The Image class stores image with customizable rows, cols, channels, dtype and device. More...

#include <Image.h>

Inheritance diagram for open3d::t::pipelines::odometry::Image:

Public Types
enum class	InterpType { Nearest = 0 , Linear = 1 , Cubic = 2 , Lanczos = 3 , Super = 4 }
	Image interpolation algorithms. More...
Public Types inherited from open3d::t::geometry::Geometry
enum class	GeometryType { Unspecified = 0 , PointCloud = 1 , VoxelGrid = 2 , Octree = 3 , LineSet = 4 , MeshBase = 5 , TriangleMesh = 6 , HalfEdgeTriangleMesh = 7 , Image = 8 , RGBDImage = 9 , TetraMesh = 10 , OrientedBoundingBox = 11 , AxisAlignedBoundingBox = 12 }
	Specifies possible geometry types. More...

Public Member Functions
	Image (int64_t rows=0, int64_t cols=0, int64_t channels=1, core::Dtype dtype=core::Float32, const core::Device &device=core::Device("CPU:0"))
	Constructor for image.
	Image (const core::Tensor &tensor)
	Construct from a tensor. The tensor won't be copied and memory will be shared.
virtual	~Image () override
Image &	Clear () override
	Clear image contents by resetting the rows and cols to 0, while keeping channels, dtype and device unchanged.
bool	IsEmpty () const override
	Returns true if rows * cols * channels == 0.
Image &	Reset (int64_t rows=0, int64_t cols=0, int64_t channels=1, core::Dtype dtype=core::Float32, const core::Device &device=core::Device("CPU:0"))
	Reinitialize image with new parameters.
int64_t	GetRows () const
	Get the number of rows of the image.
int64_t	GetCols () const
	Get the number of columns of the image.
int64_t	GetChannels () const
	Get the number of channels of the image.
core::Dtype	GetDtype () const
	Get dtype of the image.
core::Device	GetDevice () const override
	Get device of the image.
core::Tensor	At (int64_t r, int64_t c) const
	Get pixel(s) in the image.
core::Tensor	At (int64_t r, int64_t c, int64_t ch) const
	Get pixel(s) in the image. Returns a tensor with shape {}.
void *	GetDataPtr ()
	Get raw buffer of the Image data.
const void *	GetDataPtr () const
	Get raw buffer of the Image data.
core::Tensor	AsTensor () const
	Returns the underlying Tensor of the Image.
Image	To (const core::Device &device, bool copy=false) const
	Transfer the image to a specified device.
Image	To (core::Dtype dtype, bool copy=false, utility::optional< double > scale=utility::nullopt, double offset=0.0) const
	Returns an Image with the specified `dtype`.
Image	Clone () const
	Returns copy of the image on the same device.
Image &	LinearTransform (double scale=1.0, double offset=0.0)
	Function to linearly transform pixel intensities in place.
Image	RGBToGray () const
	Converts a 3-channel RGB image to a new 1-channel Grayscale image.
Image	Resize (float sampling_rate=0.5f, InterpType interp_type=InterpType::Nearest) const
	Return a new image after resizing with specified interpolation type.
Image	Dilate (int kernel_size=3) const
	Return a new image after performing morphological dilation.
Image	Filter (const core::Tensor &kernel) const
	Return a new image after filtering with the given kernel.
Image	FilterBilateral (int kernel_size=3, float value_sigma=20.0f, float distance_sigma=10.0f) const
	Return a new image after bilateral filtering.
Image	FilterGaussian (int kernel_size=3, float sigma=1.0f) const
	Return a new image after Gaussian filtering.
std::pair< Image, Image >	FilterSobel (int kernel_size=3) const
	Return a pair of new gradient images (dx, dy) after Sobel filtering.
Image	PyrDown () const
	Return a new downsampled image with pyramid downsampling.
Image	PyrDownDepth (float diff_threshold, float invalid_fill=0.f) const
	Edge and invalid value preserving downsampling by 2 specifically for depth images.
Image	ClipTransform (float scale, float min_value, float max_value, float clip_fill=0.0f) const
	Return new image after scaling and clipping image values.
Image	CreateVertexMap (const core::Tensor &intrinsics, float invalid_fill=0.0f)
	Create a vertex map from a depth image using unprojection.
Image	CreateNormalMap (float invalid_fill=0.0f)
	Create a normal map from a vertex map.
Image	ColorizeDepth (float scale, float min_value, float max_value)
	Colorize an input depth image (with Dtype UInt16 or Float32).
core::Tensor	GetMinBound () const
	Compute min 2D coordinates for the data (always {0, 0}).
core::Tensor	GetMaxBound () const
	Compute max 2D coordinates for the data ({rows, cols}).
open3d::geometry::Image	ToLegacy () const
	Convert to legacy Image type.
std::string	ToString () const
	Text description.
Public Member Functions inherited from open3d::t::geometry::Geometry
virtual	~Geometry ()
GeometryType	GetGeometryType () const
	Returns one of registered geometry types.
int	Dimension () const
	Returns whether the geometry is 2D or 3D.
std::string	GetName () const
void	SetName (const std::string &name)
Public Member Functions inherited from open3d::core::IsDevice
	IsDevice ()=default
virtual	~IsDevice ()=default
bool	IsCPU () const
bool	IsCUDA () const
bool	IsSYCL () const

