Meng Fanqi and Wang Kai have communicated a lot during this period and have become quite familiar with each other.
However, when Meng Fanqi saw this sentence, he still felt that it was quite nonsensical: "You are a muscular man about 1.8 meters tall. Why should I take you away? Should I take you away or should you carry me?"
"Brother has been arranged to come to you again." Wang Kai said as he sent Meng Fanqi a photo of the airport: "I just arrived at the airport. The evening flight goes to Shanghai."
"What's going on?"
"Isn't it because the leader is worried? He doesn't dare to ask you to come to Yanjing, so he can only send me to Shanghai." Wang Kai couldn't help himself. Although he had just made a contribution, he still felt that there was no improvement in his work. .
"The leader said that you want me to understand this part of the code by your side, and then go back and explain it to everyone."
"It's still the same job as last time." Meng Fanqi understood that he was not a white person after all. Robin Li was eager to get government projects, so he would definitely need to send someone to hand over the code to him.
Just looking at that last look, I guess I am worried. Therefore, he sent Wang Kai, who had been communicating with him before, to follow him and seize every minute to fully grasp this part of Meng Fanqi's new development.
"Okay, that's okay. Let's meet at the Baidu branch near Hechuan Road tomorrow."
"No problem, brother." Wang Kai agreed readily: "Brother, if you need someone in the future, you must think of me. No matter what time or place, brother will definitely do it with you."
"What do you mean by taking me away for a long time?" Meng Fanqi finally came to his senses.
"That's for sure. There's definitely no big future in working part-time. You can't save much money in a lifetime if you earn a dead salary safely." Wang Kai didn't hide it, and directly expressed his truest thoughts.
"I don't remember where I read it. It says that a person can encounter at least one opportunity to change his or her destiny every day in his life, but most of the time I don't know it at all."
"It depends on how you change your life. Being hit by a dump truck when you go out also changes your life." Meng Fanqi has heard another version: "The version I have heard is that a person will definitely encounter someone in Silicon Valley throughout his life. An opportunity to get rich overnight, commonly known as five ones."
"I think you are my opportunity. Brother, if you really want to start a business in the future, you must call me. I will definitely do my best to be a cow and a horse, and I will die with all my heart." After experiencing these two technological shocks from Meng Fanqi, Wang Kai has completely Impressed by his technical prowess.
Many times the question of an algorithm is not whether it can be written, but whether it can be imagined.
In the past few days, Baidu AI Research Institute has recalled a group of technical staff who were traveling abroad, hoping to gradually shift its focus back. But after almost a week of busy work, there has been absolutely no meaningful progress in multi-target detection and tracking.
Although Baidu AI's business has been booming recently and the situation is great, many internal AI technicians are actually quite critical.
The YOLO algorithm is very strong, but it was not independently developed by Baidu at all. It just borrowed the power of Meng Fanqi to build a large-scale implementation project.
It is indeed a rapid progress in the short term. If you look at it in the long term, it is not a good thing. The accumulation of real technology is not enough.
Especially when too many projects and clients came in at once, researchers were greatly distracted. Many people complained privately that they were doing things other than their responsibilities.
The new demand from the government is a good example. Baidu, who lacked time to accumulate technology, was at a loss for a while, so he had to turn to Meng Fanqi to fulfill this demand.
Wang Kai felt that instead of going to Meng Fanqi repeatedly, he might as well just go and settle with him.
Based on his perception of Meng Fanqi's technical strength, making a lot of money by starting a business is just a matter of how much you make, and there is no possibility of failure at all.
After listening to Wang Kai's series of inside stories, Meng Fanqi was actually quite able to understand that many of the management's ideas and decisions were indeed not very friendly to low-level employees.
Because they do not, and cannot really consider issues from the perspective of low-level employees. To put it bluntly, others don't care at all about the thoughts of wage earners.
"I really don't have any idea of starting a business in the short term. I may have a plan in two years." Meng Fanqi can earn enough wealth through technology sharing alone, and supplemented by some investments, this will be much easier than starting a business.
If he really wants to start his own company, Meng Fanqi currently has only one plan, which is to leave Google with Ilya at the end of 2015 and start OpenAI together in two years.
But even this matter was not something he had thought through, because in the planning, Meng Fanqi would publish the core technologies needed for ChatGPT 15 years ago, such as the attention mechanism and the generative pre-training Transformer.
At that time, maybe just invest in OpenAI based on technology and capital, and become a technical consultant. There is no need to start a business by yourself.
"Let's talk about it then. If I make any moves, I won't hide it. If you see it then, remember to contact me."
Someone recognized him so early, and Meng Fanqi was willing to let him join the team. After all, he was also Baidu's AI engineer, so there was no problem with his level.
It's just that I really don't have this need in the short term. Even if I did, I wouldn't be too embarrassed to directly poach the people Mr. Li sent to connect with him.
After the brief phone call, Meng Fanqi began to search for the Hungarian algorithm and the Kuhn-Mankels (KM) algorithm, both of which are actually old methods.
For example, the Hungarian algorithm was proposed in 1955. It is a combinatorial optimization algorithm that solves allocation problems in polynomial time.
The SORT multi-target tracking method only uses these two, as well as YOLO proposed by Meng Fanqi, and does not require many changes to the three themselves.
The YOLO algorithm is used as a feature extractor to extract all vehicle parts.
The Hungarian or KM algorithm performs maximum matching on several targets between adjacent frames, and the Kalman filter makes some corrections to the results based on the prediction of the target trajectory.
Especially in government scenarios, where the movement trajectories of vehicles are relatively regular, the effect of this method will be quite excellent and stable.
Moreover, in adjacent frames, Meng Fanqi also introduced the IOU distance, which is the area of intersection between two detection frames, and used it as the weight of bipartite graph matching.
At the same time, it can also calculate the vehicle speed based on the distance between frames, and reasonably design the thresholds of relevant parameters based on the vehicle speed range. It can be said that it kills two birds with one stone.
On this basis, there is also an improved version of DeepSORT. This method is inspired by pedestrian recognition and determines whether the two are the same person by comparing the distance between the two hidden layer parameter vectors.
However, Meng Fanqi did not prepare so thoroughly and achieve so much. Meeting the basic needs was enough. As for continuing to improve this part, he only plans to write down an idea here for Baidu.
Time is better spent on Google.
