This top international computer vision conference in December 2013 actually took a little longer.
The conference schedule is divided into two days of seminars on the 1st and 2nd, followed by a four-day main session on the 3rd and 6th, and finally two days of seminars on specific topics on the 7th and 8th.
As for the IMAGENET-2013 event where Meng Fanqi won the championship, its seminar is scheduled to be held on December 7, 2013.
Therefore, strictly speaking, Meng Fanqi does not need to participate in the main part of the conference.
But he didn't plan to go too late. The group's tickets were scheduled for the 4th.
There was nothing special about the first three days of December.
Meng Fanqi went to Baidu's Yanjing headquarters again, this time because of Baidu's propaganda needs.
Robin Li's Deep Learning Research Institute moved very quickly. He quickly studied the key points of Meng Fanqi's method and began to try to reproduce it on his own data.
Wang Kai, the Baidu researcher who is responsible for connecting with Meng Fanqi, has frequently visited Meng Fanqi near Yanjing Electric Power University during this period, and has learned almost all the relevant knowledge.
Meng Fanqi still prefers to deal with relatively pure people like Wang Kai. Although it stands to reason that Wang Kai's contact frequency is slightly higher and the questions he asks sometimes exceed the reasonable range, his methods are still very particular. .
Since he had nothing particularly important to do, Meng Fanqi didn't care too much.
Shortly after the last technology contract was signed, with Wang Kai’s active promotion, Baidu had mastered the algorithm.
Robin Li immediately decided to prepare materials and produce a promotional video with the theme of "Real-time detection can subvert the world."
To put it more clearly, it is to show the usage scenarios of Meng Fanqi’s YOLO algorithm in various industries.
Its core is to highlight several characteristics, such as wide use, fast speed and high efficiency.
After the boss gave an order, the visual algorithm researchers at the institute sat in a row and started brainstorming.
They need to think about requirements within the context of these scenarios, collect images, and run detection algorithms on them.
These materials are then given to the marketing department below, which is responsible for designing copywriting, editing videos and other operations.
"First of all, it must be automatic driving, road signs, vehicles and pedestrians. The detection of these aspects is not only of great significance, but also most consistent with the 'real-time' feature. The real-time demand for automatic driving is the most indispensable." Wang Haifeng said Got a head.
"Autonomous driving not only fits the theme, but also has a huge market. It is one of the four basic needs for food, clothing, housing and transportation, and its scope of influence is also the most exaggerated. Moreover, its implementation is also very intelligent, giving it a feeling of the third industrial revolution. "A researcher nodded in agreement. The scene display of autonomous driving is the most consistent with the theme and the most shocking.
"There is also industrial inspection. Industrial inspection may be defect detection, checking defective and defective products, or it may be statistics and counting, such as counting the quantity of certain materials and counting them." Wang Kai believes that different industries have their own demand, there will be great business opportunities in this area.
"However, the real-time attribute of industrial testing is not that strong. For many operations, results can be obtained within a few seconds."
"But we can emphasize efficiency and make a slightly shorter video. The left side shows the worker's own point detection, the middle is a very slow detection algorithm, and the right side is our new algorithm. Speed up the video and compare the number of their results."
"That's right. Finally, a percentage numerical result is given, such as a 160% efficiency improvement or something like that."
"Medical images are also a big market, and we definitely need to mention this part. But compared with autonomous driving and industry, the concept of real-time classification and detection of medical images is even weaker. There are basically no situations that require real-time recognition. ." Some researchers raised this difficulty.
"Furthermore, regarding medical responsibility and ethics, we can only say that it is an auxiliary method and cannot be promoted as a substitute for doctors. This requires great attention." Yu Kai passed by and listened for a while before reminding him aloud.
"There are also security inspections..."
"Intelligent marking of key periods of long-term recording...."
"Food testing..."
"Animal testing...."
"Face Detection...."
"Pedestrian detection...."
Everyone works together, whether it really has a chance to be implemented immediately, or it is just a painting, and it feels that basically no one will pay for it within three to five years, all of them are proposed first and counted.
"Let's get here first today. Within two days, we will prepare at least twenty picture test results in each direction, and five short videos of actual camera measurements will be provided as material."
Yu Kai finally made the final decision and concluded, "All the picture materials will be ready and released today, and the short video can be a little slower."
In the short term, the Baidu Research Institute’s primary task is to assist the marketing and publicity departments to promote the versatility, high performance, and disruptive nature of the YOLO algorithm.
According to Robin Li's idea, this promotional video must be released quickly.
The development speed of AI technology is so fast that it would be more shocking if it were started one day earlier.
Mid- and long-term plans can only be formulated and planned after seeing real feedback from the market and contacting and talking to major industry leaders.
After Meng Fanqi came to Baidu headquarters this time, it was not the three giants who were responsible for the connection, nor did he go to the Deep Learning Institute, but was directly pulled to the marketing department.
As soon as I entered the door, I was pushed directly onto the chair by my two sisters and started applying makeup in front of the mirror.
"How about such a big battle?" Meng Fanqi had the experience of hosting in the hospital when he was a freshman. As a boy, it was also his first time to come into contact with makeup. With this experience, he probably knew what to do.
I just didn't expect it to be so exaggerated. There were three professionals just smearing it on my face.
This made Meng Fanqi really flattered.
"We still have a lot to take pictures today." Li Yanhong was very busy, so he sent his personal Secretary Liu to accompany him this time. "I heard from Mr. Li that you are going to attend the International Computer Vision Conference in a few days. Before you leave, these pieces of material must be shot. They will be ready to be distributed soon."
"Moving so fast?" Meng Fanqi knew that Robin Li had always wanted to make practical achievements in AI. His YOLO algorithm gave him what he needed most, but even so, Robin Li's movements seemed too fast. Will the promotional video be released in a week?
"First take a formal half-length photo. This will definitely be used at the press conference. Later, I will take a few more videos explaining algorithms, writing codes, and doing testing experiments. Finally, when Mr. Li comes back, I will take a few more A piece of material for you to communicate and discuss. That’s almost enough.”
