🧠 Face Recognization System

🔍 Overview

This project implements a full-stack face identification system using deep learning and web deployment.

It uses the VGGFace2 dataset to train a face recognition model, capable of identifying individuals from images. The system includes:

• ML Pipeline: Preprocessing, embedding extraction, classifier training, evaluation.
• Web Application: FastAPI backend + Django frontend for real-time image uploads and identification.

The integrated setup allows a user to upload an image, and the system will predict the person’s identity if present in the database

🏗️ Project Structure

missing_person_detection/
├── data/
│   └── vggface2/                     # VGGFace2 images organized by person/
├── models/
│   └── classifier_checkpoint.pth     # Trained classifier checkpoint
├── src/
│   ├── preprocessing.py              # Detect & crop faces
│   ├── feature_extraction.py         # Generate embeddings with VGGFace2
│   ├── datasets.py                   # PyTorch Dataset class
│   ├── models.py                     # MLP classifier architecture
│   ├── training.py                   # Training loop & checkpoint saving
│   ├── evaluation.py                 # Accuracy
│   ├── utils.py                      # Helper functions
│   ├── main.py                       # Full ML pipeline
│   └── inference.py                  # Model wrapper for serving
│
├── app_backend/
│   ├── main.py                       # FastAPI backend
│   ├── requirements.txt              # Backend-specific dependencies
│   └── model_store/                  # Optional folder for uploaded images
│
├── app_frontend/
│   ├── manage.py
│   ├── requirements.txt              # Frontend dependencies
│   ├── frontend_project/
│   │   ├── __init__.py
│   │   ├── settings.py
│   │   ├── urls.py
│   │   └── wsgi.py
│   └── webapp/
│       ├── __init__.py
│       ├── views.py
│       ├── urls.py
│       ├── templates/
│       │   └── webapp/
│       │       └── index.html
│       └── static/
│
├── requirements.txt                  # Optional global dependencies
└── README.md

📦 Dataset Description

VGGFace2 Dataset:

Kaggle Link

• ~3.3 million images of 9,000+ individuals.
• Each folder corresponds to one person with multiple images across pose, age, illumination, and expression.
• Used to train a robust face recognition model, forming the foundation for missing-person identification.

Note: For real-world deployment, this can be replaced with a specialized person dataset.

⚙️ How It Works

• Preprocessing (src/preprocessing.py)
  • Detects faces using MTCNN.
  • Crops and resizes them to uniform dimensions.
• Feature Extraction (src/feature_extraction.py)
  • Converts cropped faces into 512-dimensional embeddings using pretrained VGGFace2 models.
• Dataset Loading (src/datasets.py)
  • Organizes embeddings and labels into PyTorch Dataset objects.
  • Supports train/validation/test splits.
• Model Training (src/training.py)
  • Defines a lightweight MLP classifier.
  • Uses cross-entropy loss and Adam optimizer.
  • Saves best model as models/classifier_checkpoint.pth.
• Evaluation (src/evaluation.py)
  • Computes accuracy, and confusion matrix for performance analysis.
• Inference (src/inference.py)
  • Loads the trained model checkpoint.
  • Accepts new images and predicts identity.
• Backend (FastAPI) (app_backend/main.py)
  • Accepts image uploads via HTTP POST.
  • Runs inference and returns predicted identity + confidence score.
• Frontend (Django) (app_frontend/webapp/)
  • Provides a simple web interface for uploading images.
  • Displays predicted identity returned by the backend.
• Main Controller (src/main.py)
  • Runs the full ML pipeline: preprocessing → embedding extraction → training → evaluation → launches backend API.

🧮 Training & Running the System

Install dependencies

pip install -r requirements.txt

Run the full pipeline

python src/main.py

• Trains the model.
• Saves classifier_checkpoint.pth

launch web app

Run backend (FastAPI)

• From repo root, activate venv and install backend deps:

  pip install -r app_backend/requirements.txt

• Start backend:

uvicorn app_backend.main:app --reload --host 0.0.0.0 --port 8000

• Run frontend (Django)
  • In a separate venv, install:

  pip install -r app_frontend/requirements.txt
  

  • Start Django devserver:

  cd app_frontend
  python manage.py runserver 8001
  

Upload an image via the Django frontend

• Open http://127.0.0.1:8001/ and upload an image — backend should be running at http://127.0.0.1:8000/.
• The system predicts the identity of the person in the image.
• Returns the name

📊 Evaluation Metrics

• Accuracy
• Classification Report
• Confusion Matrix

🧑‍🔬 Research & Impact

• Demonstrates expertise in deep learning for facial recognition.
• Implements end-to-end face recognition with production-ready deployment.
• Full-stack integration shows ability to bridge ML research and practical application.