DeepFace

GIF
Tested Data
Landing page
Landing page-2
home page
Video Stream Protection
Parameter configuration

🌟 Inspiration

The rise of deepfakes has created a pressing issue for online creators, celebrities, and brands, with malicious users exploiting their content for fraud, scams, sexual content, and false advertising. Inspired by the need to protect digital identities and restore trust in online media, we developed DeepFace—a tool to combat deepfake misuse and empower creators to safeguard their content.

🎥 What it does

DeepFace reads a video stream, adds perturbations to prevent deepfake manipulation, and returns the protected stream to OBS for seamless broadcasting. By subtly altering the video frames in real-time, it ensures that the content cannot be easily exploited by deepfake algorithms while maintaining visual quality for viewers.

🛠️ How we built it

We implemented a minimalist face-tracking system to identify hash locations within video frames, ensuring precise targeting of key areas. Using the PGD (Projected Gradient Descent) attack method, we applied perturbations to the original frames across different machine learning models, making them resistant to deepfake manipulation. Finally, we integrated the solution into OBS using the OBS WebSocket API, allowing creators to stream protected content directly from their broadcasting software.

⚠️ Challenges we ran into

Generating a hash for each frame took a significant amount of time, slowing down the process. To address this, we reduced the video resolution to decrease runtime, though this impacted quality. In the future, we plan to identify frame sequences with minimal differences and generate a single hash for those sequences to optimize performance without sacrificing accuracy.

😀 Accomplishments that we're proud of

This tool not only enhances the safety of individual creators but also strengthens the integrity of digital ecosystems, encouraging brands to invest in safer online spaces and helping platforms comply with emerging regulations on deepfake content.

📚 What we learned

Throughout the development of DeepFace, we gained valuable insights into the complexities of real-time video processing and the trade-offs between performance and quality. We learned how to balance computational efficiency with effective deepfake protection, optimize machine learning models for practical use, and integrate seamlessly with existing creator tools like OBS. This experience also deepened our understanding of the evolving deepfake landscape and the importance of adaptive solutions.

🚀 What's next for DeepFace

Looking ahead, we aim to integrate DeepFace into mobile devices for broader accessibility, transform it into an API for easy adoption by platforms and developers, and generalize our implementation to support a wider range of machine learning models, ensuring robust protection against evolving deepfake technologies.