Static Public Member Functions
static Image	FromLegacy (const open3d::geometry::Image &image_legacy, const core::Device &Device=core::Device("CPU:0"))
	Create from a legacy Open3D Image.

Static Public Attributes
static constexpr bool	HAVE_IPP = false
	Do we use IPP for accelerating image processing operations?

Protected Attributes
core::Tensor	data_

Additional Inherited Members
Protected Member Functions inherited from open3d::t::geometry::Geometry
	Geometry (GeometryType type, int dimension)
	Parameterized Constructor.

Detailed Description

The Image class stores image with customizable rows, cols, channels, dtype and device.

Member Enumeration Documentation

◆ InterpType

enum class open3d::t::geometry::Image::InterpType

strong

Image interpolation algorithms.

Enumerator
Nearest	Nearest neighbors interpolation.
Linear	Bilinear interpolation.
Cubic	Bicubic interpolation.
Lanczos	Lanczos filter interpolation.
Super	Super sampling interpolation (only downsample).

Constructor & Destructor Documentation

◆ Image() [1/2]

open3d::t::geometry::Image::Image	(	int64_t	rows = 0,
		int64_t	cols = 0,
		int64_t	channels = 1,
		core::Dtype	dtype = core::Float32,
		const core::Device &	device = core::Device("CPU:0") )

Constructor for image.

Row-major storage is used, similar to OpenCV. Use (row, col, channel) indexing order for image creation and accessing. In general, (r, c, ch) are the preferred variable names for consistency, and avoid using width, height, u, v, x, y for coordinates.

Parameters

rows	Number of rows of the image, i.e. image height. `rows` must be non-negative.
cols	Number of columns of the image, i.e. image width. `cols` must be non-negative.
channels	Number of channels of the image. E.g. for RGB image, channels == 3; for grayscale image, channels == 1. `channels` must be greater than 0.
dtype	Data type of the image.
device	Device where the image is stored.

◆ Image() [2/2]

open3d::t::geometry::Image::Image ( const core::Tensor & tensor )

Construct from a tensor. The tensor won't be copied and memory will be shared.

Parameters

tensor Tensor of the image. The tensor must be contiguous. The tensor must be 2D (rows, cols) or 3D (rows, cols, channels).

◆ ~Image()

virtual open3d::t::geometry::Image::~Image ( )

inlineoverridevirtual

Member Function Documentation

◆ AsTensor()

core::Tensor open3d::t::geometry::Image::AsTensor ( ) const

inline

Returns the underlying Tensor of the Image.

◆ At() [1/2]

core::Tensor open3d::t::geometry::Image::At	(	int64_t	r,
		int64_t	c ) const

inline

Get pixel(s) in the image.

If channels == 1, returns a tensor with shape {}, otherwise returns a tensor with shape {channels,}. The returned tensor is a slice of the image's tensor, so when modifying the slice, the original tensor will also be modified.

