DeepFake Detection 🕵️‍♂️🧠

A robust deep learning-based system for detecting AI-generated faces using pairwise learning and advanced CNN architectures. This project leverages real vs. GAN-generated image comparisons to effectively distinguish DeepFakes from authentic images.

🚀 Features

• 🔍 Pairwise Learning: Compares real and fake image pairs using a contrastive loss
• 🧬 Custom CNN Backbone: Utilizes DenseNet and CFFN for feature extraction
• 🎭 Multi-GAN Dataset: Trained on fakes generated from multiple GANs like DCGAN, CycleGAN, StyleGAN, etc.
• 🧠 PyTorch-based: Built with modular, scalable PyTorch code
• 📊 Evaluation Tools: Includes accuracy, AUC, precision-recall, and confusion matrix

📁 Project Structure

DeepFake-Detection/
│
├── data/                  # Real and Fake image directories
├── models/                # CNN, DenseNet, and CFFN models
├── utils/                 # Utility functions and dataset loaders
├── train.py               # Training script
├── test.py                # Evaluation script
├── config.py              # Configuration and hyperparameters
└── README.md              # You're here!

📦 Installation

git clone https://github.com/Taskmaster-1/DeepFake-Detection.git
cd DeepFake-Detection
pip install -r requirements.txt

Make sure you have a CUDA-enabled GPU and PyTorch installed.

🧠 Model Architecture

• Input: Pair of images (Real, Fake)
• Backbone: CFFN or DenseNet-based encoder
• Head: Fully connected layers + Sigmoid
• Loss: Contrastive loss / Binary Cross Entropy

🧪 Sample Fake Sources

• StyleGAN
• CycleGAN
• DCGAN
• ProGAN
• Celeb-DF

🙌 Acknowledgements

• FaceForensics++
• Kaggle Deepfake Datasets

📄 License

This project is licensed under the MIT License.