Inspiration
- Generative AI has become very good, we are not far from a future where mass misinformation will be a big problem, so we are building a general purpose tech to tackle this!
What it does
- Deepfake detection software to find spoof videos
- Gives Bounding boxes and confidence scores on the videos for real.fake
How we built it
- Implemented SOTA spatio-temporal tranformer encoder to classify real and fake videos
- We can detect face swaps, lip morphed, etc deformations on original videos
Challenges we ran into
- Poor results from some open source deepfake detection projects
- Latency issues with api
- React and Flask connectivity issues
Accomplishments that we're proud of:
-Developed an innovative solution addressing the imminent threat of deepfake technology to information integrity. -Achieved high accuracy in identifying/flagging synthetic media with our spatio-temporal transformer classifier. -Successfully delivered a user-friendly platform that democratizes access to deepfake detection technology.
What we learned
- Implemented a State of the art research paper
- Video data processing
- Created a nice web interface
What's next for FrameForge AI
-Add Voice Cloned audio detection -Add AI generated text detection -Make video detection more robust to other morphing techniques
Log in or sign up for Devpost to join the conversation.