Skip to content

Latest commit

 

History

History
132 lines (113 loc) · 4.36 KB

INSTALL_OPENCV_CUDA.md

File metadata and controls

132 lines (113 loc) · 4.36 KB

Use CUDA to Accelerate OpenCV

This OpenCV CUDA support is mainly for accelerating Leopard Imaging USB3.0 raw camera image processing capability Refer to:https://docs.opencv.org/3.4/d6/d15/tutorial_building_tegra_cuda.html

Test Environment

  • ubuntu 18.04 64-bit
  • NVIDIA GTX 1080
  • OpenCV 3.4.3
  • cuda 10.1

Preinstalled Libraries

Some of them might be unnecessary but I installed these all...

sudo apt-get-repository universe
sudo apt-get update
sudo apt-get install libtbb-dev
sudo apt-get install libpng12-dev
sudo apt-get install libswscale-dev
sudo apt-get install libavcodec-dev
sudo apt-get install flake8
sudo apt-get install pylint
sudo apt-get install libvtk6-dev
sudo apt-get install libopenblas-dev
sudo apt-get install libopenblas-base
sudo apt-get install x11-apps
sudo apt-get install libhdf5-serial-dev
sudo apt-get install libavformat-dev
sudo apt-get install libjasper-dev
sudo apt-get install libtiff5-dev
sudo apt-get install libjpeg8-dev
sudo apt-get install ffmpeg libv4l-dev
sudo apt-get remove libeigen3-dev # couldn't find the right directory while make, so I remove it

Install Driver

  1. remove the GPU driver you previously installed
sudo apt-get --purge remove nvidia-*
sudo apt-get --purge xserver-xorg-video-nouveau
  1. NVIDIA driver couldn't be installed under GUI environment, switch to command line, press Ctrl + Alt + F1 or try F1 until F7 see which can switch to x server. After switching to your x server, type the following
sudo /etc/init.d/lightdm stop
sudo ./NVIDIA-Linux-x86_64-390.25.RUN
  1. For installation you just choose yes or no. Most important thing is that when it asks your if using nv config file, choose yes, otherwise, x-window won't start when restarting pc. After finishing installing, run
# to restart your pc
sudo update-initramfs -u

If you run

sudo lcpci | grep nouveau

and see no output, it means you have disabled open source nouvear correctly

Install CUDA

  1. Similarly, we enter the command line x-server for installing CUDA.(Enter it by trying from ctl+alt+F1~F6 to enter, ctrl+alt+F7 is to return)
sudo /etc/init.d/lightdm stop
# switch to cuda installation package directory, and run
sudo ./cuda_10.1xxxxxx.run
sudo /etc/init.d/lightdm start # open GUI environment
  1. Add this following lines in your ~/.bashrc (or ~/.zshrc)
# for cuda
export PATH=/usr/local/cuda-10.1/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-10.1/lib64:$LD_LIBRARY_PATH
export CUDA_ROOT=$CUDA_ROOT:/usr/local/cuda-10.1/
export CUDA_INC_DIR=/usr/local/cuda-10.1/include:$CUDA_INC_DIR

and then source it

source ～/.bashrc # for normal bash terminal
source ~/.zshrc   # for zsh terminal

Install cuDNN

cuDNN is a deep learning framework designed to work on GPU accelerating.

  1. go to website: https://developer.nvidia.com/cudnn to download it
  2. then run the following
tar xvf cudnn-10.1-linux-x64-v7.1.tgz
cd cuda
sudo cp cuda/include/cudnn.h /usr/local/cuda-10.1/include
sudo cp -a cuda/lib64/libcudnn*  /usr/local/cuda-10.1/lib64
sudo chmod a+r /usr/local/cuda-10.1/lib64/libcudnn*

Testing on CUDA Installation

cd ~/samples/NIVIDIA_CUDA_10.1_Samples/1_Utilities/deviceQuery
make
./deviceQuery # after run that, you will able to see Result = PASS

nvidia-smi # see info about your nvidia GPU's current workload
nvcc -V i  # see info about nvidia CUDA compiler

Rebuild OpenCV

mkdir opencv_cuda
cd opencv_cuda
git clone https://github.com/opencv/opencv.git
cd opencv
git checkout -b 3.4.3 # I tried 4.0.0, couldn't compile ><

cd ..
git clone https://github.com/opencv/opencv_contrib.git
cd opencv_contrib
git checkout -b 3.4.3 #opencv_contrib should be same version as your OpenCV

cd ../opencv
mkdir build 
cd build

# just my builds, you can adjust it to fit your need
cmake -D WITH_CUDA=ON -D WITH_OPENGL=ON -D ENABLE_PRECOMPILED_HEADERS=OFF -D BUILD_opencv_cudacodec=OFF -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local -D WITH_TBB=ON -D BUILD_EXAMPLES=ON -D WITH_QT=ON -D WITH_OPENGL=ON -D WITH_GTK=ON -D WITH_GTK3=ON -D WITH_GTK_2_X=OFF -D WITH_VTK=OFF -D CUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-10.1 -D OPENCV_EXTRA_MODULES_PATH=../../opencv_contrib/modules ..

# this will still take a long time to make
# I did that before I got off work, so don't know the exact time it will take
make -j12
sudo make install -j12