Pose estimation methods are divided into two paradigms—top-down and bottom-up—where top-down generally achieves higher accuracy but speed scales linearly with the number of people, while bottom-up speed is largely independent of the number of people.