FakeCatcher by Intel

Description:

Intel’s deepfake detector examines “blood flow” within video pixels to deliver results quickly with 96% accuracy [15]. By analyzing natural indicators in real videos—such as slight color changes in facial veins from blood flow—this system captures signals across the face. Algorithms then convert these signals into spatiotemporal maps, which deep learning models use to identify if a video is genuine or manipulated. [17]

 

Advantage:

A real-time deepfake detector with 96% accuracy, delivering results within millisecond [15]

 

Limitations:

In BBC experiments, as indicated in [16], when tested on real, authentic videos, the system began to encounter issues.

On multiple occasions, it incorrectly identified genuine videos as fake [16].

System also does not analyze audio when checks the fake cues in video [16]

System is able to work at its best when analyzing full frontal face portraits [17]

 

References:

[15] Intel introduces Real-Time Deepfake Detector. (2022, November 15). Intel. https://www.intel.com/content/www/us/en/newsroom/news/intel-introduces-real-time-deepfake-detector.html#gs.ipro6g

[16] Clayton, B. J. (2023, July 22). Intel’s deepfake detector tested on real and fake videos. https://www.bbc.com/news/technology-66267961

[17] Powell, N. (2022, November 17). Intel unveils new FakeCatcher platform to tackle deep fake videos. Techreport. https://techreport.com/software-news/intel-unveils-new-fakecatcher-platform-to-tackle-deep-fake-videos/

 

Tool For:

Video


How to Access:

Runs on a server using Intel hardware and software, and interfaces through a web-based platform.

 


Resources:

Resource 1

Resource 2

Resource 3


Last Accessed:

11/16/2024

NSF Award #2346473