I can run the given example by Apple (https://developer.apple.com/documentation/vision/vndetecthumanhandposerequest). However, I am not sure what are the options to fine tune its behaviour? I know you can control by the confidence level, is there other ways to control the points to make the detection more consistent?
