1). I am training a network for 2D pose detection, and current plan is add MPII, LSP and Human3.6 in the training set. It looks like each dataset has slightly different set of marked keypoints. How should I deal with it? To my understanding, I should fix the output dimension of the network .
2). I am using tensorflow. How should I write the loss function over the 2D heatmaps ? (if use the heatmap regression)
Thanks.
