Facial Recognition App

This is a full-stack application that uses a FastAPI backend for face recognition and a React frontend for the user interface, packaged into a standalone main.exe. The React frontend running in a browser acts as the client, capturing and sending images to the backend via a RESTful API for processing. The backend processes the image using the face_recognition library and returns a JSON response to the frontend.

Features

• Real-time Camera Access: Images are captured directly from users' brower webcam. Users can alternatively upload a local image file.
• Face Recognition: The FastAPI backend, powered by the face_recognition and dlib libraries, processes the captured image to detect and recognize faces.
• Dynamic Database: The application supports dynamically adding and deleting "known" faces to its recognition database via API endpoints in a session.
• Standalone Executable: The entire backend, including all Python dependencies and the React build, is bundled into a single .exe file.

Running the App

Download the executable packaged in the latest release. It should be ready for use without any dependencies. If you're running the web version, it should start a terminal (this can be avoided with the usage of the included .vbs script). Make sure to Ctrl+Click the IP address that pops up to open the app hosted locally. The local version shouldn't require anything else than starting the .exe.

The developer setup is in detail below.

Developer Setup

Backend

1. Make sure you have Python installed (3.10+ recommended).

2. Install dependencies:

   pip install -r backend/requirements.txt

Frontend

1. Make sure you have Node.js and npm installed.

2. Navigate to the frontend folder and install dependencies:

   npm install

Dev App Run

Put images of faces you wish to recognize in the backend/data/ folder.

You need to run both the frontend and backend simultaneously in separate terminals. Navigate to their respective folders before running these commands. You also need to create a /data folder under /backend with face photos to be recognized, and addition /train and /test folders if you choose to use the in-house MobileNetV2 algorithm (you'll also have to run model.py to train first).

• face_recognition: produces a 128-D encoding vector per face, and you manually compare with known encodings. Flexible but not trainable.
• Keras classifier: learns to directly map raw images → class labels. Requires a fixed training dataset and retraining if you add new people.

1. Backend:

   face_recognition algorithm

   uvicorn main:app --reload --host 127.0.0.1 --port 8000

   in-house training algorithm

   uvicorn main_classifier:app --reload --host 127.0.0.1 --port 8000

2. Frontend:

   npm start

The React app should open automatically in your browser. If your device has a camera, you can start using it to capture images and send them to the backend for recognition.