"???" Meng Fanqi walked while stuffing the contract documents into his bag. Just as he opened the door, he heard someone say this.

He stood still and looked closely at the person in front of him, who was about half a head taller than him, with a strong build.

He was wearing loose shirts and sweatshirts, with obvious training marks. At this moment, he was staring at Meng Fanqi with a smile, and his body movements seemed a bit awkward.

"Who are you...?" Meng Fanqi raised his left eyebrow and asked tentatively.

"Wang Kai, bro." Wang Kai grinned and asked Meng Fanqi to sit aside. "Please have a seat. I am specifically responsible for handing over the code."

Wang Kai pulled his laptop to the middle of the two of them and said, "I have created the internal account and password for you. You need to upload the entire project to the designated location and add some necessary code comments, introductions, and running methods and procedures.

You need to ensure that others can understand the code in various parts, so the final task today is for me to understand the entire algorithm and go through this code and technology. This will be considered as the completion of the handover."

"No problem, I understand this." Meng Fanqi was very familiar with code handover, internal tool version control, code specifications, where comments must be written, and the rules and restrictions for naming variables and functions.

Not to mention, he had worked at Baidu in Shanghai before, so he definitely wouldn't make any mistakes in terms of coding habits that didn't meet the requirements.

He spent more than a week optimizing the code again. The reason why it took so long was largely because the handover required many necessary code comments and document writing.

In order for others to quickly understand and comprehend.

"Wang Kaige, I still have some understanding of these things. I have written the comments and documents." Meng Fanqi said as he operated on Wang Kai's computer, pulling the latest version of the code from his own private repository on GitHub.

"Let me briefly go through the entire process with you. We definitely won't be able to understand all the details today, so let's exchange contact information. We will definitely need to communicate multiple times in the future."

"Yes, Wang Kai, bro." Wang Kai had a habit of calling people "bro". With his burly figure, it created a somewhat contrasting effect. "My name is the same as the CEO's, both with the character 'Kai'."

"Oh, I see." Meng Fanqi suddenly realized, "You guys are all master's or doctoral students. You must be at least twenty-five or twenty-six. Why are you calling me 'bro'?"

"I'm twenty-four this year, bro." Wang Kai replied, "Each has their own expertise. When it comes to code, the one who excels is the 'bro'. In our team, I am the youngest brother. If you look around, everyone else is my 'bro'."

"Does that mean you're the least experienced?" When the three executives were present earlier, Meng Fanqi always felt like he was in the presence of a leader, which he wasn't used to. Now, joking with Wang Kai, an ordinary employee, gave him more of a sense of being at work.

"That's why they sent you?"

"Of course, the slowest bird flies first." Wang Kai didn't feel ashamed of his lack of skills. On the contrary, he felt proud, "If I can understand it through the handover, it means there are no problems at all. Everyone in the entire research institute will definitely be able to handle it easily."

"Keep talking, keep talking. Let's hear the explanation from the big bro." Seeing that Meng Fanqi had already pulled down the project, opened the documents and code, Wang Kai quickly stood up and greeted.

In response to the energetic greeting, more than a dozen research members came over and surrounded the two of them.

"Sister-in-law, please sit here!" Wang Kai saw Han Ci standing on the side watching, and suddenly felt that it was time for him to perform, so he quickly stood up and offered his seat.

"Don't talk nonsense, I don't have such good luck." Meng Fanqi found this guy quite interesting, "Let's get to the point. I have given you the drafts of the documents and papers. You guys should read them carefully. I won't explain them directly."

"Today, the main thing is to talk about the main structure and logic. The YOLO detection algorithm, which stands for 'you only look once', means that the image is scanned only once. Its main speed improvement comes from this.

The process of a classic detection method is as follows: first, analyze the image using computer graphics (or deep learning) methods to find several regions where objects may exist. These regions are then cropped and put into an image classifier for classification.

The most foolish approach is to repeatedly scan the image with multiple sizes of bounding boxes. Later, some methods are used to filter out the most likely regions. These methods are done in two steps: looking at the image once to select the regions of interest, and then looking at the image again to classify these regions.

My approach is to directly divide the entire image into evenly spaced grids, and each grid can output the category of the object and the coordinates of the detection box. The grid responsible for predicting the object is determined by where the object falls.

The output of my network is one-dimensional, with the first N numbers representing the probabilities of N categories, the following X numbers representing the probabilities of X detection boxes, and the last 4X numbers used to describe the positions of the detection boxes on the image."

"How is it possible to be two hundred times faster?" A quick-thinking Baidu researcher asked, "Traditional methods that focus on speed will reduce the number of proposed regions, usually no more than a few thousand. According to what you said, you would need at least dozens of grids, and each grid would require several detection boxes. In that case, it doesn't seem to make a big difference.""If we adopt a 7x7 grid with two detection boxes, the total number would only be 98 detection boxes. Moreover, there is a lot of room for adjustment in the design of the backbone network and the specific operation sequence of non-maximum suppression."

"Wouldn't that greatly affect the performance?"

"How should we handle multiple types of targets?"

"In that case, wouldn't the detection of small targets be a challenge?"

Read this novel and other amazing translated novels from the original source at the "[pawread.com]"

As Meng Fanqi explained, the researchers present began to understand more deeply, and the questions became more numerous.

Meng Fanqi had to admit, Li Yanhong's research institute really gathered a group of quite outstanding and creative young programmers.

They had only listened for less than half an hour, but they were already able to ask quite deep and meaningful questions.

There were questions about the source of speed improvement and parameter settings and meanings, which Meng Fanqi could explain well.

There were also concerns about the detection of small targets, which is indeed a hard problem for the YOLO algorithm.

When discussing technical content, time always flies by. Unconsciously, two or three hours had passed, and Meng Fanqi's mission for this trip was finally completed.

"Let's go. We signed a big deal today, it took much longer than I expected." After such intense negotiations and technical discussions, he was indeed a bit mentally exhausted. "How about I treat you to a good meal?"

"Let's go to the city." Han Ci suggested with a smile.

In fact, for most girls, after this trip, it would be hard to develop any good feelings. After several hours, someone like Mai Mang would be bored to death.

However, for Han Ci, this kind of professional talent and dedication was actually a big plus.