Skip to content

🛡 cross internet tracer and deepfake detector for face-swapped media using CNNs, React, FastAPI

Notifications You must be signed in to change notification settings

Spirizeon/streeshield

Repository files navigation

ezgif com-video-to-gif-converter

Stree Shield: AI-driven Deepfake Defense

🌐 Real-time detection of face-swap deepfake videos and morphed images.


Inspiration

With the increasing prevalence of deepfakes, digital media integrity is under threat. Deepfakes are often used for misinformation, fraud, and other malicious purposes. Stree Shield addresses this issue by offering a robust AI/ML solution to detect manipulated media effectively, ensuring digital authenticity and safeguarding public discourse.


How is Stree Shield Different?

  • Dual Media Detection: Supports both images and videos with distinct models for each, enhancing versatility and accuracy.

  • CNN and 3D CNN Models: Utilizes CNN for images and 3D CNN for videos to improve detection accuracy for different media types.

  • Real-time Results: Provides results in under 1 second for images and less than 4 seconds for videos, ensuring rapid analysis.

  • User-friendly Design: Features an intuitive interface with clear visual results and confidence percentages.

  • Multi-lingual Support: Accessible to users in multiple languages, broadening its usability.


Problem it Solves

  • Detecting Deepfakes: Helps prevent the misuse of deepfake technology by accurately detecting manipulated media.

  • Media Integrity: Ensures the authenticity of digital content, crucial for maintaining trust in digital communications.


Approach

The workflow involves:

  1. Media Upload: Users upload images or videos via a straightforward interface supporting formats like JPEG, PNG, MP4, and AVI.

  2. Media Type Detection: The backend identifies the media type and triggers the appropriate processing model.

  3. Preprocessing and Data Augmentation: Media files are resized, videos are frame-extracted, and both media types undergo augmentation to ensure reliable input.

  4. Model Inference:

    • Images: Classified using a CNN.
    • Videos: Analyzed using a 3D CNN for deepfake detection.
  5. Real-time Results: Confidence percentages are calculated and displayed for both images and videos.

Desktop (1) Desktop (2)


Unique Features

  • Fast Detection: Results are quickly generated, suitable for real-time applications.

  • Data Augmentation: Consistent augmentation across images and videos for enhanced model performance.

  • Detailed Confidence Metrics: Provides clear confidence percentages for each detection outcome.


Technologies Used

My Skills

  • Frontend: ReactJS for a dynamic user interface (Akhil & Jayanth), Figma for UI/UX (Jayanth)
  • Backend: FastAPI for setting up communications (Ayush)
  • Models: Tensorflow for model inference (Praneeth & Balaswitha)
    • CNN for image classification.
    • 3D CNN for video analysis.
  • Database: MongoDB for data storage. (Ayush)
  • API: REST APIs for communication. (Ayush & Akhil)

Challenges We Faced

  • Handling large video datasets and real-time analysis.
  • Ensuring accuracy and managing potential prediction glitches.
  • Scaling the system to accommodate various media sizes and formats.

What’s Next?

  • Enhanced Algorithms: Developing more robust algorithms to tackle advanced deepfake techniques and reduce glitches.

  • Scaling Up: Improving processing capabilities for larger datasets and higher-quality media.


Team

About

🛡 cross internet tracer and deepfake detector for face-swapped media using CNNs, React, FastAPI

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published