The flower grower also saw the "XX video screenshot", still wondering why this man looked a little familiar, but the author who did not pay much attention to the entertainment circle did not see who this was at the time, which was not august 28, Liu Haoran studio issued an alarm statement!
"Recently, a large number of netizens have spread insulting videos and screenshots of Liu Haoran through online channels and using AI face-changing technology, as well as chat records containing defamatory remarks."
When the content was maliciously disseminated in January this year, the studio had made clarifications and statements, but the rumors still caused a wider spread, with an extremely bad impact. In order to safeguard Liu Haoran's legitimate rights and interests, the studio has reported the case to the police on the 28th.
Perhaps the PS face-changing technology is already very familiar to everyone, but the video technology face-changing technology may not be familiar to everyone, but in fact, the earliest "civilian technology" of AI face-changing originated from the most evil technology in 2017: DeepFake!
<h1 class="pgc-h-arrow-right" data-track="7" > was spurned globally within five days of going online: How evil is black technology DeepFake? </h1>
In December 2017, Reddit's adult communication community suddenly exploded a video that spread across global adult video sites in just a few days, not because of the exaggerated content of the video, but because the protagonist was the popular actress Gayle Gadot.
On June 2, 2017, Wonder Woman was released simultaneously in the United States and Chinese mainland, playing Gayle Gadot as Diana Prince, and although this is not the first time Gayle Gadot has participated in a Hollywood blockbuster, Wonder Woman has given Gayle Gadot's protagonist an aura.
It's obvious that Gail Gadot couldn't have made such a film, either by an actor who resembled Gale Gadot, or someone had perfectly transplanted Gail Gadot's facial features onto the heads of actresses in adult movies! However, within a few days of the film's circulation, Reddit's adult communication community, which was under great pressure, suspended the anonymous account "DeepFakes" that uploaded the video for violating the privacy of others!
"DeepFakes" was extremely dissatisfied with the official ban, and the result was that "DeepFakes" released the software source code for making this film on GitHub in a free and public form, "You can seal me, but the technology has been made public, blossoming everywhere, see how you can seal it?" ”
What exactly does DeepFake work?
The principle of PS seems that everyone can understand, if you want to change your face and change your head, then cut the head out, and then graft it onto the target picture according to the background color and environmental tone, etc., in a little modification, the perfect transition, a fake real PS picture is ready.
But it is obvious that this technology can not be applied to video, first of all, a high-definition high-frame rate movie of 60 frames per second, that is to say, to process 60 pictures a second, 3600 a minute, it is estimated that the brothers of image processing can be crazy, and PS is difficult to modify people's expressions and postures, so it is necessary to constantly imitate the protagonist to take photos from various angles.
Of course, DeepFake does not use this low-level technology, its process is to model the key features of the face, the face is digitized, such as the current popular face payment is based on the key features of the face is identified, although some people look very similar, but the key size is still relatively large differences.
Modeling is learning, modeling only confirms the facial features, but people are alive, such as the shape of the face is not the same when speaking or moving, just like laughing, some people like to laugh, but some people just smile, so through a large number of target people's videos, you can sample the characteristics of this person's various expressions, of course, you can also make up for other features that have not been collected according to the public database, but there is no doubt that the more you learn, the more realistic it will be.
Finally, there is dynamic splicing, such as the difference between the face shape of the "actor" and the face shape of the target group, then it is necessary to dynamically stitch the model of the target person into the "actor" after modification, and the transition must also be natural in order not to be seen.
Of course, the DeepFake technology of early videos is still relatively rough, especially in some shoddy videos, the excessive parts of the face and body are very unnatural, and even when they are covered by hands, you can still see the face, but in the later stage, the level is getting higher and higher, and it can even be done to the extent of fake reality.
What exactly is DeepFake used for?
This technology is of great significance to the film industry, in fact, long before the popularity of "civilian" DeepFake technology, Hollywood CG technology has long been applied similar technology, "Fast and Furious" protagonist Paul Walker died in a car accident in Los Angeles at 3 p.m. on November 30, 2013, but at this time, "Fast and Furious 7" although most of the shots have been filmed, there are still partly based on Paul Walker's two younger brothers, Cody Walker and Caleb Walker. It was later treated similarly into Paul Walker.
The shot of Paul Walker driving farther and farther down a fork in the road at the end doesn't know how many Fast and Furious fans cried! The song "See You Again" has therefore become popular all over the world, including the author, and I like it very much, and I will turn it out to listen to it when I have time.
