primary_cap: Apply Conv2D 'n_channels' times and concatenate all capsules

In dfalbel/capsnet: Dynamic Routing Between Capsules implementation in Keras

Description

Apply Conv2D 'n_channels' times and concatenate all capsules

Usage

primary_cap(inputs, dim_vector, n_channels, kernel_size, strides, padding)

Arguments

inputs	4D tensor, shape=[None, width, height, channels]
dim_vector	the dim of the output vector of capsule
n_channels	the number of types of capsules
kernel_size	kernel size for convoltions
strides	strides for convolution
padding	padding for convolutions

Value

output tensor, shape=[None, num_capsule, dim_vector]

dfalbel/capsnet documentation built on May 29, 2019, 12:37 p.m.