[P] Using Keras, TensorFlow, CoreML, and ARKit to create marker-less 3D interaction on an iPhone : MachineLearning

[P] Using Keras, TensorFlow, CoreML, and ARKit to create marker-less 3D interaction on an iPhone from MachineLearning

Hi. This is a quite impressive work. May I know some details of pose estimation? I think the correspondence between 2d and 3d points is neccesary for pose estimation. Coordinates of 2d points can be obtained by MobileNets. How do you obtain the 3d coordinate of key points of hands? Do you use PnP algorithm to estimate the pose? Thanks in advance.