FakeCatcher, Intel’s new deepfake-detecting platform, claims real-time deepfake detection with 96 % accuracy. And the key is our blood.
A brand new deepfake detector that claims nearly fast, extremely correct outcomes has been revealed by Intel, utilizing what the corporate believes is a way more dependable technique than reverse engineering. What FakeCatcher does otherwise, which Intel believes makes it extra distinctive, is to take a look at the video itself slightly than the uncooked information behind it.
This new strategy analyzes potential deepfakes in real-time, utilizing numerous instruments like face detection AI fashions and algorithms. It additionally appears to be like for (and may allegedly detect) “blood movement” by delicate modifications in pixels, which happen as our blood naturally flows. These pulses are picked up by FakeCatcher, translated into seen maps, then checked for inconsistencies, all inside a really brief time.
Actual-time evaluation is one other side Intel believes will make a giant distinction, as different deepfake detectors can take hours to course of and sometimes require importing questionable media first.
Intel hopes that FakeCatcher will assist forestall companies, social media platforms, and particular person customers from falling for deepfakes sooner or later, whether or not it is by stopping deepfakes from being uploaded or flagging these movies earlier than they’re probably shared.
Intel Senior Employees Analysis Scientist, Ilke Demir, will share extra details about FakeCatcher and its potential in a Twitter Spaces event this Wednesday, November 16, at 2:30 pm ET.