OpenCV released OpenCV-3.4.4 and OpenCV-4.0.0 on 20th November. There have been a lot of bug fixes and other changes in these versions. The release highlights are as follows:
- OpenCV is now C++11 library and requires C++11-compliant compiler. Minimum required CMake version has been raised to 3.5.1.
- A lot of C API from OpenCV 1.x has been removed.
- Persistence (storing and loading structured data to/from XML, YAML or JSON) in the core module has been completely reimplemented in C++ and lost the C API as well.
- New module G-API has been added, it acts as an engine for very efficient graph-based image procesing pipelines.
- dnn module now includes experimental Vulkan backend and supports networks in ONNX format.
- The popular Kinect Fusion algorithm has been implemented and optimized for CPU and GPU (OpenCL)
QR code detector and decoder have been added to the objdetect module. - Very efficient and yet high-quality DIS dense optical flow algorithm has been moved from opencv_contrib to the video module.
In this post, we will provide a bash script for installing OpenCV-4.0 (C++ and Python 3.5) on Ubuntu 16.04. We will also briefly study the script to understand what’s going in it. Note that this script will install OpenCV in a local directory and not on the entire system.
1. Install OpenCV 4.0
Step 0: Select OpenCV version to install
echo "OpenCV installation by learnOpenCV.com"
# Define OpenCV version to install
cvVersion="master"
We are also going to clean build directories and create installation directory.
# Clean build directories
rm -rf opencv/build
rm -rf opencv_contrib/build
# Create directory for installation
mkdir installation
mkdir installation/OpenCV-"$cvVersion"
Finally, we will be storing the current working directory in cwd variable. We are also going to refer to this directory as OpenCV_Home_Dir throughout this post.
# Save current working directory
cwd=$(pwd)
Step 1: Update Packages
sudo apt -y update
sudo apt -y upgrade
Step 2: Install OS Libraries
sudo apt -y remove x264 libx264-dev
## Install dependencies
sudo apt -y install build-essential checkinstall cmake pkg-config yasm
sudo apt -y install git gfortran
sudo apt -y install libjpeg8-dev libjasper-dev libpng12-dev
sudo apt -y install libtiff5-dev
sudo apt -y install libtiff-dev
sudo apt -y install libavcodec-dev libavformat-dev libswscale-dev libdc1394-22-dev
sudo apt -y install libxine2-dev libv4l-dev
cd /usr/include/linux
sudo ln -s -f ../libv4l1-videodev.h videodev.h
cd $cwd
sudo apt -y install libgstreamer0.10-dev libgstreamer-plugins-base0.10-dev
sudo apt -y install libgtk2.0-dev libtbb-dev qt5-default
sudo apt -y install libatlas-base-dev
sudo apt -y install libfaac-dev libmp3lame-dev libtheora-dev
sudo apt -y install libvorbis-dev libxvidcore-dev
sudo apt -y install libopencore-amrnb-dev libopencore-amrwb-dev
sudo apt -y install libavresample-dev
sudo apt -y install x264 v4l-utils
# Optional dependencies
sudo apt -y install libprotobuf-dev protobuf-compiler
sudo apt -y install libgoogle-glog-dev libgflags-dev
sudo apt -y install libgphoto2-dev libeigen3-dev libhdf5-dev doxygen
Step 3: Install Python Libraries
sudo apt -y install python3-dev python3-pip
sudo -H pip3 install -U pip numpy
sudo apt -y install python3-testresources
We are also going to install virtualenv and virtualenvwrapper modules to create Python virtual environments.
cd $cwd
############ For Python 3 ############
# create virtual environment
python3 -m venv OpenCV-"$cvVersion"-py3
echo "# Virtual Environment Wrapper" >> ~/.bashrc
echo "alias workoncv-$cvVersion=\"source $cwd/OpenCV-$cvVersion-py3/bin/activate\"" >> ~/.bashrc
source "$cwd"/OpenCV-"$cvVersion"-py3/bin/activate
# now install python libraries within this virtual environment
pip install wheel numpy scipy matplotlib scikit-image scikit-learn ipython dlib
# quit virtual environment
deactivate
Step 4: Download opencv and opencv_contrib
git clone https://github.com/opencv/opencv.git
cd opencv
git checkout $cvVersion
cd ..
git clone https://github.com/opencv/opencv_contrib.git
cd opencv_contrib
git checkout $cvVersion
cd ..
