How can we assure a media pipe pose joint landmarks are reliable and not predicted?
We are using the python MP SDK, complexity = 1. Our application requires very high reliability of visible joints in the frame. The MP predicts joints present landmarks even if there is a vision interference, which is amazing but not accurate. We were expecting to find significant lower presence scoring where they are predicated and not reliable tracked, but found very similar presence scoring in both cases. Did someone face this issue and found a valid way to assure a joint's landmarks is exactly where it is :) ???
