🕵️ Deepfake & AI-Generated Face Detection

A deep learning system that detects deepfake and AI-generated facial images using a fine-tuned Xception CNN model trained on 100,000 images.

🏆 Results

Metric	Score
Test Accuracy	99%
AUC-ROC Score	0.9992
F1-Score	0.99
Precision (Fake)	0.99
Recall (Fake)	0.98

🏗️ Model

• Architecture: Xception (Transfer Learning from ImageNet)
• Fine-tuned: Top 20 layers unfrozen
• Dataset: 140k Real and Fake Faces (Kaggle)
• Training: Kaggle Notebooks — Tesla T4 x2 GPU
• Epochs: 13 (early stopping)
• Optimizer: Adam (lr=1e-4)

🚀 Run Locally

# Clone
git clone https://github.com/YOUR_USERNAME/DetectFake.git
cd DetectFake

# Install
pip install -r requirements.txt

# Run
streamlit run app.py

On first run, the model (~100MB) downloads automatically from Google Drive.

📁 Structure

DetectFake/
├── app.py            # Streamlit web application
├── requirements.txt  # Dependencies
└── README.md

🛠️ Tech Stack

• Python, TensorFlow/Keras, Xception CNN
• Streamlit, OpenCV, NumPy, PIL
• Trained on Kaggle (Tesla T4 GPU)

📚 Honours Project

Built as part of a Bachelor of Engineering (Honours) in Computer Science, 2025-2026.