Human Activity Recognition Based on a Modified Capsule Network

Table 3

Detailed structural parameters of the MCN model on the WISDM dataset.

Model	Layers	Parameters
MCN	Conv2d-1	Kernel size = 3 × 3 step = 1 filters = 32 padding = 1
	Conv2d-2	Kernel size = 3 × 3 step = 1 filters = 64 padding = 1
	Conv2d-3	Kernel size = 3 × 3 step = 1 filters = 128
	Primary caps	Kernel size = 2 × 2 step = 2 filters = 32 × 8
	Activity caps	Routing iterations = 3