With the arrival of the creation of videos using artificial intelligence known as deepfakes (nothing to do with the videos generated using prompts in tools like Midjourney and more), it is feared that they will be used to generate disinformation. Youtubers like Álvaro Wasabi have made videos in which they put their faces in movies through these deepfakes, and even a Star Wars fan has improved certain scenes from The Mandalorian. But it is feared that it will be used by people with more power and influence, and that it will reach a point where it is not easy to detect it, and for this reason Intel has launched Fake Catcher technology.
As part of Intel's work on Responsible Artificial Intelligence, they have produced FakeCatcher. It is a technology that can detect fake videos in real time with 96% accuracy rate.
Faced with the dangers of deepfakes, it's time to try to detect them
Intel's FakeCatcher uses a detector designed by Demir in collaboration with Umur Ciftci of the State University of New York at Binghamton. It runs on a server and is interconnected through a web-based platform. As for the software, a set of specialized tools form an optimized architecture. The teams used OpenVino to run AI models for face and landmark detection algorithms.
The computer vision blocks were optimized with Intel Integrated Performance Primitives and OpenCV, while the inference blocks were optimized with Intel Deep Learning Boost and Intel Advanced Vector Extensions 512. Then, the multimedia blocks were optimized with Intel Advanced Vector Extensions 2. The teams also drew on the Open Visual Cloud project to provide an integrated software stack for the Intel Xeon Scalable processor family.
On the hardware side, the real-time detection platform can run up to 72 different detection streams simultaneously on 3rd generation Intel Xeon Scalable processors. Most deep learning-based detectors examine raw data to try to find signs of inauthenticity and identify what is wrong with a video. But Intel FakeCatcher look for signs of authenticity in real videos, assessing what makes us human. This would be the subtle “blood flow” in the pixels of a video because when our heart pumps blood, our veins change color. These blood flow signals are collected from all over the face and algorithms translate them into spatiotemporal maps, and using deep learning, could instantly detect if a video is real or fake.
Deepfake videos pose as a growing threat in the wrong hands. Companies are expected to spend up to $188.000 billion on cybersecurity solutions. It is also difficult to detect these fake videos in real time, as detection apps require you to upload videos for analysis and then wait several hours for the results. That time can be vital before a video with harmful information is disseminated before being disseminated in a massive way and being able to classify it as manipulation.
FakeCatcher is being postulated for use on social media sites that could take advantage of it to prevent users from uploading harmful deepfake videos. International news organizations could use the detector to prevent the inadvertent dissemination of doctored videos they receive.
