Conv2D< T > Class Template Reference

Dynamic implementation of a 2-dimensional convolution layer with no activation. More...

Inheritance diagram for Conv2D< T >:

Public Member Functions
	Conv2D (int in_num_filters_in, int in_num_filters_out, int in_num_features_in, int in_kernel_size_time, int in_kernel_size_feature, int in_dilation_rate, int in_stride, bool in_valid_pad)
	Conv2D (std::initializer_list< int > sizes)
	Conv2D (const Conv2D &other)
Conv2D &	operator= (const Conv2D &other)
virtual	~Conv2D ()=default
RTNEURAL_REALTIME void	reset () override
	Reset the layer's state. More...
std::string	getName () const noexcept override
	Returns the name of this layer. More...
constexpr bool	isActivation () const noexcept
	Returns false since convolution is not an activation layer. More...
RTNEURAL_REALTIME void	forward (const T *input, T *output) noexcept override
	Performs forward propagation for this layer. More...
RTNEURAL_REALTIME void	setWeights (const std::vector< std::vector< std::vector< std::vector< T >>>> &inWeights)
	Sets the layer weights. More...
RTNEURAL_REALTIME void	setBias (const std::vector< T > &inBias)
	Sets the layer biases. More...
RTNEURAL_REALTIME int	getKernelSizeTime () const noexcept
	Returns the size of the convolution kernel (time axis). More...
RTNEURAL_REALTIME int	getKernelSizeFeature () const noexcept
	Returns the size of the convolution kernel (feature axis). More...
RTNEURAL_REALTIME int	getStride () const noexcept
	Returns the convolution stride (feature axis) More...
RTNEURAL_REALTIME int	getDilationRate () const noexcept
	Returns the convolution dilation rate (time axis) More...
Public Member Functions inherited from Layer< T >
	Layer (int in_size, int out_size)
	Constructs a layer with given input and output size. More...
virtual	~Layer ()=default

Public Attributes
const int	num_filters_in
const int	num_features_in
const int	num_filters_out
const int	kernel_size_time
const int	kernel_size_feature
const int	dilation_rate
const int	stride
const int	num_features_out
const int	receptive_field
const bool	valid_pad
Public Attributes inherited from Layer< T >
const int	in_size
const int	out_size

Detailed Description

template<typename T>
class RTNEURAL_NAMESPACE::Conv2D< T >

Dynamic implementation of a 2-dimensional convolution layer with no activation.

Template Parameters

T	Type of the layer (float, double, int ...)

Constructor & Destructor Documentation

Conv2D() [1/3]

Conv2D	(	int	in_num_filters_in,
		int	in_num_filters_out,
		int	in_num_features_in,
		int	in_kernel_size_time,
		int	in_kernel_size_feature,
		int	in_dilation_rate,
		int	in_stride,
		bool	in_valid_pad
	)

Parameters

in_num_filters_in	number of input filters (channels)
in_num_filters_out	number of output filters (channels)
in_num_features_in	number of input features
in_kernel_size_time	size of the convolution kernel (time axis)
in_kernel_size_feature	size of the convolution kernel (feature axis)
in_dilation_rate	dilation_rate (time axis)
in_stride	convolution stride (feature axis)
in_valid_pad	whether the padding is "valid" or not ("same" otherwise)

Conv2D() [2/3]

Conv2D

(

std::initializer_list< int >

sizes

)

Conv2D() [3/3]

Conv2D

(

const Conv2D< T > &

other

)

~Conv2D()

virtual ~Conv2D ( )

virtualdefault

Member Function Documentation

operator=()

Conv2D& operator=

(

const Conv2D< T > &

other

)

reset()

RTNEURAL_REALTIME void reset ( )

overridevirtual

Reset the layer's state.

Reimplemented from Layer< T >.

References Conv2D< T >::receptive_field.

getName()

std::string getName ( ) const

overridevirtualnoexcept

Returns the name of this layer.

Reimplemented from Layer< T >.

isActivation()

constexpr bool isActivation ( ) const

constexprnoexcept

Returns false since convolution is not an activation layer.

forward()

RTNEURAL_REALTIME void forward ( const T *  input, T *  output  )

overridevirtualnoexcept

Performs forward propagation for this layer.

Implements Layer< T >.

References Conv2D< T >::dilation_rate, Conv2D< T >::kernel_size_time, Conv2D< T >::num_features_out, Conv2D< T >::num_filters_out, and Conv2D< T >::receptive_field.

setWeights()

RTNEURAL_REALTIME void setWeights

(

const std::vector< std::vector< std::vector< std::vector< T >>>> &

inWeights

)

Sets the layer weights.

The weights vector must have size weights[num_filters_out][num_filters_in][kernel_size]

setBias()

RTNEURAL_REALTIME void setBias

(

const std::vector< T > &

inBias

)

Sets the layer biases.

The bias vector must have size bias[num_filters_out]

getKernelSizeTime()

RTNEURAL_REALTIME int getKernelSizeTime ( ) const

noexcept

Returns the size of the convolution kernel (time axis).

References Conv2D< T >::kernel_size_time.

getKernelSizeFeature()

RTNEURAL_REALTIME int getKernelSizeFeature ( ) const

noexcept

Returns the size of the convolution kernel (feature axis).

References Conv2D< T >::kernel_size_feature.

getStride()

RTNEURAL_REALTIME int getStride ( ) const

noexcept

Returns the convolution stride (feature axis)

References Conv2D< T >::stride.

getDilationRate()

RTNEURAL_REALTIME int getDilationRate ( ) const

noexcept

Returns the convolution dilation rate (time axis)

References Conv2D< T >::dilation_rate.

Member Data Documentation

num_filters_in

const int num_filters_in

num_features_in

const int num_features_in

num_filters_out

const int num_filters_out

Referenced by Conv2D< T >::forward().

kernel_size_time

const int kernel_size_time

Referenced by Conv2D< T >::forward(), and Conv2D< T >::getKernelSizeTime().

kernel_size_feature

const int kernel_size_feature

Referenced by Conv2D< T >::getKernelSizeFeature().

dilation_rate

const int dilation_rate

Referenced by Conv2D< T >::forward(), and Conv2D< T >::getDilationRate().

stride

const int stride

Referenced by Conv2D< T >::getStride().

num_features_out

const int num_features_out

Referenced by Conv2D< T >::forward().

receptive_field

const int receptive_field

Referenced by Conv2D< T >::forward(), and Conv2D< T >::reset().

valid_pad

const bool valid_pad

---

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

• conv2d.h