Conv2D

Conv2D is a layer that executes 2-dimensional convolution with biases.

Conv2D(;input_shape::Tuple, filter::Int64, padding::Int64=0, kernel_size::Tuple{Int64, Int64}, strides::Tuple{Int64, Int64}=(1,1), activation_function::Module, randomization::Bool=true)

input_shape: a tuple of input data shape, (s1, s2, channels), can be set automatically, unless it is the first layer in the network

filter: the number of dimensions for output data

padding: the number of outer rings for padding, default 0

kernel_size: the shape of kernel, a tuple of two integers, default true

strides: the moving step size in vertical and horizontal direction, default (1,1)

activation_function: a chosen activation function

randomization: generate the filters with Glorot Uniform or set as zeros