Deepfake app? How can you tell a Deepfake? How do you spot a Deepfake?
These are the questions we recently feel the need to be answered as growing numbers of deepfakes are being crafted to support biased and destructive political agendas and to mould the information in sinister ways, Deepfakes are the new technological war front that will act as a breeding ground for fake news in the coming years.
So by now, we all are familiar with deepfakes you can take a photograph and impose a video over it to make it act and say in the way you want to for example look at this First Order Motion Model for Image Animation
Researchers have now found a way to carefully detect these deepfakes that fool human eyes with the help of an AI. they created an AI to beat another AI. We have entered a new age of internet security and forensics as seen here the AI can identify the deepfakes pretty easily.
shown here are the video samples that are used one is the original and the other is a deepfake and the AI correctly identifies the tampered part of the video, here the red part is the correctly identified tampered area of the deep fakes
so the war is won over the Deepfakes?
NO Because using the algorithm that identifies the deepfakes researchers have found a way to fool it with the help of the addition of noise in the video that will Not fool any human eyes but the AI detective!
This is the AI detecting correctly the deepfake
and this is after the addition of noise, The AI is fooled!
here is how the noise is added to fool the AI, They added noise in every section of the video by learning from the detective classifier such that humans won't notice a thing but the classifier will be fooled
( more in details in the sources )
look at these images the AI detector detects them correctly at first but after the addition of noise the AI cant identify them correctly and is fooled easily into thinking its an ostrich!
Because of the noise added the internal working of the detective AI is now exploited into concluding the bus is an ostrich!
so the war is on and the innovation is leading like always .so always think twice before you see something nowadays check the source to the papers.
check out the playlist of our favourite deepfakes, they are hilarious and confusing on so many levels :
sources :
https://arxiv.org/abs/2002.12749
https://adversarialdeepfakes.github.io/
https://www.youtube.com/watch?v=hFZlxpJPI5w