The blessings of the film and television industry may turn into disasters
DeepFake changed its face at will, making your videos fake and real, after being released on GitHub, the software script was continuously improved by global enthusiasts and professional structures, and the fake videos produced became more and more realistic, even familiar friends could not distinguish, what would happen?
It's okay to make a joke with a friend, but what would be the result if you grafted the face of a public figure onto an obscene video, or a leader's face on a video that speaks at an important event but has a completely opposite attitude, especially when there is unrest in a certain country?
Einstein's speech on the coronavirus
The former will waste a lot of public resources to investigate whether such videos are true or false, while the latter may trigger riots, and even trigger a revolution of "XX flowers" or "XX velvets" in a different world, and the enthusiasm but the unknown masses are the most likely to be instigated, and the probability of this is very high.
<h1 class="pgc-h-arrow-right" data-track="8" > After 4 years of development, how far has AI face-changing technology developed? </h1>
How far has this technology evolved? In fact, everyone is watching every day, the live broadcast of all the Internet celebrities, all of them are beautiful women, and the live broadcaster who has skin like gelatin and is beautiful as a flower may be a big mother.
The most famous is the Qiao Biluo overturning incident, enough horror, can let people pay a hundred thousand, occupy the first brother of the list, it is too sad, can tip out so much money fans, should be a study of His Highness Qiao Biluo, how did not see whether this is true or false?
Since then, there have been more and more incidents of flipping in the live broadcasting industry, some because they forgot to open, such as Qiao Biluo, and some are unable to do the AI face-changing algorithm at a certain angle, so the original form is exposed.
But in the end, there is one, technology can be compensated, with the hardware technology and algorithms more and more powerful, the possibility of future errors will be lower and lower, is not very scary? Even if it's a direct video, you can't tell if the person across from you is your friend or your family, and you may have to agree on a code word to distinguish whether it is a friend or not.
<h1 class="pgc-h-arrow-right" what to do after data-track="53" >? Can it still be distinguished? </h1>
This is a very acute problem, related to the basic trust between people in the future society, if not solved, then the future can really only return to the era of face-to-face solutions.
Of course, scientists are not idle, since AI technology can do face change, then the same can also do recognition, the algorithm can not do exactly the same as natural people, in order to be fast and natural, it must miss some details, so this is the KEY to the AI algorithm to crack the face change.
In June 2019, researchers at the University of California, Berkeley and the University of Southern California took advantage of this difference, building a database of target person characteristic data and then comparing it with AI face-swapping videos, with a recognition rate of up to 92%, but this technology has a fatal problem, if there is no database of target people, then its recognition rate will drop sharply, especially in some deliberately blurred videos.
The other is that in August 2020, Ali Security Turing Laboratory and the Institute of Computing of the Chinese Academy of Sciences cooperated to develop a technology, this recognition threshold is relatively low, just from the video can identify whether it has undergone an AI face change, because after the AI face change will be cut into the face because of each frame of different positions, this is dynamic, and flashed, it is difficult for the human eye to identify.
But the computer is very good at dealing with this kind of thing that needs to be compared with each frame, this recognition does not need to be learned, does not need modeling, basically can be developed into an APP, the future mobile phone will definitely appear on an AI recognition is a real person APP software, interested friends can develop, maybe it will sell well!
Read more: Harvey the Crosser
On February 13, 2021, a Spanish man named Xavi posted his first video on social media, and since then the video has grown more and more, and the content has become more and more terrifying.
Xavi claims that he has come to Valencia, Spain in 2027, but this empty city uses the same time and space as Valencia in 2021, which means that he can see everything in Valencia in 2021, but no one.
Many friends are convinced of this, such as he entered the empty Apple store to take away the mobile phone, entered the police station to drive away the police car, etc., is Valencia in 2027 really become such a world?
In fact, there are great flaws here, first of all, some netizens found that the hand that took away the mobile phone and drove away the police car is not the same, so the question is, how is the empty street discharged?
In fact, it is very simple, we can take a picture of the street without people, using the superposition technology! Using AI technology to remove these moving people is much simpler than face-changing technology, one is dynamic transformation, and the algorithm that "deletes" people from the video is much simpler to stitch with the video!
So Harvey is just a liar, of course, everyone is also a hilarity, but it is just entertainment!