Table 2.

The training procedure of hierarchical processing network.

Input: The training set of robotic dance pose (including three subsets: HSV training subset, RGB training subset, and depth training subset), and the initial hierarchical processing network (including four initial CNNs: HSV-CNN, RGB-CNN, depth-CNN, and the synthesis CNN)

Output: The trained hierarchical processing network (including four trained CNNs: HSV-CNN, RGB-CNN, depth-CNN, and the synthesis CNN)

Steps:

1. HSV training subset is used to train HSV-CNN.

2. RGB training subset is used to train RGB-CNN.

3. Depth training subset is used to train depth-CNN.

4. Based on the above three trained CNNs (HSV–CNN, RGB-CNN, and depth-CNN), the training set of robotic dance pose (including three subsets: HSV training subset, RGB training subset, and depth training subset) is used to train the synthesis CNN.