◆ At() [2/2]

core::Tensor open3d::t::geometry::Image::At	(	int64_t	r,
		int64_t	c,
		int64_t	ch ) const

inline

Get pixel(s) in the image. Returns a tensor with shape {}.

◆ Clear()

Image & open3d::t::geometry::Image::Clear ( )

inlineoverridevirtual

Clear image contents by resetting the rows and cols to 0, while keeping channels, dtype and device unchanged.

Implements open3d::t::geometry::Geometry.

◆ ClipTransform()

Image open3d::t::geometry::Image::ClipTransform	(	float	scale,
		float	min_value,
		float	max_value,
		float	clip_fill = 0.0f ) const

Return new image after scaling and clipping image values.

This is typically used for preprocessing a depth image. Images of shape (rows, cols, channels=1) and Dtypes UInt16 and Float32 are supported. Each pixel will be transformed by

x = x / scale
x = x < min_value ? clip_fill : x
x = x > max_value ? clip_fill : x

Use INFINITY, NAN or 0.0 (default) for clip_fill.

Returns: Transformed image of type Float32, with out-of-range pixels clipped and assigned the clip_fill value.

◆ Clone()

Image open3d::t::geometry::Image::Clone ( ) const

inline

Returns copy of the image on the same device.

◆ ColorizeDepth()

Image open3d::t::geometry::Image::ColorizeDepth	(	float	scale,
		float	min_value,
		float	max_value )

Colorize an input depth image (with Dtype UInt16 or Float32).

The image values are divided by scale, then clamped within [min_value, max_value] and finally converted to an RGB image using the Turbo colormap as a lookup table.

Returns: Full color depth map of shape (rows, cols, channels=3) and Dtype UInt8.

◆ CreateNormalMap()

Image open3d::t::geometry::Image::CreateNormalMap ( float invalid_fill = 0.0f )

Create a normal map from a vertex map.

The input vertex map image should be of shape (rows, cols, channels=3) and Dtype Float32. This uses a cross product of \(V(r, c+1)-V(r, c)\) and \(V(r+1, c)-V(r, c)\). The input vertex map is expected to be the output of CreateVertexMap. You may need to start with a filtered depth image (e.g. with FilterBilateral) to obtain good results.

Parameters

invalid_fill Value to fill in for invalid points, and to fill-in if no valid neighbor is found. Use NAN, INFINITY or 0.0 (default). Must be consistent with clip_fill in CreateVertexMap.

Returns: Normal map of shape (rows, cols, channels=3) and Dtype Float32, with invalid normals assigned the invalid_fill value.

◆ CreateVertexMap()

Image open3d::t::geometry::Image::CreateVertexMap	(	const core::Tensor &	intrinsics,
		float	invalid_fill = 0.0f )

Create a vertex map from a depth image using unprojection.

The input depth (of shape (rows, cols, channels=1) and Dtype Float32) is expected to be the output of ClipTransform.

Parameters

intrinsics	Pinhole camera model of (3, 3) in Float64.
invalid_fill	Value to fill in for invalid depths. Use NAN, INFINITY or 0.0 (default). Must be consistent with `clip_fill` in ClipTransform.

Returns: Vertex map of shape (rows, cols, channels=3) and Dtype Float32, with invalid points assigned the invalid_fill value.

◆ Dilate()

Image open3d::t::geometry::Image::Dilate ( int kernel_size = 3 ) const

Return a new image after performing morphological dilation.

Supported datatypes are UInt8, UInt16 and Float32 with {1, 3, 4} channels. An 8-connected neighborhood is used to create the dilation mask.

Parameters

kernel_size An odd number >= 3.

◆ Filter()

Image open3d::t::geometry::Image::Filter ( const core::Tensor & kernel ) const

Return a new image after filtering with the given kernel.

◆ FilterBilateral()

Image open3d::t::geometry::Image::FilterBilateral	(	int	kernel_size = 3,
		float	value_sigma = 20.0f,
		float	distance_sigma = 10.0f ) const

Return a new image after bilateral filtering.

