"Okay." Yang Ming no longer insisted, but he always felt that something was not quite right, and his intuition told him that something bad seemed to be about to happen.

"Meng's new technology is really quite useful!" In the past few days of experimental iteration, Facebook's DeepFace team has made several major breakthroughs.

First of all, the originally cumbersome model size has been greatly reduced, and the speed has also increased significantly due to the simple and easy-to-use operators.

Looking at the model's training log, the algorithm's performance has also been steadily improving, and it is very likely to surpass their previous best results.

"I really don't know how he came up with it." Wolf sighed deeply, a strategy so concise that it could solve a major problem that had plagued the entire field for several years.

Support us at pawread.com .

"In fact, this situation is the most annoying. If he really came up with a very complex theory and operation, it would be fine." Yang Ming could understand everyone's thoughts: "If his strategy is really very complex, we would be completely convinced, and there would be no illusions or regrets."

"But he managed to get things done with such a simple and straightforward method, giving the feeling that it's like Columbus sailing on the sea and discovering a new continent. There may be a lot of wisdom and effort behind it, but it always gives us the illusion that 'I could have done it too.'"

Yang Ming's self-awareness was still clear, knowing that this 'I could have done it too' mentality was just a mirage, and he had not fallen into becoming a keyboard warrior.

Several researchers were quite appreciative of Meng Fanqi's research results, but they had no idea that Meng Fanqi had prepared such a surprise gift for their latest algorithm that had not yet been released.

Facial recognition can be said to be the most widely researched, widely applied, and largest-scale direction in the entire field of computer vision.

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

However, it has always been plagued by an unsolvable problem, which is the issue of occlusion types in images.

Ordinary photos are not layers in Photoshop. If a certain object is occluded, the information in that part is completely lost.

There is no technical means to restore the scene at that time, just like removing a mosaic, it is impossible, the destroyed layer is irreversible.

But now, we have a new way to break new ground and solve this very tricky problem, which is Meng Fanqi's generative adversarial network.

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

But it looks like it's the real thing and does not affect people's understanding of the image.

Moreover, if it has seen many other photos of the target person in advance, the generated part is very likely to be almost identical to reality.

And generation is only the most basic aspect. By understanding the parameters of some dimensions in the hidden layer, people can even finely adjust some attributes of the image.

Editing specific attributes of an image is a more advanced application of generative technology.

For example, in an image, the emotions and facial expressions of the characters can be adjusted according to your preferences.

The later very popular fake smile emoji is created by controlling the emotions of the characters in the image, making a person who was originally sad or angry open their mouth and laugh exaggeratedly.

Some people can even consistently modify the entire video, making the whole scene look very funny and amusing.

In addition to emotional expressions, another later popular application is age adjustment. By modifying certain specific parameters, people can directly turn gray overnight, or even become younger.

These are just ordinary image edits, and only edit the facial part.

There are more advanced algorithm applications later, such as intelligent face swapping, intelligent beauty filters, and so on.

The former is very popular in many film and television derivative works and meme videos, and everyone can 'act' in a three-minute blockbuster.

It is possible to have anyone make a cameo appearance, such as Guan Gong fighting Qin Qiong.

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

"However, due to the current limitations of hardware and other technical means, I still cannot generate very high-resolution clear fake images, and it is a bit difficult to directly manipulate the entire video."

Meng Fanqi clearly knows what he can do at this stage. 4k resolution high-definition images are not possible, but ordinary resolutions of around two to three hundred are almost enough.

It is more than enough for the current stage of facial algorithms.

Moreover, Meng Fanqi now has the research privileges of Google Brain, and it is not a problem to call dozens of graphics cards. The speed of these experiments has increased by tens of times compared to before.

After this model is trained, Meng Fanqi plans to build a website called "[These People Do Not Exist]" to generate hype and make everyone pay attention to this technology of creating fake portraits of people's faces.

Every time you open this website, a lifelike but completely non-existent portrait of a person will be randomly generated.Meng Fanqi was ready to use his relationship with Google and Baidu to promote this website, and in this way, remind people to pay attention to the fact that the identification algorithm must have the ability to distinguish whether the image is real or artificially generated by artificial intelligence.

It could be said that he would kill two birds with one stone, promoting his generative technology and presenting a difficult problem that cannot be solved in the short term for the current direction of facial recognition.

By taking advantage of the year-long time difference, Meng Fanqi, who could easily and properly handle the problem, would form an unbreakable technological barrier and quickly seize the market in the domestic facial recognition direction.

"However, even if this matter is relatively important, it still has to be placed behind the recommended advertising algorithm," Meng Fanqi frowned, knowing that if he wanted to start a business, there would be a lot of trivial matters in the future.

The matter of the recommendation algorithm could not be delayed any longer. It was his biggest money tree at the moment and would also be the core start-up capital source for multiple companies in the future.

He couldn't afford to lose sight of the big picture. Tomorrow, he would go to the Shanghai branch of Google to start the development of this part of the content.

Until the test results were successful, everything else had to take a back seat to it.

Whether it was the FaceGAN facial generation technology, establishing a company, or the subsequent plans to establish the website "These People Do Not Exist" to promote this virtual generation technology, all had to come after the recommendation algorithm.

After all of these tasks were completed, he would find time to discuss the issue of intelligent processing of medical images with the Shanghai Health Center.