Assume I want to use the model named full_human, but I also want to track hands and other elements. Should I start from scratch (make a new model), or add the points and train the model?
I’m facing a similar situation and trying to determine next steps. I already have some labeled data, and I want to try using a pretrained model from the zoo as an initialization point. I have all the keypoints in the pretrained model plus more, and I’m wondering if (a) there’s a way to tell the network which keypoints match between my labeled data and the pretrained model and (b) whether that’s even worth doing (vs. just setting the pretrained model as the initialization point and going from there without doing anything with the keypoints).