Real-Time Deepfake Detection Application

This application uses MesoNet and FastAPI to detect deepfakes in uploaded videos, images, or live webcam streams.

Features

• Upload videos for deepfake analysis
• Upload images for instant deepfake detection
• Live webcam deepfake detection
• Per-frame confidence scores
• Visual timeline of detection results
• Face detection and highlighting

Installation

Prerequisites

• Python 3.8 or higher
• macOS users with Apple Silicon (M1/M2/M3) require special TensorFlow installation

Setup Instructions

1. Clone this repository

   git clone <repository-url>
   cd deepfakerealtimedetection

2. Create and activate a virtual environment (recommended)

   python3 -m venv venv
   source venv/bin/activate

3. Install the required packages

   # For Intel-based systems
   pip install -r requirements.txt
   
   # For Apple Silicon (M1/M2/M3) Macs
   # Note: You might need to install packages individually if you encounter issues
   pip install fastapi uvicorn
   pip install numpy opencv-python mediapipe python-multipart pillow jinja2 aiofiles
   pip install tensorflow-macos tensorflow-metal

Usage

Full Application

Start the server:

python3 run.py

Then open your browser and go to http://localhost:8000

Using the Application

1. Upload a video file, image file, or use your webcam for real-time detection
2. View the analysis results showing confidence scores for deepfake probability
3. For videos, examine the frame-by-frame timeline for detailed analysis
4. For images, see highlighted faces with individual confidence scores

Technologies Used

• FastAPI for the backend API
• MesoNet model for deepfake detection
• MediaPipe for face detection
• ONNX Runtime for optimized inference (when available)
• JavaScript/HTML for the frontend interface

Troubleshooting

• If you encounter "Module not found" errors, ensure all dependencies are installed
• For Apple Silicon Macs, use the TensorFlow Metal version for GPU acceleration
• If port 8000 is unavailable, modify the port in run.py