Hand pose estimation

A mobile electronic device with an embedded implementation of a neural network, comprising:

• a mobile communication device;
• and
• an embedded implementation of a neural network in multi-task deep learning paradigm for machine vision, wherein the neural network in the embedded implementation comprises:
• a vision processing unit having a plurality of super hi-vision (SHV) processors or processor cores;
• an encoder comprising a first tier, a second tier, and a third tier, wherein each tier comprises a respective number of unit blocks,
• each succeeding tier receives a respective output from an immediately preceding tier and has a higher network complexity than the immediately preceding tier,
• wherein the embedded implementation of the neural network comprises:
• a visibility decoder layer that is operatively coupled to the encoder and is used to determine one or more masks based at least in part upon an encoder output received from the encoder and a binary cross-entropy loss of the visibility decoder layer;
• a keypose decoder layer that is operatively coupled to the encoder, that is used to receive the encoder output from the encoder, and that corresponds to a classification loss of a keypose predicted from one or more keypoints in an input to the embedded implementation;
• and
• an orientation decoder layer that is operatively coupled to the encoder and is used to regularize predictions by the embedded implementation based at least in part upon the encoder output received from the encoder and a categorical orientation loss of the orientation decoder layer.