Deepfake Detection using Vision Transformers

📌 Project Overview

This project focuses on detecting deepfake images and videos using deep learning techniques.
We use a combination of Convolutional Neural Networks (CNN) and Vision Transformers (ViT) to identify manipulated facial content.

---

🎯 Objective

• Detect whether a video/image is Real or Fake
• Identify subtle inconsistencies in facial features
• Build an automated deepfake detection system

---

📂 Dataset

We used the Celeb-DF (v2) dataset, which contains real and deepfake videos of celebrities.

🔗 Dataset Link: https://github.com/yuezunli/celeb-deepfakeforensics

⚠️ Note:

• Dataset is not included in this repository due to large size (~10GB)
• Users must download it manually and place it in the project directory

---

⚙️ Project Workflow

1. Video Input (Celeb-DF dataset)
2. Frame Extraction using OpenCV
3. Face Detection using MTCNN
4. Preprocessing (crop, resize to 224x224)
5. Dataset Splitting (Train / Test / Validation)
6. Model Training (CNN + Vision Transformer)
7. Prediction (Real / Fake)

---

🛠️ Tools & Technologies

🔹 AI / Machine Learning

• Python
• OpenCV
• MTCNN
• NumPy
• TensorFlow / PyTorch
• Matplotlib

🔹 Backend

• FastAPI (recommended)
• Uvicorn

🔹 Frontend

• HTML
• CSS
• JavaScript

🔹 Development Tools

• VS Code
• Git & GitHub

---