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.
-
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.
-
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.
The workflow involves:
-
Media Upload: Users upload images or videos via a straightforward interface supporting formats like JPEG, PNG, MP4, and AVI.
-
Media Type Detection: The backend identifies the media type and triggers the appropriate processing model.
-
Preprocessing and Data Augmentation: Media files are resized, videos are frame-extracted, and both media types undergo augmentation to ensure reliable input.
-
Model Inference:
- Images: Classified using a CNN.
- Videos: Analyzed using a 3D CNN for deepfake detection.
-
Real-time Results: Confidence percentages are calculated and displayed for both images and videos.
-
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.
- 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)
- 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.
-
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.