Struct ConvolutionDescriptor 

pub struct ConvolutionDescriptor {
    pub padding: Vec<u32>,
    pub stride: Vec<u32>,
    pub dilation: Vec<u32>,
    pub groups: u32,
}

Describes a convolution operation’s hyper-parameters.

All vectors are indexed by spatial dimension (e.g. for 2-D convolutions they have length 2, for 3-D length 3).

Fields

padding: Vec<u32>

Zero-padding applied to each spatial dimension (symmetric).

stride: Vec<u32>

Stride of the convolution kernel in each spatial dimension.

dilation: Vec<u32>

Dilation factor in each spatial dimension.

groups: u32

Number of groups for grouped/depthwise convolution.

Implementations

impl ConvolutionDescriptor

pub fn conv2d( pad_h: u32, pad_w: u32, stride_h: u32, stride_w: u32, dilation_h: u32, dilation_w: u32, groups: u32, ) -> DnnResult<Self>

Creates a standard 2-D convolution descriptor.

Errors

Returns DnnError::InvalidArgument if stride or dilation contains a zero value, or if groups is zero.

pub fn spatial_dims(&self) -> usize

Returns the number of spatial dimensions this descriptor covers.

pub fn output_size( input: u32, kernel: u32, pad: u32, stride: u32, dilation: u32, ) -> DnnResult<u32>

Computes the output spatial size for a single dimension.

Formula: floor((input + 2*pad - dilation*(kernel-1) - 1) / stride) + 1

Errors

Returns DnnError::InvalidDimension if the computation underflows (i.e. the kernel is too large for the padded input).

Trait Implementations

impl Clone for ConvolutionDescriptor

fn clone(&self) -> ConvolutionDescriptor

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ConvolutionDescriptor

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

Formats the value using the given formatter.