A company is developing a solution to create images of protective eyewear. The solution requires high accuracy and must reduce the risk of erroneous annotations as much as possible. Which approach will satisfy these requirements?