TBB=ON \
-D ENABLE_FAST_MATH=1 \
-D CUDA_FAST_MATH=1 \
-D WITH_CUDA=OFF \
-D BUILD_opencv_cudacodec=OFF \
-D WITH_OPENGL=ON \
-D OPENCV_GENERATE_PKGCONFIG=ON \
-D OPENCV_PC_FILE_NAME=opencv.pc \
-D OPENCV_ENABLE_NONFREE=ON \
-D OPENCV_EXTRA_MODULES_PATH=~/build/opencv_contrib-4.8.0/modules \
-D INSTALL_PYTHON_EXAMPLES=OFF ..

This assumes opencv_contrib-4.8.0 is at ~/build/opencv_contrib-4.8.0 and installs OpenCV at /usr/local.

make -j $(nproc)
sudo make install