oakutils.nodes.models package

Submodules

oakutils.nodes.models.gaussian module

Module for creating gaussian models.

Functions

create_gaussian: Creates a gaussian model with a specified kernel size.

oakutils.nodes.models.gaussian.create_gaussian(pipeline: dai.Pipeline, input_link: dai.Node.Output, kernel_size: int = 3, shaves: int = 1, *, grayscale_out: bool | None = None) → dai.node.NeuralNetwork

Use to create a gaussian model with a specified kernel size.

Parameters:

pipeline (dai.Pipeline) – The pipeline to add the gaussian to
input_link (dai.node.XLinkOut) – The input link to connect to the gaussian node. Example: cam_rgb.preview.link Explicitly pass in the link as a non-called function.
kernel_size (int, optional) – The size of the gaussian kernel, by default 3 Must be an odd integer Must be between 3 and 15 (since these are the compiled model sizes)
shaves (int, optional) – The number of shaves to use, by default 1 Must be between 1 and 6
grayscale_out (bool, optional) – Whether or not to use graycale output, by default False

Returns:

The gaussian node

Return type:

dai.node.NeuralNetwork

oakutils.nodes.models.gftt module

Models for the gftt node.

Functions

create_gftt: Creates a gftt model as a node.

oakutils.nodes.models.gftt.create_gftt(pipeline: dai.Pipeline, input_link: dai.Node.Output, blur_kernel_size: int = 3, shaves: int = 1, *, use_blur: bool | None = None, grayscale_out: bool | None = None) → dai.node.NeuralNetwork

Use to create a gftt model with a specified kernel size.

Parameters:

pipeline (dai.Pipeline) – The pipeline to add the gftt to
input_link (dai.node.XLinkOut) – The input link to connect to the gftt node. Example: cam_rgb.preview.link Explicitly pass in the link as a non-called function.
blur_kernel_size (int, optional) – The size of the blur kernel, by default 3 Only used when use_blur is True Must be an odd integer Must be between 3 and 15 (since these are the compiled model sizes)
shaves (int, optional) – The number of shaves to use, by default 1 Must be between 1 and 6
use_blur (bool, optional) – Whether or not to use a blur before the gftt, by default False
grayscale_out (bool, optional) – Whether or not to use graycale output, by default False

Returns:

The gftt node

Return type:

dai.node.NeuralNetwork

oakutils.nodes.models.harris module

Models for the harris node.

Functions

create_harris: Creates a harris model as a node.

oakutils.nodes.models.harris.create_harris(pipeline: dai.Pipeline, input_link: dai.Node.Output, blur_kernel_size: int = 3, shaves: int = 1, *, use_blur: bool | None = None, grayscale_out: bool | None = None) → dai.node.NeuralNetwork

Use to create a harris model with a specified kernel size.

Parameters:

pipeline (dai.Pipeline) – The pipeline to add the harris to
input_link (dai.node.XLinkOut) – The input link to connect to the harris node. Example: cam_rgb.preview.link Explicitly pass in the link as a non-called function.
blur_kernel_size (int, optional) – The size of the blur kernel, by default 3 Only used when use_blur is True Must be an odd integer Must be between 3 and 15 (since these are the compiled model sizes)
shaves (int, optional) – The number of shaves to use, by default 1 Must be between 1 and 6
use_blur (bool, optional) – Whether or not to use a blur before the harris, by default False
grayscale_out (bool, optional) – Whether or not to use graycale output, by default False

Returns:

The harris node

Return type:

dai.node.NeuralNetwork

oakutils.nodes.models.hessian module

Models for the hessian node.

Functions

create_hessian: Creates a hessian model as a node.

oakutils.nodes.models.hessian.create_hessian(pipeline: dai.Pipeline, input_link: dai.Node.Output, blur_kernel_size: int = 3, shaves: int = 1, *, use_blur: bool | None = None, grayscale_out: bool | None = None) → dai.node.NeuralNetwork

Use to create a hessian model with a specified kernel size.

Parameters:

pipeline (dai.Pipeline) – The pipeline to add the hessian to
input_link (dai.node.XLinkOut) – The input link to connect to the hessian node. Example: cam_rgb.preview.link Explicitly pass in the link as a non-called function.
blur_kernel_size (int, optional) – The size of the blur kernel, by default 3 Only used when use_blur is True Must be an odd integer Must be between 3 and 15 (since these are the compiled model sizes)
shaves (int, optional) – The number of shaves to use, by default 1 Must be between 1 and 6
use_blur (bool, optional) – Whether or not to use a blur before the hessian, by default False
grayscale_out (bool, optional) – Whether or not to use graycale output, by default False

Returns:

The hessian node

Return type:

dai.node.NeuralNetwork

oakutils.nodes.models.laplacian module

Module for creating a laplacian model with a specified kernel size.

Functions

create_laplacian: Creates a laplacian model with a specified kernel size.

oakutils.nodes.models.laplacian.create_laplacian(pipeline: dai.Pipeline, input_link: dai.Node.Output, kernel_size: int = 3, blur_kernel_size: int = 3, shaves: int = 1, *, use_blur: bool | None = None, grayscale_out: bool | None = None) → dai.node.NeuralNetwork

Use to create a laplacian model with a specified kernel size.

