Deepfake Detection Web Application

This web application allows users to upload a video, which is then analyzed to detect whether it contains deepfake content. The app processes the video, extracts frames, and uses a pre-trained deep learning model to evaluate each frame's authenticity.

Features

• Video upload via a web interface
• Frame extraction from the uploaded video
• Deepfake detection on extracted frames
• Results displayed with confidence scores

Requirements

To run this application, you'll need the following libraries installed:

opencv-python-headless
numpy
tensorflow
django

Installation

1. Clone the repository:

   git clone https://github.com/gitone912/deepfaker_deepfake_detector.git
   cd deepfaker_deepfake_detector

2. Create a virtual environment (optional but recommended):

   python -m venv venv
   source venv/bin/activate   # On Windows: venv\Scripts\activate

3. Install the required libraries:

   pip install -r requirements.txt

4. Set up Django:

   • Apply migrations:

     python manage.py migrate

   • Create a superuser for the admin interface:

     python manage.py createsuperuser

Usage

1. Run the Django development server:

   python manage.py runserver

2. Open your web browser and navigate to:

   http://127.0.0.1:8000/
   

3. Upload a video:

   • Click on upload video to upload a video file.
   • The server will process the video, extract frames, and evaluate them using the deepfake detection model.

4. View the results:

   • The results page will display each frame's evaluation, including the confidence score and the overall prediction for the video.

Project Structure

• models/deepfake_detection_model.h5: The pre-trained deepfake detection model.
• media/: Directory for storing uploaded videos and extracted frames.
• app_name/: Replace app_name with your actual Django app name containing:
  • views.py: Contains the logic for video upload, frame extraction, and evaluation.
  • forms.py: Contains the form for video upload.
  • templates/: Directory for HTML templates.

Code Overview

• FrameCapture(path): Extracts frames from the video located at path.
• evaluate_frames(directory): Evaluates extracted frames for deepfake content.
• upload_video(request): Handles video upload and initiates the frame extraction and evaluation process.

Notes

• Ensure the pre-trained model deepfake_detection_model.h5 is placed in the models directory.
• The app is configured to delete and recreate the media directory with each new video upload to ensure a clean processing environment.

License

This project is licensed under the MIT License.

Contact

For any questions or feedback, please contact akash verma on my insta @frozen_blink.