🎭 DeepFake Video Detection System (CNN + ViT)

A high-accuracy deepfake detection system using Xception CNN + Vision Transformer (ViT), capable of detecting manipulated videos and providing confidence scores with a user-friendly Streamlit interface.

📌 Project Overview

This project detects whether a given video is REAL or FAKE by analyzing spatial and temporal inconsistencies using deep learning.

🧠 Model Architecture

Input Video → Face Extraction (MTCNN) → Xception CNN → Vision Transformer → Classifier → REAL / FAKE + Probability

📊 Datasets

• FaceForensics++ (C23) – Training & Validation
• WildDeepfake – Final Testing

🚀 Technologies

• Python
• PyTorch
• Xception
• Vision Transformer (ViT)
• MTCNN
• OpenCV
• Streamlit
• Scikit-learn

📁 Project Structure

Deepfake-Detection/ │ ├── model.py ├── app.py ├── train.py ├── extract_faces.py ├── xception_vit.pth ├── requirements.txt └── README.md

⚙️ Installation

pip install -r requirements.txt

🌐 Run Web App

streamlit run app.py

📈 Output

• Prediction: REAL / FAKE
• Fake Probability
• Risk Warning

🎓 Author

Devendra Prajapat

📜 License

MIT License