Parameters

value_sigma	Standard deviation for the image content.
distance_sigma	Standard deviation for the image pixel positions.

Note: CPU (IPP) and CUDA (NPP) versions use different algorithms and will give different results:
CPU uses a round kernel (radius = floor(kernel_size / 2)),
while CUDA uses a square kernel (width = kernel_size).
Make sure to tune parameters accordingly.

◆ FilterGaussian()

Image open3d::t::geometry::Image::FilterGaussian	(	int	kernel_size = 3,
		float	sigma = 1.0f ) const

Return a new image after Gaussian filtering.

Parameters

kernel_size	Odd numbers >= 3 are supported.
sigma	Standard deviation of the Gaussian distribution.

◆ FilterSobel()

std::pair< Image, Image > open3d::t::geometry::Image::FilterSobel ( int kernel_size = 3 ) const

Return a pair of new gradient images (dx, dy) after Sobel filtering.

Parameters

kernel_size Sobel filter kernel size, either 3 or 5.

◆ FromLegacy()

Image open3d::t::geometry::Image::FromLegacy	(	const open3d::geometry::Image &	image_legacy,
		const core::Device &	Device = core::Device("CPU:0") )

static

Create from a legacy Open3D Image.

◆ GetChannels()

int64_t open3d::t::geometry::Image::GetChannels ( ) const

inline

Get the number of channels of the image.

◆ GetCols()

int64_t open3d::t::geometry::Image::GetCols ( ) const

inline

Get the number of columns of the image.

◆ GetDataPtr() [1/2]

void * open3d::t::geometry::Image::GetDataPtr ( )

inline

Get raw buffer of the Image data.

◆ GetDataPtr() [2/2]

const void * open3d::t::geometry::Image::GetDataPtr ( ) const

inline

Get raw buffer of the Image data.

◆ GetDevice()

core::Device open3d::t::geometry::Image::GetDevice ( ) const

inlineoverridevirtual

Get device of the image.

Implements open3d::t::geometry::Geometry.

◆ GetDtype()

core::Dtype open3d::t::geometry::Image::GetDtype ( ) const

inline

Get dtype of the image.

◆ GetMaxBound()

core::Tensor open3d::t::geometry::Image::GetMaxBound ( ) const

inline

Compute max 2D coordinates for the data ({rows, cols}).

◆ GetMinBound()

core::Tensor open3d::t::geometry::Image::GetMinBound ( ) const

inline

Compute min 2D coordinates for the data (always {0, 0}).

◆ GetRows()

int64_t open3d::t::geometry::Image::GetRows ( ) const

inline

Get the number of rows of the image.

◆ IsEmpty()

bool open3d::t::geometry::Image::IsEmpty ( ) const

inlineoverridevirtual

Returns true if rows * cols * channels == 0.

Implements open3d::t::geometry::Geometry.

◆ LinearTransform()

Image & open3d::t::geometry::Image::LinearTransform	(	double	scale = 1.0,
		double	offset = 0.0 )

inline

Function to linearly transform pixel intensities in place.

\(image = scale * image + offset\).

Parameters

scale	First multiply image pixel values with this factor. This should be positive for unsigned dtypes.
offset	Then add this factor to all image pixel values.

Returns: Reference to self.

◆ PyrDown()

Image open3d::t::geometry::Image::PyrDown ( ) const

Return a new downsampled image with pyramid downsampling.

The returned image is formed by a chained Gaussian filter (kernel_size = 5, sigma = 1.0) and a resize (ratio = 0.5) operation.

Returns: Half sized downsampled depth image.

◆ PyrDownDepth()

Image open3d::t::geometry::Image::PyrDownDepth	(	float	diff_threshold,
		float	invalid_fill = 0.f ) const

Edge and invalid value preserving downsampling by 2 specifically for depth images.

Only 1 channel Float32 images are supported. The returned image is formed by a chained Gaussian filter (kernel_size = 5, sigma = 1.0) and a resize (ratio = 0.5) operation.

Parameters

diff_threshold	The Gaussian filter averaging ignores neighboring values if the depth difference is larger than this value.
invalid_fill	The Gaussian filter ignores these values (may be specified as NAN, INFINITY or 0.0 (default)).

