Install opencv for Python 3.3

tim picture tim · Jan 6, 2014 · Viewed 118k times · Source

Is OpenCV still not available for Python 3.3 and do I really have to downgrade to Python 2.7 to use it? I didn't find much about it on the internet, only some posts from 2012 that OpenCV wasn't yet ported to be used in Python 3.x. But now it's 2014 and after trying to install the latest OpenCV 2.4.x and copying the cv2.pyd file to C:\Program Files (x86)\Python333\Lib\site-packages this still yields the error in Python IDLE:

>>> import cv2
Traceback (most recent call last):
  File "<pyshell#0>", line 1, in <module>
    import cv2
ImportError: DLL load failed: %1 ist keine zulässige Win32-Anwendung.

Answer

midopa picture midopa · Jan 19, 2014

EDIT: first try the new pip method:

Windows: pip3 install opencv-python opencv-contrib-python

Ubuntu: sudo apt install python3-opencv

or continue below for build instructions

Note: The original question was asking for OpenCV + Python 3.3 + Windows. Since then, Python 3.5 has been released. In addition, I use Ubuntu for most development so this answer will focus on that setup, unfortunately

OpenCV 3.1.0 + Python 3.5.2 + Ubuntu 16.04 is possible! Here's how.

These steps are copied (and slightly modified) from:

Prerequisites

Install the required dependencies and optionally install/update some libraries on your system:

# Required dependencies
sudo apt install build-essential cmake git libgtk2.0-dev pkg-config libavcodec-dev libavformat-dev libswscale-dev
# Dependencies for Python bindings
# If you use a non-system copy of Python (eg. with pyenv or virtualenv), then you probably don't need to do this part
sudo apt install python3.5-dev libpython3-dev python3-numpy
# Optional, but installing these will ensure you have the latest versions compiled with OpenCV
sudo apt install libtbb2 libtbb-dev libjpeg-dev libpng-dev libtiff-dev libjasper-dev libdc1394-22-dev

Building OpenCV

CMake Flags

There are several flags and options to tweak your build of OpenCV. There might be comprehensive documentation about them, but here are some interesting flags that may be of use. They should be included in the cmake command:

# Builds in TBB, a threading library
-D WITH_TBB=ON
# Builds in Eigen, a linear algebra library
-D WITH_EIGEN=ON

Using non-system level Python versions

If you have multiple versions of Python (eg. from using pyenv or virtualenv), then you may want to build against a certain Python version. By default OpenCV will build for the system's version of Python. You can change this by adding these arguments to the cmake command seen later in the script. Actual values will depend on your setup. I use pyenv:

-D PYTHON_DEFAULT_EXECUTABLE=$HOME/.pyenv/versions/3.5.2/bin/python3.5
-D PYTHON_INCLUDE_DIRS=$HOME/.pyenv/versions/3.5.2/include/python3.5m
-D PYTHON_EXECUTABLE=$HOME/.pyenv/versions/3.5.2/bin/python3.5
-D PYTHON_LIBRARY=/usr/lib/x86_64-linux-gnu/libpython3.5m.so.1

CMake Python error messages

The CMakeLists file will try to detect various versions of Python to build for. If you've got different versions here, it might get confused. The above arguments may only "fix" the issue for one version of Python but not the other. If you only care about that specific version, then there's nothing else to worry about.

This is the case for me so unfortunately, I haven't looked into how to resolve the issues with other Python versions.

Install script

# Clone OpenCV somewhere
# I'll put it into $HOME/code/opencv
OPENCV_DIR="$HOME/code/opencv"
OPENCV_VER="3.1.0"
git clone https://github.com/opencv/opencv "$OPENCV_DIR"
# This'll take a while...

# Now lets checkout the specific version we want
cd "$OPENCV_DIR"
git checkout "$OPENCV_VER"

# First OpenCV will generate the files needed to do the actual build.
# We'll put them in an output directory, in this case "release"
mkdir release
cd release

# Note: This is where you'd add build options, like TBB support or custom Python versions. See above sections.
cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local "$OPENCV_DIR"

# At this point, take a look at the console output.
# OpenCV will print a report of modules and features that it can and can't support based on your system and installed libraries.
# The key here is to make sure it's not missing anything you'll need!
# If something's missing, then you'll need to install those dependencies and rerun the cmake command.

# OK, lets actually build this thing!
# Note: You can use the "make -jN" command, which will run N parallel jobs to speed up your build. Set N to whatever your machine can handle (usually <= the number of concurrent threads your CPU can run).
make
# This will also take a while...

# Now install the binaries!
sudo make install

By default, the install script will put the Python bindings in some system location, even if you've specified a custom version of Python to use. The fix is simple: Put a symlink to the bindings in your local site-packages:

ln -s /usr/local/lib/python3.5/site-packages/cv2.cpython-35m-x86_64-linux-gnu.so $HOME/.pyenv/versions/3.5.2/lib/python3.5/site-packages/

The first path will depend on the Python version you setup to build. The second depends on where your custom version of Python is located.

Test it!

OK lets try it out!

ipython

Python 3.5.2 (default, Sep 24 2016, 13:13:17) 
Type "copyright", "credits" or "license" for more information.

IPython 5.1.0 -- An enhanced Interactive Python.
?         -> Introduction and overview of IPython's features.
%quickref -> Quick reference.
help      -> Python's own help system.
object?   -> Details about 'object', use 'object??' for extra details.

In [1]: import cv2

In [2]: img = cv2.imread('derp.png')
i
In [3]: img[0]
Out[3]: 
array([[26, 30, 31],
       [27, 31, 32],
       [27, 31, 32],
       ..., 
       [16, 19, 20],
       [16, 19, 20],
       [16, 19, 20]], dtype=uint8)