Master Generative AI for CV
Step 5: Compile and install OpenCV with contrib modules
First we navigate to the build directory.
cd opencv
mkdir build
cd build
Next, we start the compilation and installation process.
cmake -D CMAKE_BUILD_TYPE=RELEASE \
-D CMAKE_INSTALL_PREFIX=$cwd/installation/OpenCV-"$cvVersion" \
-D INSTALL_C_EXAMPLES=ON \
-D INSTALL_PYTHON_EXAMPLES=ON \
-D WITH_TBB=ON \
-D WITH_V4L=ON \
-D OPENCV_PYTHON3_INSTALL_PATH=$cwd/OpenCV-$cvVersion-py3/lib/python3.5/site-packages \
-D WITH_QT=ON \
-D WITH_OPENGL=ON \
-D OPENCV_EXTRA_MODULES_PATH=../../opencv_contrib/modules \
-D BUILD_EXAMPLES=ON ..
make -j4
make install
2. How to use OpenCV in C++
Using CMakeLists.txt
The basic structure of your CMakeLists.txt will be as follows:
cmake_minimum_required(VERSION 3.1)
# Enable C++11
set(CMAKE_CXX_STANDARD 11)
set(CMAKE_CXX_STANDARD_REQUIRED TRUE)
You will have to set OpenCV_DIR as shown below.
SET(OpenCV_DIR <OpenCV_Home_Dir>/installation/OpenCV-master/lib/cmake/opencv4)
Make sure that you replace OpenCV_Home_Dir with correct path. For example, in my case:
SET(OpenCV_DIR /home/hp/OpenCV_installation/installation/OpenCV-master/lib/cmake/opencv4)
Once you have made your CMakeLists.txt, follow the steps given below.
mkdir build && cd build
cmake ..
cmake --build . --config Release
This will generate your executable file in build directory.
3. How to use OpenCV in Python
To use the OpenCV version installed using Python script, first we activate the correct Python Virtual Environment.
For OpenCV-4 : Python 3
workon OpenCV-master-py3
Once you have activated the virtual environment, you can enter Python shell and test OpenCV version.
ipython
import cv2
print(cv2.__version__)
Hope this script proves to be useful for you :). Stay tuned for more interesting stuff. In case of any queries, feel free to comment below and we will get back to you as soon as possible.
Satya,
I have been looking for something like this! I will modify it as I compile all the Cuda in and I don’t need python 2.7.
For a student working here I have been building OpenCV 3.4.1 with out Cuda or Python bindings, she is used to finding a opencv_world341d.dll (.lib) file and I do not get one. What am I missing>
Thanks,
Doug
Hi Douglas! Thanks for the reply. In Windows systems, you have to make sure that OpenCV_DIR environment variable has been setup. Can you make sure that’s the case?
Hi Vishwesh,
It wasn’t the environment variable, I just need to click the option to build the opencv_world3XX.dll file.
Hi Douglas!
Does it work fine now?
I have opencvXXX_world.dll’s not so yes it is working.
Excellent! I tried the 3.4.1 version and installed OK! I’ll try now the Master version.
Just a question: after completing the installation, I have this tree:
/installOpenCV.sh
installation
opencv
opencv_contrib
Can I delete the ‘opencv’ folder? It occupies almost 8 Gb and don’t know if it is necessary.
Many thanks, Satya
Thank you Cesar! It’s great to see that the script was useful. Sure you can remove the opencv and opencv_contrib folders.
Hi Satya, just an issue in the script: I’d like to have both OpenCV-3.4.1 and OpenCV-4 installed, as you say at the beginning, but the script deleted the ‘installation’ directory (line 9) when I ran the script for second time to install OpenCV 4. Then I managed to change the script a bit to have both versions installed.
Bests,
Cesar
Hi Cesar! Thank you for pointing it out. I will make the corresponding change in the script.
Update: I have removed the “rm -rf installation” line from the post as well as the script on github. Thanks!
Thanks. I executed the bash script and got the following errors
make[2]: *** No rule to make target ‘/home/indusai/anaconda3/lib/libQt5Widgets.so.5.6.0’, needed by ‘lib/libopencv_cvv.so.4.0.0’. Stop.
