After the groundless claims that grew like an avalanche, social media is here with another deepfake rumor. But this time, we will look at how tools that eliminate this uncertainty work.
Princess of Wales Kate MiddletonConcerns increased when he was not seen in public after his abdominal surgery. Not hearing from her for weeks, manipulated photos being presented as real as concerns increased, and finally, a video whose authenticity was questioned saying “The Princess is fine.” Sharing it as “triggered conspiracy theories.” Fortunately, everyone was relieved when Catherine was spotted outside.
But the events are not over. Kate Middletonmade a statement on the official social media accounts of the Prince and Princess of Wales and said that he was diagnosed with cancer. stated. Catherine, he said, is in the early stages of chemotherapy treatment. But rumors circulating on social media say that this video is a deepfake and made with artificial intelligence It was in the direction.
Deepfake detection sites showed that the video was original. So, how do these sites make their decisions?
The most popular among these sites is undoubtedly deepware. The application, which assigned a deepfake score to the video in question using different models, claimed that there was no Deepfake in the Kate Middleton video in question. There are four different models on which this result is based. These; Avatarify, Deepware, Seferbekov and Ensemble It is mentioned as . Apart from this, when we look at the result screen, we can see both video and audio details.
Avatarifyironically actually create deepfake videos provides. Using machine learning, the application adds animation and sound to a photo you give it and animates it. Its biggest convenience is that it processes videos directly on the mobile device, not in the cloud. Its knowledge in the field of deepfakes is also used in deepfake detection on Deepware.
Deepware Scanner, which has been used since 2018, is known for constantly researching different methods.
From convolutional neural networks architecture EfficientNet-B7 Deepware Scanner, which works on the model, offers high accuracy with low resource usage. In addition, it contains 120,000 video data. CFDC data set uses. These videos include examples from 4chan Real, Celeb-DF, YouTube and many other platforms. In this context, since it feeds on both organic and live videos, it becomes easier to distinguish deepfakes.
Let’s also take a look at Seferbekov and Ensemble.
Developed by Selim Seferbekov, Seferbekov evaluates videos by examining them frame by frame. Consisting of 3-stage neural networks MTCNN facial recognition system, is found in Seferbekov’s model. With this system, Seferbekov boxes the faces in the videos, extracts them from the videos, determines the outline of the face, then extracts SSIM masks for comparison and performs the comparisons. Seferbekov’s Using these values, Deepware produces and presents its own results.
According to the information on the site, Ensemble is a combination of Deepware’s browser and Seferbekov’s browser. joins forces.
Let’s sum it up in a single denominator.
Although the systems and the solutions they offer vary, it is stated that the most effective are convolutional neural networks that help distinguish real images from fake ones. Apart from this, although the methods used in the systems may differ, they are all essentially without deep learning Let’s say you benefited from it. Deepfake, on the other hand, is created with machine learning.
Although it may seem simple to our eyes, there are different processes and models behind the system in question. For videos you are suspicious of, you can reach Deepware from this link.
Let’s take a look at those who confuse the concepts of machine learning and deep learning:
RELATED NEWS
Key Differences Between Artificial Intelligence, Deep Learning and Machine Learning Concepts That Are Constantly Confused With Each Other
“Okay, these systems can understand deepfake in this way, but how can we understand it?” Let’s take those who think like this:
RELATED NEWS