🛡️ Deepfake Detection using AI 🛡️

(Final Submission - KFold Model)

Welcome to our Deepfake Detection Model submission for the competition. Our AI-powered deepfake detection system analyzes images to distinguish between real and fake images with high accuracy. The model is trained using a K-Fold Cross-Validation approach, ensuring robustness and generalizability.

🎯 Objectives

• Detect Deepfake Images:: Accurately classify images as real or fake using advanced deep learning techniques.
• Improve Model Robustness:: Implement K-Fold Cross-Validation for better generalization.
• Optimize for Performance:: Use CNN and data augmentation for improved classification accuracy.
• Provide an Easy-to-Use Solution:: Ensure smooth deployment with a structured codebase.

🔍 Features

• ✅ High Accuracy:: Trained with CNN for superior performance.

• ✅ K-Fold Cross-Validation:: Enhances generalizability by training on multiple data splits.

• ✅ Data Augmentation:: Ensures robustness against varied deepfake patterns.

• ✅ Scalability:: Designed to handle large datasets with batch processing.

• ✅ Optimized Performance:: Uses early stopping and learning rate scheduling for efficient training.

• ✅ Structured Codebase:: Includes a requirements.txt file and a clear setup guide.

📁 Folder Structure

Deepfake-Detection/

│── data/ # Data folder (Test images should be placed here) │── models/ # Saved trained models │── scripts/ # Python scripts for preprocessing & evaluation ├── train_model.py # Training script ├── evaluate_model.py # Evaluation script (to generate predictions) │── outputs/ # Folder for storing generated JSON predictions │── requirements.txt # Dependencies for running the model │── Spades_prediction.json # Final JSON file for submission │── README.md # This file │── Spades_presentation.pdf # Final Report + Presentation (single PDF)

🛠️ Running Requirements

Ensure you have Python 3.9+ installed on your system.

Install Dependencies:

pip install -r requirements.txt

Hardware Requirements:

• CPU: Minimum Intel i5 (Recommended: Intel i7 or AMD Ryzen 7)
• GPU (Optional, Recommended for training): NVIDIA RTX 3060+ (CUDA enabled)
• RAM: Minimum 8GB (Recommended: 16GB+)

📥 Installation & Running the Model

Step 1: Clone the Repository

git clone https://github.com/Mushmat/Predicathon/tree/main cd deepfake-detection

Step 2: Install Dependencies

pip install -r requirements.txt

Step 3: Place Test Images

Ensure all test images are stored in the /data/test/ folder.

Step 4: Run the Model

Execute the evaluation script to generate predictions:

python evaluate_model.py Then Run python generate_predictions.py

Step 5: Check Predictions

The predictions will be saved in the outputs/teamname_prediction.json file.

📚 Model Details

Model Architecture

• Backbone Model: CNN
• Input Size: 32x32
• Loss Function: Categorical Cross-Entropy
• Optimizer: Adam (Adaptive Learning Rate)
• Learning Rate Schedule: Cosine Decay with Warmup
• Regularization: L2 weight decay, dropout
• Data Augmentation: Albumentations library (random rotations, flips, brightness adjustments)

📊 Submission Components

Our final submission includes: ✅ 1. Predicted JSON File (Spades_prediction.json)

• Format:

• [ { "index": "1.png", "prediction": "fake" }, { "index": "2.png", "prediction": "real" } ]

• Stored in /outputs/

• ✅ 2. Final Report + Presentation (Spades_presentation.pdf)

• Includes methodology, preprocessing steps, model details, challenges, and results.

• ✅ 3. Code Repository (Optional)

• GitHub repository for reproducibility.

🚧 Challenges Faced

• Data Imbalance: Adjusted by class weighting and augmentation.
• Overfitting: Reduced using dropout layers, L2 regularization, and early stopping.
• Computation Bottlenecks: Optimized by batch normalization and mixed precision training.

🔮 Future Scope

-🔹 Improve Deepfake Generalization: Train on larger datasets for better generalizability.

-🔹 Enhance Detection with Video Input: Extend the model for real-time deepfake detection.

-🔹 Optimize Model for Mobile Deployment: Convert to TensorFlow Lite for mobile applications.

-🔹 Explainable AI: Integrate Grad-CAM to visualize model decision-making.

❓ FAQs

• 1. How do I run the model on my system?

• Follow the Installation & Running the Model section.

• 2. Can I train the model on my own dataset?

• Yes! Modify train_model.py and provide your dataset in /data/train/.

• 3. What format should the test images be in?

• Images should be JPEG/PNG, and stored in /data/test/.

🙌 Acknowledgements

• 🔹 TensorFlow: Framework used for model development.

• 🔹 Albumentations: Data augmentation techniques.

• 🔹 OpenAI & Research Papers: Reference materials.

• 🔹 GitHub & Community Support: Collaboration and resources.

🔗 Contact Us For queries, reach out via GitHub Issues.

🚀 Spades | Deepfake Detection AI | Competition Submission 🚀