Returns: Half sized downsampled Float32 depth image.

◆ Reset()

Image & open3d::t::geometry::Image::Reset	(	int64_t	rows = 0,
		int64_t	cols = 0,
		int64_t	channels = 1,
		core::Dtype	dtype = core::Float32,
		const core::Device &	device = core::Device("CPU:0") )

Reinitialize image with new parameters.

◆ Resize()

Image open3d::t::geometry::Image::Resize	(	float	sampling_rate = 0.5f,
		InterpType	interp_type = InterpType::Nearest ) const

Return a new image after resizing with specified interpolation type.

Downsample if sampling rate is < 1. Upsample if sampling rate > 1. Aspect ratio is always preserved.

◆ RGBToGray()

Image open3d::t::geometry::Image::RGBToGray ( ) const

Converts a 3-channel RGB image to a new 1-channel Grayscale image.

Uses formula \(I = 0.299 * R + 0.587 * G + 0.114 * B\).

◆ To() [1/2]

Image open3d::t::geometry::Image::To	(	const core::Device &	device,
		bool	copy = false ) const

inline

Transfer the image to a specified device.

Parameters

device	The targeted device to convert to.
copy	If true, a new image is always created; if false, the copy is avoided when the original image is already on the targeted device.

◆ To() [2/2]

Image open3d::t::geometry::Image::To	(	core::Dtype	dtype,
		bool	copy = false,
		utility::optional< double >	scale = utility::nullopt,
		double	offset = 0.0 ) const

Returns an Image with the specified dtype.

Parameters

dtype	The targeted dtype to convert to.
copy	If true, a new tensor is always created; if false, the copy is avoided when the original tensor already has the targeted dtype.
scale	Optional scale value. This is 1./255 for UInt8 -> Float{32,64}, 1./65535 for UInt16 -> Float{32,64} and 1 otherwise
offset	Optional shift value. Default 0.

◆ ToLegacy()

open3d::geometry::Image open3d::t::geometry::Image::ToLegacy ( ) const

Convert to legacy Image type.

◆ ToString()

std::string open3d::t::geometry::Image::ToString ( ) const

Text description.

Field Documentation

◆ data_

core::Tensor open3d::t::geometry::Image::data_

protected

Internal data of the Image, represented as a contiguous 3D tensor of shape {rows, cols, channels}. Image properties can be obtained from the tensor.

◆ HAVE_IPP

bool open3d::t::geometry::Image::HAVE_IPP = false

staticconstexpr

Do we use IPP for accelerating image processing operations?

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

cpp/open3d/t/geometry/Image.h
cpp/open3d/t/geometry/Image.cpp

Public Types

Public Member Functions

Static Public Member Functions

Static Public Attributes

Protected Attributes

Additional Inherited Members

Detailed Description

Member Enumeration Documentation

◆ InterpType

Constructor & Destructor Documentation

◆ Image() [1/2]

◆ Image() [2/2]

◆ ~Image()

Member Function Documentation

◆ AsTensor()

◆ At() [1/2]

◆ At() [2/2]

◆ Clear()

◆ ClipTransform()

◆ Clone()

◆ ColorizeDepth()

◆ CreateNormalMap()

◆ CreateVertexMap()

◆ Dilate()

◆ Filter()

◆ FilterBilateral()

◆ FilterGaussian()

◆ FilterSobel()

◆ FromLegacy()

◆ GetChannels()

◆ GetCols()

◆ GetDataPtr() [1/2]

◆ GetDataPtr() [2/2]

◆ GetDevice()

◆ GetDtype()

◆ GetMaxBound()

◆ GetMinBound()

◆ GetRows()

◆ IsEmpty()

◆ LinearTransform()

◆ PyrDown()

◆ PyrDownDepth()

◆ Reset()

◆ Resize()

◆ RGBToGray()

◆ To() [1/2]

◆ To() [2/2]

◆ ToLegacy()

◆ ToString()

Field Documentation

◆ data_

◆ HAVE_IPP