My technology affects the whole world

How much? 99% and 1 second? Zheng Shijian was desperate when he heard this data.

The accuracy of the algorithm made by Hanwang Company cannot even reach 50%, and it also requires a person to look directly at the camera, and the recognition speed also takes 5 seconds to 10 seconds.

Compared with Ma Tian's, it's not an exaggeration to say it's garbage.

From the hard work in 2003 to now, the results of 8 years are all rubbish!

Seeing this, Zheng Shi felt like his energy had suddenly been drained out of him. He stooped and pursed his lips. What he wanted to say finally turned out to be: Thank you, Mr. Ma, for informing me. I don't have any questions. I'm sorry to bother you today!

After finishing his weak words, Zheng Shi stood up and was about to leave.

Ma Tian watched all this, but his eyes were flashing.

He thought for a while, then stopped Zheng Shijian from leaving and said, Don't leave in a hurry. Tell me how far facial recognition technology has developed at home and abroad!

Zheng Shijian looked at Ma Tian, ​​then returned to his original position and said:

Dr. Shasuha from abroad proposed face recognition and rendering technology based on commercial images in 2001. The definition of face signature images under various lighting conditions can be used for face recognition under constant lighting conditions. , making people aware of the research on light image algorithms for face recognition.

Later, Dr. Baslay and Dr. Jacobs used spherical harmonics to represent illumination and used a convolution process to describe reflection. They proved that: the set of all Lambert reflection functions obtained from any far point light source forms a linear space! Their proof provides ideas for solving the problem of face illumination transformation.

Now we basically continue to optimize the accuracy of face recognition under different lighting based on these two bases.

The currently leading international technologies are Cognitec and Idetiix. They are based on rectangular feature and AdaBoost feature detection and can achieve an accuracy of 80%!

Zheng Shijian told the story clearly, and he obviously has deep research on international facial recognition technology.

Ma Tian nodded with satisfaction: What about your Hanwang company's research?

Zheng Shijian:......

Zheng Shi saw that he was silent. As the saying goes, hitting someone is not a slap in the face. What is the difference between Ma Tian's move and a slap in the face.

However, he still answered truthfully: Now we have relatively little research on face recognition algorithms. It is not live detection yet, and the accuracy is about 30%!

This time it was Ma Tian's turn to be silent. He didn't expect that the country was so good. In addition to being a little behind other countries, this country is a bit behind other countries. Everyone has uploaded it to the point where the algorithm has been implemented, but domestic research is still just beginning.

Seeing Ma Tian's silence, Zheng Shijian finally couldn't help but ask his doubts: Mr. Ma, I want to ask, how did you overcome so many technologies at once and achieve 99% accuracy?

It's not that he doesn't believe Ma Tian, ​​but it's really a bit outrageous. Technology must be accumulated from generation to generation, and it is not so easy to reach the top in one step.

He didn't know that Ma Tian was a loser. Level 3 mathematics had almost sent the final version of the face recognition algorithm to Ma Tian's mind. Ma Tian could write it out just by looking at the face pictures and making associations.

Because the technology I use is deep learning, or neural network! Ma Tian replied.

Deep learning is familiar to many people who are engaged in AI. It is a complex machine learning algorithm that achieves results in language and image recognition that far exceed previous technologies.

In short, it is like a neural network given to a robot, with the ability to learn and summarize. For example, for a face picture, it can allow the machine to divide it into three layers like a person to speed up recognition.

When we normal people recognize a face, we first look at the facial features, right? Secondly, we may look at the hair and skin wrinkles, and finally we judge whether it is a human face through the body below the head.

Machines use multi-layer neural networks for deep learning and can generally have this ability.

Are you talking about what the machine learning master and Toronto professor Kilfry published in the journal Science in 2006? Zheng Shijian asked.

Ma Tian nodded. He did not expect that the concept of machine learning had been proposed abroad, but before Zheng Shi met, he must pretend to know about this paper.

However, Zheng Shijian knew a lot. It seemed that he did have a deep research on facial recognition technology. Ma Tian thought.

Zheng Shijian was silent. It turned out that this was the problem. In the final analysis, Hanwang has done too little research, and has not carefully studied so many foreign technology publications.

Thinking of this, Zheng Shi couldn't help but look at Ma Tian with gratitude. At least Ma Tian pointed out a correct path to catch up.

Thank you, Mr. Ma, for informing me. If not for your reminder, we would have taken many detours! Zheng Shijian said sincerely.

Ma Tian smiled and didn't say much. It wasn't that he looked down on Zheng Shi for meeting them, but that the foreign founders were just getting started with machine learning.

He didn't think that with his reminder today, Zheng Shi would have a big breakthrough when he returned.

Seeing Zheng Shijian leave in excitement, Ma Tian also showed a meaningful smile.

He thinks Zheng Shijian is a talent, and since he is currently selling facial recognition doors and is short of a R\u0026D manager, he thinks Zheng Shijian is a good candidate.

Now let Zheng Shijian go back and hit the wall, and then recruit him to sell facial recognition automatic doors and engage in machine learning. Ma Tian thinks this idea is very good.

The field of artificial intelligence algorithms requires long-term research and accumulation. It is definitely unrealistic to rely on him alone. After all, just typing code and making wheels is a huge amount of work.

When Zheng Shi saw that it was exactly as Ma Tian expected, he was so excited that he immediately took a plane and flew back.

When he returned to Hanwang Company, he organized meetings with his team without stopping.

I went to find Ma Tian! His first words from his seat at the top attracted everyone's attention.

Ma Tianzhen has really developed our ideal product. Its accuracy is 99% and its recognition speed is about 1 second! Zheng Shi continued without stopping.

As soon as this data came out, the reaction of the people below was exactly the same as the answer he heard from Ma Tian at that time, and he immediately became desperate.

How is it possible? 99%, 1 second, wuwuwu—— Someone is in pain.

It's over, it's all over!

…

The current international leading level cannot reach this data. Tang Jiefei looked unconvinced and stood up to question. However, when he thought that Zheng Shijian could not joke and lie, he immediately put on a mask of pain and became paralyzed in despair. Slumped in the seat.

Don't give up, everyone, Ma Tian told me what method he used! Zheng Shijian said in time, one blow is enough, but he can't let his team despair.

What method did he use?

Deep learning! Zheng Shijian said the answer. Seeing that many people had not been exposed to this, he explained, This is the 2006 machine learning master and Toronto professor Kilfry published in the journal Science A method, and Ma Tian applied this method.

After the conference is over, I will post this paper in the group for everyone to study it carefully and see if we can also figure out a face recognition algorithm of our own.