In January 2014, faced with the security issues of face data generated by the generative technology FaceGAN, Facebook's DeepFace and Hong Kong Chinese's Tang Xiaoou Laboratory withstood the pressure and forced the release of their latest progress.
The ideas on both sides are actually similar. Release of technology may face some security issues and doubts, but if it is not released, there will definitely be nothing left.
After all, both sides have made some breakthroughs in terms of performance, and they won't really hold back because of this kind of thing.
Compared with the human eye recognition ability of 97.53% on the LFW data set, the technologies DeepFace and DeepID released by both parties have broken through this limit.
The Facebook team, which originally had only 97.35 on the timeline, improved the accuracy to 97.89% with the help of some of Meng Fanqi's ideas.
Tang Xiaoou's team, which originally had a capacity of 98.52% on the timeline, encountered performance degradation due to its release more than two months in advance, with the release level reaching 97.95%.
However, it still narrowly defeated the Facebook team.
"Facebook's algorithm has more real human image resources than we do. As a laboratory, we have a huge disadvantage in this regard. There is a huge gap in strength between the two sides. However, we still withstood the pressure and overcame the difficulties." Tang Xiaooumian During the interview, I replied like this.
This is the first time that an artificial intelligence method has broken through human levels in the field of face recognition, causing a huge sensation around the world.
At the same time, it has also made the already hot problem of falsely generated faces reach a new high. After seeing these new technologies, the first question everyone wants to ask is whether your latest algorithm can distinguish false faces. Generated faces?
"Using facial images captured by cameras in every corner of the city to compare, automatically identify crimes and send out explosives. Quickly query character information based on gender, age and other attribute expressions, these plots in science fiction movies, Facebook is gradually making them possible. "
Facebook has some press releases to promote its technology, but it seems that the onlookers below are not very fond of this.
"Don't brag, don't brag, we just want to know whether your new algorithm can recognize the fake faces generated?"
“Can your new algorithm handle the security crisis of artificial intelligence?”
"..." Faced with the masses who don't know the inside story, Facebook's technical team is relatively helpless.
Many technologies are not completely safe, but at this point in time, this issue is too hot and too sensitive.
Facebook's publicity strategy is not to answer directly.
But everyone knows very well that not answering is actually an answer in many cases.
In this situation, if this question cannot be answered directly, then the answer to this question is actually very clear.
That is: "Sorry, can't do it."
As for Hong Kong Chinese, because it is within China, the impact of public opinion will be much smaller.
Although there are some relevant reports within China, there has not been such an exaggerated storm of public opinion.
And because Tang Xiaoou's team still beat the Facebook team despite the overall disadvantage, it is generally something worthy of publicity.
To promote our country’s prestige.
Therefore, CCTV News personally conducted relevant interviews and publicity on the matter, and the overall tone was very positive.
"As the creator of the face recognition accuracy record, Tang Xiaoou, a professor at the Chinese University of Hong Kong, led a team that recently defeated the Facebook team and the human discrimination level, allowing the machine to learn to [see everything clearly]. Professor Tang's laboratory has been selected as a global artificial intelligence It is the only laboratory selected in the top ten pioneer laboratories in Asia."
"..."
"Let us ask Professor Tang to talk about the significance of this technology and artificial intelligence to mankind and China."
"The combination of artificial intelligence and industry will become the main driving force for the next wave of industrial upgrading. Accelerating source innovation is the key for China to seize a new round of development opportunities. There is no shortcut in technology research and development. This is a long-term investment marathon. . Only by mastering the technology at the source can we avoid being choked by other countries."
Tang Xiaoou sighed, "As an intellectual, I have enjoyed the country's best educational resources along the way. I should do something meaningful for the country's innovative development, and also strive for the Chinese people in basic research. This is the sentiment that contemporary intellectuals should have.”
The teams on both sides are very aware of the major shortcomings of their current technology, but they have tacitly not mentioned it.
At the same time, the AI camera owned by Meng Fanqi Vision Company has secretly updated a version of its generative image identification function.
Right next to the age and emotion editing functions, there is no special update log or reminder, and the update was launched quietly.
Yang Ming of Facebook DeepFace finally has some free time in the past two days.
Although the face security issue, which has involved well-known media across the United States, has not been resolved, this version was finally released.
As a migrant worker, he also ushered in a long-lost relaxing vacation.
"Why do my classmates keep recommending this AI camera app to me recently?" Meng Fanqi's AI camera app is currently only available in China. Yang Ming has not seen it, but many of his classmates have returned to China and have played with this product.
"What's so fun about a camera, and it's related to my current research direction? Just brag about it." Yang Ming didn't believe this at all. In his opinion, this kind of app just adds some filters, and is different from his own. How can it be possible that highly sophisticated facial recognition is the same thing?
When he saw these messages a few days ago, he simply ignored them. DeepFace was about to be released, and he didn't have much time to watch such boring special effects.
Logically speaking, he wouldn't have watched it after he had some free time, but for some reason today, he picked up his phone and downloaded and installed this app, which is already quite popular in China.
"Age editing... expression editing...?? Is this what a camera app should do???" Yang Ming was stunned after reading it. This technology is much more professional than their Facebook team.
"Distinguish between true and false...? What is discriminating between true and false? Is it the discerning between true and false that I think of?" Yang Ming frowned, thinking no...
But if it doesn’t refer to “determining whether the image is generated by artificial intelligence methods”, how can a normal picture be said to be true or false?
What the hell is this?
"Let me test it." Yang Ming thought, logged into the [none of these people exist] website, and saved a few pictures.
After being transmitted, they are all judged to be [fake faces].
? ? ?
I turned on the camera and took a selfie, which I judged to be a [real face].
what's the situation? It's not messing with me, is it?
Yang Ming thought that if normal people saw this function, they would make some fake generated pictures to test it.
Fake pictures cannot be taken, they must be read from the photo album, while real pictures are likely to be taken directly.
In fact, it does not have this distinguishing function at all, it just mindlessly outputs authenticity based on the photo album and camera.
That must be the case!
Yang Ming thought so and uploaded another photo of himself from before, which was still determined to be [real face].
Now he was completely desperate, and with his final stubbornness, he uploaded another landscape photo.
The result was judged as [shameless].
....Your app is more professional than our dedicated face laboratory team...
Yang Ming felt that he was really shameless now.
