A workshop introducing computer vision tools in python. A python program that goes through photos in a directory, detects faces in them, warps (scale and alignement) them and overlays them to create the average face of the dataset.
No existing python knowledge is required.
Operating System: This workshop has been written on OSX, others may also work.
Other dependencies: opencv, dli, virtualenv (optional). An installation guide is below.
- Using Facemash
- Setting up a (virtual) working environment
- Installation guide for basic requirements (Beginners start here!)
basics 1
$ python basics1.py data/face1.jpg
basics 2
$ python basics2.py data/face1.jpg data/face2.jpg
facemash
$ python facemash.py data/students
[1] Navigate to a directory in which you wanna start the project.
For example like this:
$ cd ~/Desktop
$ mkdir facemash
$ cd facemash
[2] Create a new "virtualenv", this allows us to run our program within a controlled environment without interferring with the global system.
$ virtualenv env
virtualenv is the command, env the name we give our environment, typing $ ls shows you that a new directory with that name has been created, DON'T WORRY about that one, ever.
[3] Access the virtual environment
$ source env/bin/activate
A (env) at the beginning of your command line prompt signals that you are IN the virtual environment. To exit, simply enter deactivate (not for now though).
[4] Transfer our global installation of OpenCv into our virtual environment
This one is a bit strange, but a little trick does the job in my experience. First we need to find our global opencv files, chances are that
$ ls /usr/local/lib/python2.7/site-packages/cv*
will return
/usr/local/lib/python2.7/site-packages/cv.py
/usr/local/lib/python2.7/site-packages/cv2.so
/usr/local/lib/python2.7/site-packages/cv.pyc
Those are the files we are looking for, the location of them can differ. Once found, while still in the virtual environment, we copy them with
$ cp /usr/local/lib/python2.7/site-packages/cv* $VIRTUAL_ENV/lib/python2.7/site-packages/
I found this trick here.
[5] Install Numpy & Scipy and test OpenCV
Any python module we install with pip, will be installed INTO the virtual enviromnent. OpenCV requires the following
$ pip install numpy scipy
OpenCV should now work! To test open the interactive python interpreter with $ python, and enter import cv2. No error message should appear. Yes? Great!
[6] Install dlib
With boost and cmake installed (in the installation guide below), this should be easy:
$ pip install dlib
[6] Download a facial-landmarks-prediction model
Download the file you find at https://github.com/biometrics/openbr-models/blob/master/dlib/shape_predictor_68_face_landmarks.dat and just put it into your project folder.
[7] Get some datasets of faces
Store in data/[name of dataset].
For itp, I supply a set of current students' faces.
This should be it! Now get coding on the project, or use the finished pieces in this repo!
The following has only been tested on OSX. It looks like a lot, but the good news is, it has always worked super smoothly when I tried it!
Enter all commands without the $ symbol :-)
This should be installed on your machine bby default, run $ python --version to make sure. If you actually run into issues with this, this should help (chances are, you'll install python with brew (see below) if you actually don't have it yet).
Homebrew (more info)
Check if you have it with $ which brew, this should return a path.
If it doesn't, install it by running
$ /usr/bin/ruby -e "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install)"
you will need to enter your password for this.
Check with $ which pip.
Install it by running
$ sudo easy_install pip
Check with $ which virtualenv. Install it by running
$ pip install virtualenv
OpenCV (more info)
Check by starting the interactive python interpreter, enter $ python, the command line prompt will change to >>>, now enter import cv2. If nothing happens, great! Something like [...] ImportError: No module named cv2 [...] means you need to install OpenCV by entering the following commands:
$ brew tap homebrew/science
$ brew install opencv
Above test might still return the same error or a different one [...] ImportError: numpy.core.multiarray failed to import [...], but, trust me, DON'T WORRY (we'll fix that within the virtual environent later on).
Boost (more info)
This, we need to install dlib in our virtual environent later on. Install entering the following commands:
$ brew install boost --with-python
$ brew install boost-python
Test if that worked with
$ brew list | grep 'boost'
which should return
boost
boost-python
The steps to install boost are descript in more detail here.
Check with $ which cmake. Install it by running
$ brew install cmake
... Now we should be all set to set up our virtual environment!!!