Intel Corp. has developed what it claims is the world’s first real-time deepfake detector using artificial intelligence (AI) and capable of catching phony videos with an 96% accuracy rate.
Called FakeCatcher, the deepfake detector can return results in milliseconds. The platform uses Intel hardware and software and runs OpenVino to run AI models for face and landmark detection algorithms.
How it works
Computer vision blocks were optimized using Intel’s Integrated Performance Primitives and OpenCV while inference blocks were optimized using Intel’s Deep Learning Boost. On the hardware side, the real-time detection platform can run up to 72 different detection streams simultaneously on Xeon scalable processors.
FakeCatcher looks for authentic clues in real videos by assessing such factors as blood flow signals that are collected from all over the face; algorithms translate this into spatiotemporal maps. Deep learning then detects whether it is real or fake.
Intel said they developed the tool due to the growing threat of deepfake videos. Companies spend up to spend $188 billion in cybersecurity solutions, but even then, it is still tough to detect these deepfake videos in real time.
Use cases for FakeCatcher include:
- Social media platforms – to prevent uploading of phony content.
- Global news organizations – to avoid amplifying manipulated videos.
- Nonprofit organizations – to democratize detection of deepfakes
