The Kissing Specialist was born: numb to see the kissing pictures

category:Society click:518
 The Kissing Specialist was born: numb to see the kissing pictures


China and the new network, Shanghai, August 9, 29 year old Ji Heng Shan sitting in front of the computer, fingers gently click the mouse, the screen hundreds of thousands of different contents of the screen tagged into two categories - kissing and non kissing. As an AI announcer for Youku, hes training machines to recognize what a kiss lens is, so hes also nicknamed the Kiss Specialist.

At the moment, with the application of artificial intelligence and video black technology, it not only improves the users viewing experience, but also gives birth to a lot of cool new jobs: the kiss expert, the star annotate, the shooting video Analyst... The program ape, which is both interested in entertainment and sports, has become the most scarce person in Chinas video website, and reflects new employment changes brought about by the rise of the new economy and the rise of digital economy.

To search and edit intelligently, you have to train the machine to find and tag scenes from vast amounts of material. Supply diagram

At the beginning I told my girlfriend that my job was to train a computer to identify a kiss scene. She heard it was amazing, who can give you money to do this? he graduated in Beihang University in 2016, a masters computer vision, and the main purpose of this major is to teach computer to understand The world . After graduating, he worked as a programmer in a Internet Co. Last September, he joined the Alibaba technology team to become an advanced algorithm engineer of Youku. The main job was to teach the computer to recognize kissing and so on. It was also called kiss play appraiser.

This position just coincides with my major, delivers happiness and happiness to everyone, and can watch online drama in the name of work. I like it very much. Talking about his work, Ji Heng Shan has a smile. He usually likes to watch online drama, the most favorite Military Division League, is Wu Xiubos iron powder.

In the past 10 years, China has consumed nearly 600 million Internet video content users. More and more people are watching TV plays, movies and sports games with their mobile phones and computers. Every time you take the subway from work to work, you can see someone with a cell phone and a video with an earplug. Ji Heng Shan found that the needs of people are no longer simple to see the play, but also want to see more interesting pieces, collection, more actively find their favorite stars or scenes.

To achieve search and intelligent editors, you have to train machines to find the corresponding scenes from mass material and tag them, like picking crabs and turtles from the vast ocean. The kiss drama of the male and female characters conveys beauty and happiness, which is generally the climax of the development of the plot, so it has become the most common label in the drama. How to teach the machine to recognize kissing has become the first problem of gehang and colleagues.

This seemingly cold AI black technology really requires a lot of manpower in its initial stages. Ji Hengshan cut thousands of pictures from the gallery and videos, one of which was marked kissing. This process is called data annotation.

The teams Kissing Specialists typed out pictures of kissing from different angles and pasted them on the wall. A group of people had a heated discussion about what was called kissing. Supply diagram

Hundreds of kissing pictures a day, up to a thousand, staring at a computer screen for a day, and finally a little numb. Usually, the more accurate and quantitative the data is, the better the model, says Ji. Naturally, the effect of machine recognition will be better. For example, in order to make beautiful pictures, the gesture of kissing is also different, some of them look at it, and some of them have to print out the kiss picture on the wall, and a group of people talk about how to kiss.

The Kiss Specialist is a technical job requiring tedious post-commissioning based on the machines response. The machine learns the features of thousands of pictures, and then recognizes them by giving it any kissing picture. So far, Ji and his colleagues have labeled more than 20,000 kissing plays.

There are 8 people at JIhangs team. With the rapid expansion of business, the kiss expert also began to contact and study other scenes, such as crying, playing, and then identifying the shooting, throwing, shooting, celebrating and so on in the sports competition, so as to make the website make a wonderful collection. They not only need to look at the pictures, but also often visit the class, to the game or the live show to observe and experience the subtle differences in each scene, action. Currently, 5-minute full-link video processing can be completed in 10 seconds. For example, the drama three life ten miles of peach, each 50 minutes of video is cut into 3000 frames, the machine only takes 5 seconds to pick out Yang Mi and Marks kiss. In the past few years, there are some new phenomena in the entertainment circle, such as the growing image of the stars, how to train the machine to identify the ancient power, Na ZAH and Di Ali Gerba, how to distinguish the three small of TFBOYS, and the need for continuous trial and error and model training. Distinguishing star face, net red face and twin has become a new challenge for Yoshihito and his colleagues. Source: Responsible Editor of China News Network: Xun Jianguo_NN7379

There are 8 people at JIhangs team. With the rapid expansion of business, the kiss expert also began to contact and study other scenes, such as crying, playing, and then identifying the shooting, throwing, shooting, celebrating and so on in the sports competition, so as to make the website make a wonderful collection. They not only need to look at the pictures, but also often visit the class, to the game or the live show to observe and experience the subtle differences in each scene, action.

Currently, 5-minute full-link video processing can be completed in 10 seconds. For example, the drama three life ten miles of peach, each 50 minutes of video is cut into 3000 frames, the machine only takes 5 seconds to pick out Yang Mi and Marks kiss.

In the past few years, there are some new phenomena in the entertainment circle, such as the growing image of the stars, how to train the machine to identify the ancient power, Na ZAH and Di Ali Gerba, how to distinguish the three small of TFBOYS, and the need for continuous trial and error and model training. Distinguishing star face, net red face and twin has become a new challenge for Yoshihito and his colleagues.