Struct DepthwiseSeparableConv 

pub struct DepthwiseSeparableConv { /* private fields */ }

Depthwise Separable Convolution

Efficient convolution used in MobileNet architectures:

1. Depthwise convolution: one filter per input channel
2. Pointwise convolution: 1x1 conv to mix channels

Reduces parameters from O(K^2 * C_in * C_out) to O(K^2 * C_in + C_in * C_out)

Implementations

impl DepthwiseSeparableConv

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

Create a new depthwise separable convolution

pub fn set_depthwise_weights(&mut self, weights: Vec<f32>) -> CnnResult<()>

Set depthwise weights

pub fn set_pointwise_weights(&mut self, weights: Vec<f32>) -> CnnResult<()>

Set pointwise weights

Trait Implementations

impl Clone for DepthwiseSeparableConv

fn clone(&self) -> DepthwiseSeparableConv

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for DepthwiseSeparableConv

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

Formats the value using the given formatter.

impl Layer for DepthwiseSeparableConv

fn forward(&self, input: &Tensor) -> CnnResult<Tensor>

Perform the forward pass

fn name(&self) -> &'static str

Get the layer name

fn num_params(&self) -> usize

Get the number of parameters