"Okay." Yang Ming no longer insisted, but he always felt that something was not quite right, and his intuition told him that something not very good was going to happen.

"Meng's new technology is really easy to use!" In the past few days of experimental iterations, Facebook's DeepFace team has made several big leaps.

First of all, the size of the originally bloated model has been greatly alleviated, and the operator is much faster because it is simple and easy to use.

Looking at the model's training logs, the performance of the algorithm is also going all the way up, and it is likely to surpass their previous best results.

"I don't know how he came up with it." Wolf let out a long sigh, such a simple operation could solve a big problem that had plagued the entire field for several years.

"In fact, this situation is the most annoying, and it would be nice if he really came up with a very complex set of theories and operations." Yang Ming can understand everyone's thoughts very well: "If his strategy is really complicated, we will be completely convinced, and we will not have any illusions and regrets." ”

However, he did it in such a simple and simple way, it felt like Columbus had discovered the New World when he drove all the way to the sea. There may be a lot of wisdom and hard work in it, but it always gives us the illusion that I have a chance if I can't do it. ”

Yang Ming's self-understanding is quite clear, knowing that this idea of 'I can do it on the road' is just a mirage, and he has not degenerated into a keyboard warrior after all.

Several researchers still recognized Meng Fanqi's research results, but they didn't expect that Meng Fanqi had prepared a surprise gift for their latest algorithm that had not yet been released.

Face recognition can be said to be the direction with the most researchers, the widest range of applications, and the largest scale in the entire field of computer vision.

As an ancient topic, the task of face recognition has gone through the development of traditional pattern recognition to modern deep AI, and has become quite mature.

Throughout, though, it was plagued by an unsolvable problem, and that was the type of occlusion in the image.

Ordinary photos are not Photoshop layers, and if an object is obscured, then this part of the information is completely lost.

There is no technical means to restore the scene at that time, like going to mosaic, it is impossible, and the damaged layer is irreversible.

But now, we have a new way to solve this very difficult problem, and that is Meng Fanqi's generative adversarial network.

Through the repeated confrontation generation of large-scale data, the generative network will have a very terrifying image generation ability, and can completely produce content very close to reality to fill the occluded part, although the generated content may have nothing to do with the original content.

But it does look like that, and it doesn't affect people's understanding of the image.

What's more, if it had seen quite a few other photos of the target beforehand, it would most likely have produced something that was almost exactly the same as reality.

Moreover, generation is only the most basic aspect, through the understanding of the parameters of some dimensions of the hidden layer, people can even fine-tune some properties of the image.

The editing of specific attributes of an image is a more advanced way of applying generative techniques.

For example, in the image, the emotions of the characters, the joys, sorrows, and sorrows of the facial expressions can all change accordingly according to your adjustments.

Later, the very popular fake smile meme is to control the emotions of the face in the image, so that the originally sad and angry person grins and laughs exaggeratedly.

Some people will even modify the entire video coherently to make the whole scene look very funny and hilarious.

In addition to facial expressions, another popular one is age regulation, by modifying the parameters of certain dimensions, people can directly turn white overnight, or rejuvenate.

These are just normal image edits, and only the parts of the face are edited.

There are more high-end algorithm applications in the future, such as intelligent face swapping, intelligent beautification and so on.

The former is very popular among many film and television second-generation creators and ghost animals, and everyone can 'act' in a three-minute blockbuster.

Whoever you want to make a cameo can make a cameo, and it is possible for Guan Gong to fight Qin Qiong.

The latter is widely used on various live broadcast platforms, and throwing two slippers on the ground can automatically apply makeup to you.

"However, due to the limitations of hardware and other technical means, I have no way to generate a very high-resolution clear fake image, and it is a little reluctant to directly manipulate the entire video."

Meng Fanqi clearly knows what he can do at present, there is nothing he can do with a high-definition picture with 4K resolution, but the ordinary definition of about two or three hundred or even five hundred is almost enough.

For the face algorithm at this stage, it is more than enough.

Moreover, Meng Fanqi now has the researcher authority of Google Brain, and it is not a problem to call dozens of graphics cards, and the speed of doing these experiments is dozens of times faster than before.

After the model is trained, Meng Fanqi plans to build a website called "These People Don't Exist" to stir up the heat and draw attention to this kind of face forgery generation technology.

Every time you open this website, a portrait of a person is randomly generated, but it is actually completely non-existent.

Meng Fanqi plans to use his relationship with Google and Baidu to promote the website, as a way to remind people that the original recognition algorithm must be able to distinguish whether the image is real or AI-generated.

It can be said that it kills two birds with one stone, that is, it advertises its own generative technology, and puts forward a big problem that cannot be solved in a short time for the current face recognition direction.

With the help of the time difference of about a year, Meng Fanqi, who can easily and properly deal with this problem, will form an unbreakable technical barrier and quickly seize the domestic market in the direction of face recognition.

"But even if this matter is more important, it still has to be ranked behind the recommendation advertising algorithm." Meng Fanqi frowned, if you want to start a business, there must be a lot of trivial things in the future.

The recommendation algorithm can not be delayed any longer, it is currently the biggest cash cow for itself, and it is also the core source of start-up capital for many companies in the future.

I can't lose a big one because of a small one, and tomorrow I will go to Google's Shanghai branch to start the research and development of this part of the content.

Until the test results are successful, all other things have to give way to it.

Whether it's FaceGAN's face generation technology, the establishment of a company, or the follow-up plan to build a [none of these people] website to promote this virtual generation technology, it has to be ranked behind the recommendation algorithm.

Wait until these contents are completed, and then take the time to go to the Shanghai Health Center to discuss the issue of intelligent processing of medical images.