DeepDefend
Unmasking Deception, Shielding Truth.
The problem DeepDefend solves
Deepfake technology poses significant risks to media authenticity and personal security. our solution provides detection and protection against deepfake using AI, User Validation and Decentralized Technologies, ensuring trust in digital content.
Challenges we ran into
- Integrating singleton pattern for audio and video protection. (CUDA)
- Deployment of models locally and cloud source providers.
- Connecting virtual machine with local computing power to run complex models.
- Handling file uploads, temporary storage and ensuring files were properly cleaned up properly after processing.
- Long periods of waiting time for train models using the dataset.
- Faced issues with deciding what layers have to be used in designing ML models. (GEMM errors)
- Designing a Smart Contract that could handle image uploads and more importantly, secure management of votes were challenging,
- Mismatch of image and masked shapes, especially when one has fewer channels than the other (width and height parameters) while training the "Image Protection Model".
- Keeping the frontend synchronized with blockchain, for example, ensuring vote count displayed on the frontend reflected the actual count on the blockchain required careful handling of state updates.
- Achieving a satisfactory level of accuracy while training Audio files. (Detection of deepfake audio)
Tracks Applied (2)
Polygon Track
Polygon
Ethereum Track
ETHIndia
Discussion
Builders also viewed
See more projects on Devfolio