FACE RECOGNITION

Introduction

This project is a real-time face recognition system built with the power of MTCNN (Multi-task Cascaded Convolutional Networks) for accurate face detection and ArcFace for highly discriminative face recognition.

The system is capable of detecting faces from images or live video streams and recognizing individuals with high precision, even under challenging conditions such as varying lighting, angles, and facial expressions.

By combining MTCNN’s robust face localization with ArcFace’s state-of-the-art feature embeddings, this project achieves:

• Accurate face detection with bounding boxes and facial landmarks.
• Discriminative face recognition using deep feature embeddings.
• Real-time performance for practical applications.

Table of Contents

1. Introduction
2. System Architecture
3. Getting Started
4. Usage
5. API References
6. Acknowledgements

System Architecture

Getting Started

Prerequisites

• Python 3.8.0
• Dependencies are listed in requirements.txt

Installation

git clone https://github.com/salauddintapu/face_recognition.git
cd face_recognition

⚠️ Note: It is recommended to use a virtual environment (venv or conda). After creating the environment, install bcolz first:

conda create -n myenv python==3.8.0
conda install -c conda-forge bcolz
pip install -r requirements.txt

Usage

This section provides step-by-step instructions on how to use the face recognition api.

Step 1: Prepare the Environment

Prepare the environment by following the steps described in Getting Started section. Keep in mind that after setting up the venv, you must install bcolz first. Installing Python 3.8.0 and bcolz may take some time. Then install the dependencies listed in requirements.txt

Step 2: Configuration Instructions

• In the app.py, change localhost to machine's IP address if you are using a remote server.
• If cuda is avaiable the api will use your machine's GPU otherwise it will run on cpu.

Step 3: Run

To run the API, open your terminal and run,

conda activate myenv
python app.py

Prepare Your Facebank

To prepare your facebank you have to collect face data and store them in arcface/data/facebank directory. You must follow the following structure while storing data:

facebank/
├── name-of-person1/images of person1
├── name-of-person2/images of person2
├── name-of-person3/images of person3

⚠️ Note: Use cropped faces of the persons you want to recognize. Name the folder after the name of the person and store images of that person in the named folder.

Update

The api currently uses the Flask library. In the next update, Flask will be removed and FastAPI will be adopted.

API References

Endpoint: /rec

Payload Example Using cURL

curl --location 'http://localhost:5000/rec' \
--header 'Content-Type: application/json' \
--data '{
    "file": "image in base64 format"
}'

Parameters

Parameter	Type	Description
file	base64	Convert your image to base64 format. API uses opencv to process it.

Response

{
    "names": "None or list of names",
    "bboxes": "None or list of bbox of face coordinates",
    "scores": "None or list of confidence scores"
}

Acknowledgements

I would like to acknowledge the following resources and works that made this project possible:

• MTCNN Face Detector
• MTCNN Implementation
• ArcFace: Additive Angular Margin Loss for Deep Face Recognition
• InsightFace Implementation
• OpenCV
• The open-source community for providing accessible libraries and frameworks that accelerate research and development in computer vision.