tannic::transformation::Convolution2D Class Reference

Expression for 2D convolution operations. More...

#include <convolutions.hpp>

Public Member Functions
constexpr type	promote (type signal, type kernel) const
	Type promotion for 2D convolution.
constexpr Shape	transform (Shape const &signal, Shape const &kernel) const
	Computes output shape for 2D convolution.
void	forward (Tensor const &signal, Tensor const &kernel, Tensor &result) const
	Performs the forward pass of 2D convolution.

Public Attributes
std::array< std::size_t, 2 >	strides
	Stride values for height and width dimensions.
std::array< std::size_t, 2 >	padding
	Padding values for height and width dimensions.

Detailed Description

Expression for 2D convolution operations.

Implements 2D convolution with:

• Input and kernel dtype validation
• Automatic output shape calculation
• Support for strides and padding in both spatial dimensions

Member Function Documentation

forward()

void tannic::transformation::Convolution2D::forward	(	Tensor const &	signal,
		Tensor const &	kernel,
		Tensor &	result
	)		const

Performs the forward pass of 2D convolution.

Parameters

signal	Input signal tensor
kernel	Convolution kernel tensor
result	Output tensor to store convolution result

promote()

constexpr type tannic::transformation::Convolution2D::promote ( type  signal, type  kernel  ) const

inlineconstexpr

Type promotion for 2D convolution.

Requires both signal and kernel to have identical data types.

Parameters

signal	Data type of the input signal tensor
kernel	Data type of the kernel tensor

Returns

Data type (same as inputs)

Exceptions

Exception

if signal and kernel dtypes don't match

transform()

constexpr Shape tannic::transformation::Convolution2D::transform ( Shape const &  signal, Shape const &  kernel  ) const

inlineconstexpr

Computes output shape for 2D convolution.

Calculates the output shape based on input signal shape, kernel shape, strides, and padding using the standard 2D convolution formula.

Parameters

signal	Shape of the input signal tensor (N, C_in, H_in, W_in)
kernel	Shape of the kernel tensor (C_out, K_in, K_h, K_w)

Returns

Output shape after 2D convolution (N, C_out, H_out, W_out)

Exceptions

Exception

if input ranks are not 4 or channel dimensions don't match

Expected input shapes:

• Signal: [batch_size, input_channels, height, width]
• Kernel: [output_channels, input_channels, kernel_height, kernel_width]
• Output: [batch_size, output_channels, output_height, output_width]

Output dimension formulas:

• H_out = (H_in + 2*padding_h - K_h) / stride_h + 1
• W_out = (W_in + 2*padding_w - K_w) / stride_w + 1

Member Data Documentation

padding

std::array<std::size_t, 2> tannic::transformation::Convolution2D::padding

Padding values for height and width dimensions.

strides

std::array<std::size_t, 2> tannic::transformation::Convolution2D::strides

Stride values for height and width dimensions.

---

The documentation for this class was generated from the following file:

• include/tannic/convolutions.hpp