The bitterness and helplessness of the self-driving annotator: 20 cents to draw a box and a monthly salary of 3,000 yuan

Autonomous driving technology is one of the important applications in the field of artificial intelligence, which requires a large amount of data for training and optimization. The source of these data is the autonomous driving data annotation company. According to reports from Snow Leopard Finance, these companies employ thousands of employees in third- and fourth-tier cities to provide customers with pictures with data tables. After these photos are transmitted to Party A’s system, they are used to train autonomous driving technology to make it more like a real human driver.

The company and its employees provide the "water source" for the autonomous driving industry, but they do not share the benefits brought by the development of the industry. On the contrary, they face the dilemma of low income, high pressure and no development. McKinsey predicts that total domestic sales of self-driving cars will reach about $230 billion by 2030 and about $360 billion by 2040. According to the observation of a practitioner, the salary range for workers in this industry is generally between 3,000 and 6,000 yuan. “You can get 3,000 yuan when you first start a job, and you can get 4,000 yuan when you are proficient in the work. If you are in a management position or quality inspection, you can get 6000".

Autonomous driving data annotation is a low-threshold, repetitive labor. Employees earn about 20 cents each time they draw a frame with their mouse on a picture and label the type of car (SUV, sedan or truck). Annotated pictures can be divided into 2D and 3D. 2D behaves like ordinary photos, while 3D contains spatial point cloud data. According to a management of an autonomous driving data annotation company, with the popularity of lidar, 3D images will become more and more common starting in 2021.

Although this work is simple, it is very hard. In the work of autonomous driving data annotators, it is common to work overtime, stay up late or even stay up all night. An executive at an autonomous driving data annotation company said his company will not hire employees 26 and older.

In IT House’s observation, autonomous driving data annotation suppliers are also suffering from price competition and cost pressure. From 2016 to 2018, unit prices in this industry were very high, and successful companies were able to make nine-figure profits every year. As more and more teams enter the market, Party B's choices become more diverse, and industry prices become more transparent. It has become common practice to compare prices and negotiate price reductions with each other. The unit price of pulling a frame is now 40% to 50% lower than in 2016, and there will be a certain degree of decline every year. “Party A is increasingly emphasizing cost reduction and efficiency improvement, and wants to use less money to pull more frames. ”

In order to reduce costs, some practitioners try to use intelligent tools to improve data annotation efficiency and quality, but this requires a lot of investment and research and development work. Before AI completely replaces human labor, "the current method of reducing wages can achieve better cost reduction effects."

The above is the detailed content of The bitterness and helplessness of the self-driving annotator: 20 cents to draw a box and a monthly salary of 3,000 yuan. For more information, please follow other related articles on the PHP Chinese website!