Struct neuro::layers::Conv2D

pub struct Conv2D { /* fields omitted */ }

Defines a 2D convolution layer.

Methods

impl Conv2D

pub fn new(
    num_filters: u64,
    kernel_size: (u64, u64),
    stride: (u64, u64),
    padding: Padding
) -> Box<Conv2D>

Creates a 2D convolution layer with the given parameters.

By default, a ReLU activation is used and the parameters of the kernels are initialized using a HeNormal initializer and the biases of the layer a Zeros initializer.

Arguments

• num_filters - The number of filters in the layer.
• kernel_size - The height and width of the convolution kernels.
• stride - The vertical and horizontal stride used for the convolution.
• padding - The padding used for the convolution. Must be a variant of Padding.

pub fn with_param(
    num_filters: u64,
    kernel_size: (u64, u64),
    stride: (u64, u64),
    padding: Padding,
    activation: Activation,
    weights_initializer: Initializer,
    biases_initializer: Initializer
) -> Box<Conv2D>

Creates a 2D convolution layer with the given parameters.

By default, the parameters of the kernels are initialized using a HeUniform initializer and the biases of the layer a Zeros initializer.

Arguments

• num_filters - The number of filters in the layer.
• kernel_size - The height and width of the convolution kernels.
• stride - The vertical and horizontal stride used for the convolution.
• padding - The padding used for the convolution. Must be a variant of Padding.
• activation - The activation function used by the layer.
• weights_initializer - The initializer used to initialize the weights of the layer.
• biases_initializer - The initializer used to initialize the biases of the layer.

Trait Implementations

impl Display for Conv2D

fn fmt(&self, f: &mut Formatter) -> Result

Formats the value using the given formatter.

impl Layer for Conv2D

fn name(&self) -> &str

Returns the name of the layer.

fn initialize_parameters(&mut self, input_shape: Dim4)

Initializes the parameters of the layer.

fn compute_activation(&self, input: &Tensor) -> Tensor

Computes the activation of the layer during the forward pass.

fn compute_activation_mut(&mut self, input: &Tensor) -> Tensor

Computes the forward pass and stores intermediate values for efficient backpropagation.

fn compute_dactivation_mut(&mut self, input: &Tensor) -> Tensor

Computes the backward pass through the layer.

fn output_shape(&self) -> Dim4

Returns the shape of the output.

fn parameters(&self) -> Option<Vec<&Tensor>>

Returns the trainable parameters of the layer.

fn parameters_mut(&mut self) -> Option<(Vec<&mut Tensor>, Vec<&Tensor>)>

Returns the trainable parameters of the layer and their derivatives.

fn save(&self, group: &Group, layer_number: usize) -> Result<(), Error>

Writes the parameters of the layer in the HDF5 group.

fn set_regularizer(&mut self, regularizer: Option<Regularizer>)

Sets the regularizer for the layer.

fn print(&self)

Displays the properties of the layer.