ConvolutionPyramid Class Reference

Implements a multiscale scheme for approximating convolution with large filters. This is the basis of the technique described in Convolution Pyramids, Farbman et al., 2011. A set of filter parameters can be estimated through an offline optimization for each desired task: gradient field integration, seamless image cloning, background filling, or scattered data interpolation. More...

#include <ConvolutionPyramid.hpp>

Public Member Functions
	ConvolutionPyramid (uint width, uint height, uint inoutPadding)
void	setFilters (const float h1[5], float h2, const float g[3])
void	process (const Texture &texture)
void	resize (uint width, uint height)
const Texture *	texture () const
unsigned int	width ()
unsigned int	height ()

Private Attributes
Program *	_downscale
	Pyramid descending pass shader.
Program *	_upscale
	Pyramid ascending pass shader.
Program *	_filter
	Filtering shader for the last pyramid level.
Program *	_padder
	Padding helper shader.
Texture	_shifted
	Contains the input data padded to the right size.
std::vector< Texture >	_levelsIn
	The initial levels of the pyramid.
std::vector< Texture >	_levelsOut
	The filtered levels of the pyramid.
float	_h1 [5] = {0.0f, 0.0f, 0.0f, 0.0f, 0.0f}
	h1 filter coefficients.
float	_h2 = 0.0f
	h2 filter multiplier.
float	_g [3] = {0.0f, 0.0f, 0.0f}
	g filter coefficients.
glm::ivec2	_resolution = glm::ivec2(0)
	Resolution expected for the input texture.
const int	_size = 5
	Size of the filter.
int	_padding = 0
	Additional padding.

Detailed Description

Implements a multiscale scheme for approximating convolution with large filters. This is the basis of the technique described in Convolution Pyramids, Farbman et al., 2011. A set of filter parameters can be estimated through an offline optimization for each desired task: gradient field integration, seamless image cloning, background filling, or scattered data interpolation.

Constructor & Destructor Documentation

ConvolutionPyramid()

ConvolutionPyramid::ConvolutionPyramid	(	uint	width,
		uint	height,
		uint	inoutPadding
	)

Constructor.

Parameters

width	internal processing width
height	internal processing height
inoutPadding	additional padding applied everywhere except on the final result texture

Note

This is mainly used for the gradient integration task.

Member Function Documentation

height()

unsigned int ConvolutionPyramid::height ( )

inline

Returns the height expected for the input texture.

Returns

the height expected

process()

void ConvolutionPyramid::process

(

const Texture &

texture

)

Filter a given input texture.

Parameters

texture

the GPU ID of the texture

Note

Those filters are separable, and could be applied in two passes (vertical and horizontal) to reduce the texture fetches count.

resize()

void ConvolutionPyramid::resize	(	uint	width,
		uint	height
	)

Resize the internal buffers.

Parameters

width	the new width
height	the new height

setFilters()

void ConvolutionPyramid::setFilters	(	const float	h1[5],
		float	h2,
		const float	g[3]
	)

Setup the filters parameters for a given task.

Parameters

h1	the 5 coefficients of h1
h2	the multiplier coefficient h2
g	the 3 coefficients of g

Note

See Convolution Pyramids, Farbman et al., 2011 for the notation details.

texture()

const Texture * ConvolutionPyramid::texture ( ) const

inline

The GPU ID of the filter result.

Returns

the ID of the result texture

width()

unsigned int ConvolutionPyramid::width ( )

inline

Returns the width expected for the input texture.

Returns

the width expected

---

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

• src/engine/processing/ConvolutionPyramid.hpp
• src/engine/processing/ConvolutionPyramid.cpp