Conv2D

@frozen
public struct Conv2D<Scalar> : Layer where Scalar : TensorFlowFloatingPoint

A 2-D convolution layer (e.g. spatial convolution over images).

This layer creates a convolution filter that is convolved with the layer input to produce a tensor of outputs.

  • The 4-D convolution filter.

    Declaration

    public var filter: Tensor<Scalar>
  • The bias vector.

    Declaration

    public var bias: Tensor<Scalar>
  • The element-wise activation function.

    Declaration

    @noDerivative
    public let activation: Activation
  • The strides of the sliding window for spatial dimensions.

    Declaration

    @noDerivative
    public let strides: (Int, Int)
  • The padding algorithm for convolution.

    Declaration

    @noDerivative
    public let padding: Padding
  • The dilation factor for spatial dimensions.

    Declaration

    @noDerivative
    public let dilations: (Int, Int)
  • The element-wise activation function type.

    Declaration

    public typealias Activation = @differentiable (Tensor<Scalar>) -> Tensor<Scalar>
  • Creates a Conv2D layer with the specified filter, bias, activation function, strides, dilations and padding.

    Declaration

    public init(
      filter: Tensor<Scalar>,
      bias: Tensor<Scalar>? = nil,
      activation: @escaping Activation = identity,
      strides: (Int, Int) = (1, 1),
      padding: Padding = .valid,
      dilations: (Int, Int) = (1, 1)
    )

    Parameters

    filter

    The 4-D convolution filter of shape [filter height, filter width, input channel count, output channel count].

    bias

    The bias vector of shape [output channel count].

    activation

    The element-wise activation function.

    strides

    The strides of the sliding window for spatial dimensions, i.e. (stride height, stride width).

    padding

    The padding algorithm for convolution.

    dilations

    The dilation factors for spatial dimensions, i.e. (dilation height, dilation width).

  • Returns the output obtained from applying the layer to the given input.

    The output spatial dimensions are computed as:

    output height = (input height + 2 * padding height - (dilation height * (filter height - 1) + 1)) / stride height + 1

    output width = (input width + 2 * padding width - (dilation width * (filter width - 1) + 1)) / stride width + 1

    and padding sizes are determined by the padding scheme.

    Note

    Padding size equals zero when using .valid.

    Declaration

    @differentiable
    public func forward(_ input: Tensor<Scalar>) -> Tensor<Scalar>

    Parameters

    input

    The input to the layer of shape [batch size, input height, input width, input channel count].

    Return Value

    The output of shape [batch count, output height, output width, output channel count].

  • Creates a Conv2D layer with the specified filter shape, strides, padding, dilations and element-wise activation function.

    Declaration

    public init(
      filterShape: (Int, Int, Int, Int),
      strides: (Int, Int) = (1, 1),
      padding: Padding = .valid,
      dilations: (Int, Int) = (1, 1),
      activation: @escaping Activation = identity,
      useBias: Bool = true,
      filterInitializer: ParameterInitializer<Scalar> = glorotUniform(),
      biasInitializer: ParameterInitializer<Scalar> = zeros()
    )

    Parameters

    filterShape

    The shape of the 4-D convolution filter, representing (filter height, filter width, input channel count, output channel count).

    strides

    The strides of the sliding window for spatial dimensions, i.e. (stride height, stride width).

    padding

    The padding algorithm for convolution.

    dilations

    The dilation factors for spatial dimensions, i.e. (dilation height, dilation width).

    activation

    The element-wise activation function.

    filterInitializer

    Initializer to use for the filter parameters.

    biasInitializer

    Initializer to use for the bias parameters.