DeepFake Detection

A Python-based deepfake face-swapping toolkit built with TensorFlow and/or PyTorch for creating high-quality synthetic face videos.

🚀 Features

• End-to-end pipeline: Face detection ➜ Preprocessing ➜ Model training ➜ Face swapping ➜ Postprocessing.
• Multiple face encoder/decoder models: Support for autoencoder and GAN-based architectures.
• Flexible training options: Configurable epochs, batch size, learning rate via CLI/flags.
• High-resolution output: Supports HD face-swap videos.
• GPU acceleration: Optimized for CUDA-enabled NVIDIA GPUs.

🧭 Table of Contents

1. Installation
2. Usage
   • Data preparation
   • Training the model
   • Face swap inference
3. Examples
4. Project Structure
5. Configuration
6. Troubleshooting
7. License
8. Contact

---

Installation

1. Clone the repository:

   git clone https://github.com/suyashsachdeva/DeepFake.git
   cd DeepFake

2. (Optional) Create and activate a virtual environment:

   ```bash
   python3 -m venv venv
   source venv/bin/activate

3. Install dependencies:

   pip install -r requirements.txt

4. Download any required pre-trained models (links provided below or pop up in the CLI).

---

Usage

1. Data Preparation

python scripts/extract_faces.py \
  --input_video data/source.mp4 \
  --output_dir data/source_faces \
  --face_size 256

Prepare data from both source (to be swapped in) and target (original video).

2. Train the Model

python train.py \
  --source_dir data/source_faces \
  --target_dir data/target_faces \
  --model_dir models/face_swap \
  --epochs 100

Customize epochs, batch size, learning rate via flags.

3. Perform Inference

python scripts/face_swap.py \
  --model_dir models/face_swap \
  --input_video data/target.mp4 \
  --output_video results/swapped.mp4

Takes the target video and applies the trained face-swapping model.

---

📸 Examples

Include before/after visuals here! Example: results/swapped_example.mp4

---

🗂️ Project Structure

DeepFake/
├── data/
│   ├── source/
│   └── target/
├── models/
│   └── face_swap/
├── scripts/
│   ├── extract_faces.py
│   ├── align_faces.py
│   └── face_swap.py
├── train.py
├── requirements.txt
└── README.md

---

⚙️ Configuration

You can customize the workflow through CLI flags or environment variables:

• --face_size: Size of face crops (default: 256)
• --learning_rate, --batch_size, --epochs
• Paths for input/output and model storage

---

🛠️ Troubleshooting

• GPU errors: Verify CUDA toolkit is installed, and GPU drivers are up to date.
• Poor swap quality: Ensure balanced, varied samples (>2000 per identity) during prep.
• Execution stalls: Try lowering batch size or face size.

---

📬 Contact

For bug reports or inquiries, contact Suyash Sachdeva at suyashsachdeva2403@gmail.com. Also open to pull requests!

---

Happy deepfaking! 🔧🤖

---

🧩 Next Steps (Optional Enhancements)

• Web UI frontend for simplified interaction
• Support for real-time processing
• Integration of voice deepfake capabilities