In some cases, it is necessary to draw a bounding box on the screen to let the computer Know where in the frame a dog is marked. For example, when we want to mark different types of vehicles on the street view map, we need to draw the range of a car first, and then give him a marker. These marks become the "standard answer" for the computer to learn how to recognize different objects.
The work of these markings seems simple but popular database cumbersome, so most companies doing computer vision technology will outsource these marking work to companies that specialize in helping to mark various image materials, and these companies are usually located in low labor cost countries. , or often employ working-study students who receive the minimum wage, or belong to a group of low socioeconomic status in the local area, including foreign migrant workers and workers with lower education levels.
According to a study published in 2020 at the Top Symposium in Human-Computer Interaction, researchers conducted field studies and in-depth interviews with these data taggers, review taggers, project leaders, and computer managers by entering these data tagging companies. Engineers developing vision technologies have discovered that unequal power relations and business-interest-oriented workflows in the labelling of data greatly affect the basis for these computer-learning standard answers.