CMakeFiles/Makefile2:13459: recipe for target ‘modules/cvv/CMakeFiles/opencv_cvv.dir/all’ failed
make[1]: *** [modules/cvv/CMakeFiles/opencv_cvv.dir/all] Error 2
Makefile:160: recipe for target ‘all’ failed
make: *** [all] Error 2
I still find I have 3.3
>>> import cv2
>>> cv2.__version__
‘3.3.0’
Can you please help?
Walid
Hi Walid! Can you please post the complete error? Also, try to run commands one by one from your command line directly as has been described in the post. That will help us in finding the error.
There is something I do not understand:
I compared the scripts with Adrian Rosebrook script https://www.pyimagesearch.com/2018/08/15/how-to-install-opencv-4-on-ubuntu/
to install opencv-4 for python and noticed two things:
a) your script is more complete, as it is meant to wotk with both c++ and python (which is a great idea: your publishing in both languages teaches/taught me a lot)
b) pyimagesearch uses a very cryptic way (an hexa key) to select opencv4; you use a very simple keyword (“master”)
And this is the later point (b) I do not understand. I guess pyimagesearch choses a given version of cv4, and you use the latest and {greatest|buggiest|debugged} one… But, as I use git as a black box, this is only a guess…
Hi Masque!
The reason we are going with master is that it keeps on updating as and when the bugs are resolved or some new features are being added. Now, this also means that it is the least bugged but it also means that it can be highly unstable. Thankfully, the second one is not the case right now. As soon as OpenCV officially announces the OpenCV 4, we will revise the script to use that particular branch or commit.
I am really glad to know that you appreciate the installation script.
Happy hacking!
Vishwesh
pip install opencv-contrib-python==3.4.3 should also work though, right???
For the python-version – Yes! But there too, you cannot use the latest ( master ) branch. It will install the latest Release only.
Moreover, these instructions are for those who want to
1. install from source with their own modifications and
2. For running both C++ and python code
In an effort to figure out why things are not working I am running line by line. I want to load OpenCV 3.4.3 and 4.0. i also do not use Python 2 so that section has been commented out. Since I already have virtualenv installed I have commented out that section.
So in running line by line, when I execute line:
sudo apt-get -y install libjpeg8-dev libjasper-dev libpng12-dev I get the following error:
Reading package lists… Done
Building dependency tree
Reading state information… Done
Package libpng12-dev is not available, but is referred to by another package.
This may mean that the package is missing, has been obsoleted, or
is only available from another source
However the following packages replace it:
libpng12-0
E: Unable to locate package libjasper-dev
E: Package ‘libpng12-dev’ has no installation candidate
This is on Ubuntu 18.04 which I believe you used for you script.
Suggestions?
Thanks,
Doug
Hi Doug!
You are right. You can checkout the blog for Ubuntu 18.04 installation for the same.
https://learnopencv.com/install-opencv-4-on-ubuntu-18-04/
Thanks
Thanks a lot for this tutorial. I follow all the instrutions, without any problem, but when applying workon … command, I get an error : ERROR: Environment ‘OpenCV-3.4.1-py3’ does not exist. Create it with ‘mkvirtualenv OpenCV-3.4.1-py3’.
PLease, do you know how I can solve it?
Thanks in advance for your response.
Hi @Rogers_mm
Can you provide the output of :
lsvirtualenv -l
Thanks
Vishwesh,
If I wanna install only for c++ and without the script, it would be done by only omitting python sections and following the post?
Hi Jovann!
Yes. Also make sure that you modify the cmake command:
cmake -D CMAKE_BUILD_TYPE=RELEASE
-D CMAKE_INSTALL_PREFIX=$cwd/installation/OpenCV-“$cvVersion”
-D INSTALL_C_EXAMPLES=ON
-D WITH_TBB=ON
-D WITH_V4L=ON
-D WITH_QT=ON
-D WITH_OPENGL=ON
-D OPENCV_EXTRA_MODULES_PATH=../../opencv_contrib/modules
-D BUILD_EXAMPLES=ON ..
