Install

Prerequisites

To install PyMeanShift from the source code (source package or Git repository), Python 2.7 (or 3.2) and Numpy development headers are required:

• On recent Debian-based (ex. Ubuntu) Linux distributions, the Python headers are provided by the package python-dev (2.7) or python3-dev (3.2); The Numpy headers are provided by the package python-numpy-dev (2.7) or python3-numpy-dev (3.2).
• On recent Redhat-based Linux distributions (ex. Fedora), the Python headers are provided by the package python-devel (2.7) or python3-devel (3.2); The Numpy headers are provided by the package numpy (2.7) or python3-numpy (3.2).
• On MacOS X, the Python headers are installed during the Python installation process (with the official Python release, not the MacOS X version of Python); The Numpy headers are installed during the Numpy installation process.
• On Windows, the Python headers are installed during the Python installation process. The Numpy headers are installed during the Numpy installation process.
  • Also needed: Visual C++ 2008 Express Edition with SP1 (http://www.microsoft.com/visualstudio/en-us/products/2008-editions/express).
  • Also needed for Python 64 bits: Windows 7 Platform SDK is needed (with some hacks) to compile a 64 bits extension for Python 64 bits (see http://mattptr.net/2010/07/28/building-python-extensions-in-a-modern-windows-environment/ ); A 64 bits version of Numpy is also required; An unofficial release is available at http://www.lfd.uci.edu/~gohlke/pythonlibs/ (the installer contains both the binaries and the headers).

---

Build and Install

1. The Python extension is compiled as follows: * Linux and MacOS X (in a terminal window): bash cd path-to-pymeanshift-sources ./setup.py build * Windows (at the command prompt): bash cd path-to-pymeanshift-sources python setup.py build
2. The wrapper module and the extension can be installed as follows: * Systemwide, for all users (admin privileges are needed):
   • Linux and MacOS X (in a terminal window):

   sudo ./setup.py install

   • Windows (at the command prompt, from an admin account):

   python setup.py install

* For one user only:
  * **Linux** and **MacOS X** (in a terminal window):
  ```bash
  ./setup.py install --user
  ```
  * **Windows** (at the command prompt):
  ```bash
  python setup.py install --user
  ```

3. If everything went fine, it should be possible to import the pymeanshift module in Python code. The module provides a function named segment and a class named Segmenter.

Code example with OpenCV:

import cv2
import pymeanshift as pms

original_image = cv2.imread("example.png")

(segmented_image, labels_image, number_regions) = pms.segment(original_image, spatial_radius=6, 
                                                              range_radius=4.5, min_density=50)

Code example with PIL:

from PIL import Image
import pymeanshift as pms

original_image = Image.open("example.png")

(segmented_image, labels_image, number_regions) = pms.segment(original_image, spatial_radius=6, 
                                                              range_radius=4.5, min_density=50)

Code example using the Segmenter class:

import pymeanshift as pms

# [...]
# load image in "original_image"
# [...]

my_segmenter = pms.Segmenter()

my_segmenter.spatial_radius = 6
my_segmenter.range_radius = 4.5
my_segmenter.min_density = 50

(segmented_image, labels_image, number_regions) = my_segmenter(original_image)