conv2dlayer.h File Reference

#include <vector>
#include "layer/layer.h"
#include "tensor/tensor.h"
#include "compute/operation.h"
#include "compute/variable.h"
#include "compute/conv2dforward/conv2dforwardop.h"

Include dependency graph for conv2dlayer.h:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Classes
class	magmadnn::layer::Conv2dLayer< T >

Enumerations
enum	padding_t { SAME, VALID }

Functions
template<typename T >
Conv2dLayer< T > *	magmadnn::layer::conv2d (op::Operation< T > *input, const std::vector< unsigned int > &filter_shape, int out_channels, const std::vector< unsigned int > &padding, const std::vector< unsigned int > &strides, const std::vector< unsigned int > &dilation_rates, bool use_cross_correlation, bool use_bias, tensor_filler_t< T > filter_initializer, tensor_filler_t< T > bias_initializer)
template<typename T >
Conv2dLayer< T > *	magmadnn::layer::conv2d (op::Operation< T > *input, const std::vector< unsigned int > &filter_shape, int out_channels, padding_t padding, const std::vector< unsigned int > &strides, const std::vector< unsigned int > &dilation_rates, bool use_cross_correlation, bool use_bias, tensor_filler_t< T > filter_initializer, tensor_filler_t< T > bias_initializer)

Detailed Description

Author

Daniel Nichols

Version

1.0

Date

2019-07-08

Copyright

Copyright (c) 2019

Function Documentation

conv2d()

template<typename T >

Conv2dLayer< T > * magmadnn::layer::conv2d	(	op::Operation< T > *	input,
		const std::vector< unsigned int > &	filter_shape = {3, 3},
		int	out_channels = 1,
		const std::vector< unsigned int > &	padding = {0, 0},
		const std::vector< unsigned int > &	strides = {1, 1},
		const std::vector< unsigned int > &	dilation_rates = {1, 1},
		bool	use_cross_correlation = true,
		bool	use_bias = false,
		tensor_filler_t< T >	filter_initializer = {GLOROT,{0.0, 0.2f}},
		tensor_filler_t< T >	bias_initializer = {GLOROT,{0.0, 0.2f}}
	)

Create a Conv2d Layer. Computes the output on the GPU. Uses CuDNN. No host functionality.

Template Parameters

T

numeric

Parameters

input	the input data; must be a 4D tensor in format NCHW (N-batch, C-Channel, H-height, W-Width)
filter_shape	shape of the convolution kernel. must be a 2 dimensional vector; defaults to {3,3}
out_channels	number of output filters; defaults to 1
padding	the padding size to use. must be a two dimensional vector; defaults to {0,0}
strides	striding of convolution. must be a two dimensional vector; defaults to {1,1}
dilation_rates	rate of dilation. must be a two dimensional vector; defaults to {1,1}
use_cross_correlation	whether to do a cross correlation convolution or standard convolution; defaults to cross correlation
use_bias	use convolutional bias or not; defaults to false
filter_initializer	how to initialize the filter tensor; defaults to {GLOROT,{0.0f,0.2f}}
bias_initializer	how to initialize the bias tensor; defaults to {GLOROT, {0.0f,0.2f}}

Returns

Conv2dLayer<T>* a new layer