Struct Conv2D

pub struct Conv2D {
    pub in_channels: usize,
    pub out_channels: usize,
    pub kernel_size: (usize, usize),
    pub stride: (usize, usize),
    pub padding: (usize, usize),
    pub weights: Vec<Vec<Vec<Vec<f64>>>>,
    pub bias: Vec<f64>,
}

Represents a 2D convolution layer

Fields

in_channels: usize

Number of input channels

out_channels: usize

Number of output channels (number of filters)

kernel_size: (usize, usize)

Kernel size (height, width)

stride: (usize, usize)

Stride (height, width)

padding: (usize, usize)

Padding (height, width)

weights: Vec<Vec<Vec<Vec<f64>>>>

Weights (out_channels, in_channels, kernel_height, kernel_width)

bias: Vec<f64>

Biases (out_channels)

Implementations

impl Conv2D

pub fn new( in_channels: usize, out_channels: usize, kernel_size: (usize, usize), stride: (usize, usize), padding: (usize, usize), ) -> Self

Creates a new Conv2D layer

Arguments

• in_channels - Number of input channels
• out_channels - Number of output channels
• kernel_size - Size of the convolving kernel (height, width)
• stride - Stride of the convolution (height, width)
• padding - Zero-padding added to both sides of the input (height, width)

Example

use algos::ml::deep::conv::Conv2D;
let conv = Conv2D::new(3, 64, (3, 3), (1, 1), (1, 1));

pub fn output_shape( &self, input_height: usize, input_width: usize, ) -> (usize, usize)

Computes the output shape for given input dimensions

Arguments

• input_height - Height of input
• input_width - Width of input

Returns

• Tuple of (output_height, output_width)

pub fn forward( &self, input: &[Vec<Vec<Vec<f64>>>], ) -> (Vec<Vec<Vec<Vec<f64>>>>, Conv2DCache)

Forward pass of the convolution layer

Arguments

• input - Input tensor of shape (batch_size, in_channels, height, width)

Returns

• Output tensor of shape (batch_size, out_channels, output_height, output_width)
• Cache for backward pass

pub fn backward( &self, grad_output: &[Vec<Vec<Vec<f64>>>], cache: &Conv2DCache, ) -> (Vec<Vec<Vec<Vec<f64>>>>, Conv2DGradients)

Backward pass of the convolution layer

Arguments

• grad_output - Gradient of the loss with respect to the output
• cache - Cache from forward pass

Returns

• Gradient with respect to input
• Gradient with respect to weights
• Gradient with respect to bias

Trait Implementations

impl Clone for Conv2D

fn clone(&self) -> Conv2D

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Conv2D

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

Formats the value using the given formatter.