The BodyPix model can estimate which pixels in an image are part of a person, and which pixels are part of each of 24 body parts. It works on a single person, and such works best when one person is present in an image.

The output stride and model (indicated by mobileNetArchitecture) have the largest effects on accuracy/speed. A higher output stride results in lower accuracy but higher speed. A larger model, indicated by the mobileNetArchitecture dropdown, results in higher accuracy but lower speed.