Easy_Image

Motivation

OpenCV and Dlib offer some of the best facial recognition algorithms for free, as of this writing. Unfortunately, I've ended up having to write a lot of duplicate code for the most common facial recognition and image manipulation operations across my various programs. So I've built wrapper classes and functions into this library to make the process easier.

There are 2 features that differentiate this library (and my approach) from most other approaches to facial recognition:

Faces are cached in the image EXIF data so they don't need to be redetected, saving time in future executions.
Use of distinct objects for faces, images, and even images loaded from files rather than working with numpy arrays, at least at a higher level.

Installation

As of now, Easy Image has only been tested to run on Windows 10 with Anaconda3 under Python 3.6. I may test other versions of Python and other OS's in the future and report my results here. You're also welcome to try it out on other setups - feel free to report any issues here.

Windows and Mac:

First, download and install Anaconda3

Open a Terminal or Command Prompt

You'll need to set up a few dependencies before running:

conda install path.py numpy opencv Pillow scikit-image scikit-learn pandas

conda install -c pytorch pytorch (alternatively, conda install -c pytorch pytorch-cpu if you don't have a supported GPU). If you have an older GPU but pytorch complains that it's CUDA capabilities are too old but you really want to use the GPU, then your only choices are to either compile it from source if you want the latest Pytorch (see my video here) or to install Pytorch 0.3.0 to save yourself a headache by following the instructions below.

Install dlib, which may be a bit tricky:

pip install dlib

Failing that, see the instructions in this [stackoverflow post](pip install https://pypi.python.org/packages/da/06/bd3e241c4eb0a662914b3b4875fc52dd176a9db0d4a2c915ac2ad8800e9e/dlib-19.7.0-cp36-cp36m-win_amd64.whl#md5=b7330a5b2d46420343fbed5df69e6a3f).

Failing that, install from conda-forge as a last resort, but it may cause conflicts with your other libraries:

conda install -c conda-forge dlib

Continue installing the rest of the dependencies:

pip install piexif imutils face_recognition_models torchvision fake-useragent

Then download and install Easy Facial Recognition:

git clone https://github.com/xjdeng/Easy_Facial_Recognition

cd Easy_Facial_Recognition

pip install -U .

Installing Pytorch 0.3.0 in Windows

If you're using Windows and have an older GPU, you might be able to get it to work if you use Pytorch 0.3.0 on either Python 3.5 or 3.6 (2.7 and 3.7 aren't supported, unfortunately)

First, go to Peterjc123's Google Drive folder and download the version corresponding to your Python version (3.5 or 3.6). Make sure you pick the one corresponding to your version of CUDA (or lack thereof, in that case, pick a CPU version.)

Then install using conda:

conda install <the file you downloaded>

If you're using Linux or Mac and have an old GPU, you will need to install Pytorch 0.3.0 from source from their repository.

What can you do with this module?

Note: here's a Youtube Video I made on detecting and caching faces using this module.

First, open a terminal or command prompt in a working directory of your choice. Then run the following to copy some test images to this directory. Do this before running any of the following examples! Then run the following sections in order.

from Easy_Image import test

import Easy_Image

test.copytests()

Initialize a Face with the EasyImage object:

import cv2

angryimg = Easy_Image.EasyImage(cv2.imread("tests/angry-2191104_640.jpg"))

Initialize an image file using the EasyImageFile object:

woman = Easy_Image.EasyImageFile("tests/woman-3046960_640.jpg")

Detect Faces (EasyImage and EasyImageFile objects):

angryface = angryimg.detect_faces()

import time

t0 = time.time()

womanface = woman.detect_faces()

t1 = time.time()

print(str(t1 - t0) + " seconds to detect face.")

Redetect faces in an EasyImageFile object again:

t0 = time.time()

womanface2 = woman.detect_faces()

t1 = time.time()

print(str(t1 - t0) + " seconds to detect face this time")

Since the faces were stored using EXIF tags in the file tests/woman-3046960_640.jpg, it took less time after running detect_faces() again.

Draw boxes around faces:

angryfacebox = angryimg.draw_faces()

womanbox = woman.draw_faces()

cv2.imshow("face 1", angryfacebox.getimg())

cv2.waitKey(0)

Detect objects in an image (requires Pytorch):

angryimg.classify()

The above example uses the implementation from this blog post on PyImageSearch.

xjdeng/Easy_Image