There’s no need to create any environment or install any python module separately for only C++ installation.
thanks for script its works with opencv4 but when i install tensorflow on same virtual environment
with pip install tensorflow installation is successful but when i call module it show error
Illegal instruction (core dumped)
can you help me with this?
hi i solved the issue i need to downgrade the my tensorflow 1.11 to tensorflow 1.5 and it works. for ubuntu 16.04
Hello community,
I installed everything following the guide but I have the problem that it cannot import cv2, as there is ‘no such module’. I tried in workon, that happened, I tried outside workon – it again happened. Weirdly enough iPython loads inside workon. It was compiling open-cv for around 40 minutes now and I am pretty sure it is successfully compiled on my computer but to some reason it just cannot be found. Can you help me to fix that because I am a bit to new to linking and cmaking etc.
Besides that, can I use CUDA optimizations or TBB on opencv 4? Do they work with the Python version, if so, how can I activate them? Need to recompile again?
Konstantin,
I use 18.04 but it should not make that much difference. If you want to use OpenCV 4 in Python then you must link the cv2*.so file into the virtual environment that you are using. You do not say what Python you are using or if you are using the installation folder that is called out in the article. So I will use my set up as an example.
On my machine I have two hard drives. I mount my development drive, i.e. where I do all my coding, at /devel under that is a software folder that I use to hold any 3rd party software I use. I do use the installation folder that the scripts use but I do not run the scripts as I was bitten badly once upon a time. If you ran the make install command then you should find in the installation folder an OpenCV-4.0.0 folder. If you navigate to /installation/OpenCV-4.0.0/python/cv2/python-3.6 you will find the file cv2.cpython-36m-x86_64-linux-gnu.so (this assumes a Intel architecture). If this is all there then you can go to your virtual environment and link the file in. For me that is at ~/.virtualenvs/cv/lib/python3.6/site-packages. Then link the file using ln -f -s /devel/installation/OpenCV-4.0.0/python/python-3.6/cv2*.so cv2.so
From there you should do a cd ~/
workon cv
python
import cv2
And everything should work.
You can indeed compile for TBB and Cuda. Here is the cmake line I use:
cmake -D CMAKE_BUILD_TYPE=Release
-D CMAKE_INSTALL_PREFIX=$cwd/installation/OpenCV-“$cvVersion”
-D INSTALL_C_EXAMPLES=OFF
-D INSTALL_PYTHON_EXAMPLES=ON
-D WITH_TBB=ON
-D WITH_V4L=OFF
-D WITH_QT=ON
-D WITH_OPENGL=ON
-D WITH_CUDA=ON
-D BUILD_opencv_cudacodec=OFF
-D OPENCV_EXTRA_MODULES_PATH=/devel/software/opencv_contrib/modules
-D BUILD_EXAMPLES=OFF ..
I hope this helps.
Doug
Doug,
Thank you very much! Seconds ago I arrived to exactly the same solution! And it works. I am just really new to cmake and linking this way, I just get confused which file goes where while installation, which is linked, which not, where is the virtual environment, that kind of thing. In my case was exactly how you described it, I had python 3.6 and the sample was with python 3.5, so I just had to link the cython to the virtual environment.
In order for me to compile opencv 4 with TBB and Cuda, do I have to uninstall it completely or it will just rewrite whatever I’ve currently?
Another, last question I have is, can the python opencv 4 use these optimisations or it’s just for the c++? I am mainly interested whether I can use the Dense optical flow with Dense Inverse Search in python and wheather I will get anything out of TBB or it is exclusive to the C++ version of the library.
OpenCV should really make a normal installer one day apart. Thank you so much for the extremely exhaustive answer, it was my case, I am sure you will help other people after me too with it!
Hi Konstantin,
You are quite welcome! I am glad I could help. When ever I compile OpenCV, I always do a rm -rf * in the build folder. Then I know everything is “clean”. Form there the cmake then make.
In general, Python does not use the Cuda optimizations in OpenCV, you will need to use C++ for that.
As for a normal installer for OpenCV, not sure that will ever happen considering how many platforms the code is run on and ported to. I agree it would be nice but that would be a bunch of work.
Dougf
Thank you so much! Actually, the installed opencv comes with tbb. I don’t know if python uses it or not but from the first go I obtained 25-28fps compared to 15fps before running the same code. I may as well consider switching to C++ and recompiling with CUDA to see what further I can get out of it but even that for now is a miracle. Will give a go to the DIS as well! Thank you again Douglas!