Face Recognition Module

This repository contains tools and different method to peform facial recognition. The algorithms that are supported so far are:

OpenFace: https://cmusatyalab.github.io/openface/
FaceNet: https://github.com/davidsandberg/facenet
VGG-Face: https://github.com/rcmalli/keras-vggface
face_recognition (dlib): https://github.com/ageitgey/face_recognition y https://github.com/davisking/dlib-models (68 point shape predictor)
AlexNet: https://github.com/kgrm/face-recog-eval
GoogleNet (Inception-V3): https://github.com/kgrm/face-recog-eval
SqueezeNet: https://github.com/kgrm/face-recog-eval
[NOT IMPLEMENTED] SphereFace: https://github.com/clcarwin/sphereface_pytorch

The main goal of this toolbox is to compare the performance of different facial recognition methods as done in this paper

Setup

How to install library

To use this library, clone the reopository using git clone https://github.com/evd995/face_recognition_toolbox.git

General Depencencies

Libraries used:

Numpy: pip install numpy
Matplotlib: pip install matplotlib
OpenCV: pip install opencv-python
Tensorflow: pip install tensorflow
Keras: pip install keras
- AlexNet, GoogleNet and SqueezeNet can only run in Keras 1.2
dlib: pip install dlib
h5py: pip install h5py

Method Specific Depencencies

Each method has different installing instructions, each of which can be found bellow.

FaceNet

Download pre-trained models at https://github.com/davidsandberg/facenet#pre-trained-models
Move weight to the weight directory.
When running predict it is necessary to specify the model parameter as the name of the file of the pre-trained model
2018-04-08: FaceNet has two pretrained models available, one trained with the CASIA-WebFace dataset and the other with VGGFace2. Both are trained with an Inception ResNet v1 arquitecture. The name of the respective models are:
- facenet-20180408-102900.pb (for CASIA-WebFace)
- facenet-20180402-114759.pb (for VGGFace2, default)
Pre-trained models can be downloaded at https://github.com/davidsandberg/facenet#pre-trained-models

Simple example:

import cv2

# Add package to path
import sys
sys.path.append("/PATH_TO_CLONED_GIT/face_recognition_toolbox")

# Import package
from face_recognition_toolbox import predict

# Looks for "facenet-20180402-114759.pb" (default) in the "/weights" directory
descriptor = predict(image, method_name='FaceNet')

# Looks for "facenet-20180402-114759.pb" in the "/weights" directory using "model" parameter
descriptor = predict(image, method_name='FaceNet', model="facenet-20180402-114759.pb")

VGG-Face

Run pip install keras_vggface or follow instructions in https://github.com/rcmalli/keras-vggface#keras-vggface-
2019-03-21: As of today, keras_vggface supports the models: vgg16, resnet50 and senet50. The default is resnet50 but it can be specified by passing the model parameter with the desired model name.

Simple example:

import cv2

# Add package to path
import sys
sys.path.append("/PATH_TO_CLONED_GIT/face_recognition_toolbox")

# Import package
from face_recognition_toolbox import predict

# Using default model (resnet50)
descriptor = predict(image, method_name='VGGFace')

# Specifying model with "model" parameter
descriptor = predict(image, method_name='VGGFace', model='resnet50')

face_recognition (dlib)

Run pip install face_recognition or follow instructions in https://github.com/ageitgey/face_recognition#installation

Simple example:

import cv2

# Add package to path
import sys
sys.path.append("/PATH_TO_CLONED_GIT/face_recognition_toolbox")

# Import package
from face_recognition_toolbox import predict

descriptor = predict(image, method_name='face_recognition')

OpenFace

WARNING: OpenFace can only run in Python 2.7

Follow instructions at https://cmusatyalab.github.io/openface/setup/

AlexNet, GoogleNet and SphereFace

WARNING: AlexNet, GoogleNet and SphereFace can only run in Python 2.7 and Keras 1.2.2

Download weights at https://github.com/kgrm/face-recog-eval
Move weight to the weight directory.
If currently installed version of Keras is greater than 1.2.2, it has to be downgraded before running this method with the command pip install --upgrade keras==1.2.2

Tutorials/Demos

A simple impletentation can be found in this notebook

How to contribute

To contribute it is possible to perform a pull request. If your goal is to add a new method then follow the next steps:

Add a module in the methods directory that contains a class with a predict method.
Add the reference to the class here
Add instance and call to the predict method in utils.py