Inference_on_facenet_pytorch

This repo is about doing inferencing on pretrained model taken from facenet-pytorch. Running the pretrained model on an image return am embeddings of the person's face with size (1, 512). These embeddings can be used as per requirement.

How to run

• Clone the repo using git clone https://github.com/mohdsaqibhbi/Inference_on_facenet_pytorch.git.
• Go to this directory using cd Inference_on_facenet_pytorch.
• Create virtual environment.
• Install dependencies using pip install -r requirements.txt.
• Create the database and then use live face detection to test it.
• To create the database of embeddings
  • Put the images in the folder data/database/images/
  • Run the command python create_database.py -in data/database/images/ -o data/database/database.pkl
• To update the database with new person's embeddings
  • Run the command python update_database.py -in data/database/database.pkl -i data/Chadwick_Boseman.jpg -n Chadwick_Boseman.jpg"
• For live face detection
  • Need to create database embeddings first.
  • Run the command python live_detection.py -d data/database/database.pkl -th 1.0
• To understand step by step, how to create database, update database, face verification, face recognition and live face detection, follow the jupyter-notebook Face_Recognition.

Command line arguments

• create_database.py

  tag (* = required)	variable	options	default value
  -in *	in_path	path of the input images	REQUIRED
  -o	out_path	path of database to be saved	"database.pkl"

• update_database.py

  tag (* = required)	variable	options	default value
  -in *	in_path	path of the input database	REQUIRED
  -i *	image	path of input image	REQUIRED
  -n	name	name of the person	None

  Note : If name is not given, image name will be used as person's name.

• live_detection.py

  tag (* = required)	variable	options	default value
  -d *	database	path of the database	REQUIRED
  -th	threshold	threshold to euclidean distance	1.0