A Flask-powered web application that uses PyTorch, OpenCV, and a custom ResNeXt + LSTM deep learning model to detect face-swap deepfake videos. Users can upload a video, select the number of frames to analyze, and receive real-time predictions labeled “REAL” or “FAKE”, along with confidence scores and visual heatmaps for interpretability. The underlying model, designed by combining RNN and LSTM architectures, achieved 92% accuracy on benchmark test datasets.

• YouTube Walkthrough:
  🎞️ Youtube Video Link
• Hackathon Report (PDF):
  📑 SMART INDIA HACKATHON 2024 – Team Wolfenstein

• 🎥 Smart Video Processing : Upload a video, pick the number of frames, and let OpenCV + face_recognition handle face extraction.

• 🧠 Dual-Stage Prediction: Deepfake + Mask Detection : A pre-trained mask classifier filters masked faces before deepfake detection for better accuracy.

• 🔍 Custom Deepfake Detection Model : A hybrid ResNeXt-50 + LSTM model trained on FF++, CelebDF, and custom datasets with 84–97% accuracy.

• 🧪 Confidence & Calibration : Uses softmax-based prediction scores with Grad-CAM–like heatmaps to highlight manipulated regions.

• 🖼️ Visual & Interactive Results : Displays prediction labels, confidence, cropped faces, and highlighted frames — all in-browser.

• 🌐 User-Friendly Web App (Flask) : Smooth flow from landing → login → model overview → upload → result page — no technical setup needed.

1. Clone the repository

   git clone https://github.com/priyanka350/DeepfakeDetection.git
   cd DeepfakeDetection

2. Install Python dependencies

   pip install -r requirements.txt

3. Download Trained Models

   • Each model (*.pt, ~2.1 GB) is hosted on Google Drive. Download all models here
     
   • Place the .pt files into DeepfakeDetection/static/img/models/ and DeepfakeDetection/models.

4. Run the Web App

   python app.py

   Open your browser at http://127.0.0.1:5000.

1. Open the Deepfake Detection website at http://127.0.0.1:5000.
2. Click Login to access the system.
3. View the model description and project overview.
4. Click Get Started to go to the upload page.
5. Upload a video and choose the number of frames to analyze.
6. Click Submit.
7. View the prediction, confidence score, extracted frames, and detected faces.

📺 Check out the YouTube video for a full demonstration of the project in action.

If you find this project useful, please give it a ⭐ on GitHub and share it with fellow developers!

Email : priyanka.tmsl2022@gmail.com

Made with ❤️ by Priyanka Kumari