Face recognition based on Facenet
Project description
Face Recognition
Face Recognition Based on Facenet
Built using Facenet’s state-of-the-art face recognition built with deep learning. The model has an accuracy of 99.2% on the Labeled Faces in the Wild benchmark.
Features
Out of Box Working Face Recognition
Choose Any Pre-Trained Model from Facenet
For training just provide the proper folder structure
Faster than other available solutions
Prerequisites
You need Python(2.6 to 3.5) installed
X-based System supported (does work on Windows but not tested)
Installing
pip install facenet_recognitionSetup
Create setup as follows:
Create input directory eg: input_images
Create aligned images directory eg: aligned_images Create this directory we will store aligned images here
Create pre-trained model directory eg: pretrained_facenet_model Download Pre-Trained model from `Facenet` and keep it in the pre_model directory
Create my trained classifier directory eg: my_classifier In this directory we will save our trained model
Let’s Begin
For Facial Recognition we need to align images as follows:
import facenet_recognition
facenet_recognition.align_input('input_images','aligned_images')Above command will create our input images into aligned format and save it in given aligned images folder
Train & Test Classifier on Images
After we have aligned images now we can train our classifier.
pre_model='./pretrained_facenet_model/20170511-185253.pb' #locaiton of pret-trained model from Facenet
my_class ='./my_classifier/my_classifier.pkl' #location where we want to save
test_classifier_type = 'svm' #type of model either svm or nn
weight= './my_classifier/model_small.yaml' #local stored weights
facenet_recognition.test_train_classifier(aligned_images,pre_model,my_class,weight,test_classifier_type,nrof_train_images_per_class=30, seed=102)Mininum Required Image per person: 1 Number of Images for Training per Person: 30 (configurable)
Train Classifer on Images(only Training)
This API is used to Train our Classifier on Aligned Images
pre_model='./pretrained_facenet_model/20170511-185253.pb' #locaiton of pret-trained model from Facenet
my_class ='./my_classifier/my_classifier.pkl' #location where we want to save
test_classifier_type = 'nn' #type of model either svm or nn
weight= './my_classifier/model_small.yaml' #local stored weights
facenet_recognition.create_classifier(aligned_images,pre_model,my_class,weight,test_classifier_type)Mininum Required Image per person: 1 Number of Images for Training per Person: 30 (fixed)
Test Classifer on Images
This API is used to test our Trained Classifer
pre_model='./pretrained_facenet_model/20170511-185253.pb' #locaiton of pret-trained model from Facenet
my_class ='./my_classifier/my_classifier.pkl' #location where we want to save
test_classifier_type = 'nn' #type of model either svm or nn
weight= './my_classifier/model_small.yaml' #local stored weights
facenet_recognition.test_classifier(aligned_images,pre_model,my_class,weight,test_classifier_type)Mininum Required Image per person: 1
License
This project is licensed under the MIT License - see the LICENSE.md file for details
Acknowledgments
Big Thanks to David Sandberg for Facent
Inspired by Dlib based library face_recognition
History
0.1.4 (2018-28-03)
First beta release.
