Face Recognition in Tensorflow using prebuilt model from facenet
Download the prebuilt model from the below link https://drive.google.com/open?id=1EXPBSXwTaqrSC0OhUdXNmKSh9qJUQ55-
extract the zip file
You will find the model file 20180402-114759.pb
Finding the threshold for similarity
Create a folder Test_Images in the master folder Arrange your images in the following way
Step1:
Test_Images
Mahesh Babu
image1.jpg
image2.jpg
image3.jpg
Pawan Kalyan
image1.jpg
image2.jpg
image3.jpg
Allu Arjun
image1.jpg
image2.jpg
image3.jpg
Names need not be the way mentioned can have any name
Step2:
Create a folder named Face_distance_matrix in the master folder
Step3:
python src/compare.py path_to_20180402-114759.pb_file Test_Images/*
Step4:
python src/find_threshold.py 80 80
this gives two ouptus
1)what should be the threshold cutoff if we want to identify similar images with 80% accuracy
2)What should be cutoff threshold if we want to identify no similar images with 80% accuracy
Change percentage according to your needs.
REST API:
nohup python src/facesimilarity_rest.py 0.85 &
runs the REST Server at port 1120
Test using curl
curl POST -F "base_image=@Mahesh1.jpeg" -F "image1=@Mahesh2.jpg" -F "new_image2=@pavan2.jpeg" http://hostipaddress:1120/scoreimages