Parameters:

pipeline (dai.Pipeline) – The pipeline to add the laplacian to
input_link (dai.node.XLinkOut) – The input link to connect to the laplacian node. Example: cam_rgb.preview.link Explicitly pass in the link as a non-called function.
kernel_size (int, optional) – The size of the laplacian kernel, by default 3 Must be an odd integer Must be between 3 and 15 (since these are the compiled model sizes)
blur_kernel_size (int, optional) – The size of the blur kernel, by default 3 Only used when use_blur is True Must be an odd integer Must be between 3 and 15 (since these are the compiled model sizes)
shaves (int, optional) – The number of shaves to use, by default 1 Must be between 1 and 6
use_blur (bool, optional) – Whether or not to use a blur before the laplacian, by default False
grayscale_out (bool, optional) – Whether or not to use graycale output, by default False

Returns:

The laplacian node

Return type:

dai.node.NeuralNetwork

oakutils.nodes.models.laserscan module

Models for the laserscan node.

Functions

create_laserscan: Creates a laserscan model as a node.

oakutils.nodes.models.laserscan.create_laserscan(pipeline: dai.Pipeline, input_link: dai.Node.Output, width: int = 10, scans: int = 1, shaves: int = 1) → dai.node.NeuralNetwork

Use to create a laserscan model with a specified width.

Parameters:

pipeline (dai.Pipeline) – The pipeline to add the laserscan to
input_link (dai.node.XLinkOut) – The input link to connect to the laserscan node. Example: stereo.depth.link Explicitly pass in the link as a non-called function.
width (int, optional) – The width of the laserscan, by default 10 Options are [5, 10, 20]
scans (int, optional) – The number of scans to use, by default 1 Options are [1, 3, 5] Scans are horizontal lines of depth data, each scan is sampled from a different row of the depth image, with even spacing. A center scan is always generated and is always the middle entry in the output scans.
shaves (int, optional) – The number of shaves to use, by default 1 Must be between 1 and 6

Returns:

The laserscan node

Return type:

dai.node.NeuralNetwork

Raises:

ValueError – If the width is invalid

oakutils.nodes.models.laserscan.get_laserscan(data: dai.NNData) → np.ndarray

Use to get the laserscan data from the output tensor.

Parameters:: data (dai.NNData) – The data from the output tensor
Returns:: The laserscan data
Return type:: np.ndarray

oakutils.nodes.models.point_cloud module

Module for creating a point cloud model onboard.

Functions

create_xyz_matrix: Use to create a constant reprojection matrix for the given camera matrix and image size.
create_point_cloud: Use to create a point_cloud model.

oakutils.nodes.models.point_cloud.create_point_cloud(pipeline: dai.Pipeline, depth_link: dai.Node.Output, calibration: CalibrationData, input_stream_name: str = 'xyz_to_pcl', shaves: int = 4) → tuple[dai.node.NeuralNetwork, dai.node.XLinkIn, Callable[[dai.Device], None]]

Use to create a point_cloud model with a specified kernel size.

Parameters:

pipeline (dai.Pipeline) – The pipeline to add the point_cloud to
depth_link (dai.Node.Output) – The output link of the depth node Example: stereo.depth Explicity pass the object without calling (i.e. not stereo.depth())
calibration (CalibrationData) – The calibration data for the camera
input_stream_name (str, optional) – The name of the input stream, by default “xyz_to_pcl”
shaves (int, optional) – The number of shaves to use, by default 4 Must be between 1 and 6

Returns:

dai.node.NeuralNetwork – The point_cloud node
dai.node.XLinkIn – The input link to connect to the point_cloud node.
Callable[[dai.Device], None] – Function to pass the device, which will start the point cloud generation

oakutils.nodes.models.point_cloud.create_xyz_matrix(width: int, height: int, camera_matrix: ndarray) → ndarray

Use to create a constant reprojection matrix for the given camera matrix and image size.

Note: This is for generating the input to the point cloud generation model.

Parameters:

width (int) – The width of the image
height (int) – The height of the image
camera_matrix (np.ndarray) – The camera matrix to use for the reprojection This should be a 3x3 matrix

Returns:

The reprojection matrix

Return type:

np.ndarray

oakutils.nodes.models.point_cloud.get_point_cloud_buffer(data: NNData, frame_size: tuple[int, int] = (640, 400), scale: float = 1000.0, *, remove_zeros: bool | None = None) → ndarray

Use to convert the raw data output from a neural network execution and converts it to a point cloud.

Parameters:

data (dai.NNData) – Raw data output from a neural network execution.
frame_size (tuple[int, int], optional) – The size of the buffer, by default (640, 400) Usually this will be the size of the depth frame. Which inherits its shape from the MonoCamera resolutions.
scale (float, optional) – The scale to apply to the point cloud, by default 1000.0 This will convert from mm to m.
remove_zeros (bool, optional) – Whether to remove zero points, by default None If None, then True is used Recommended to set to True to remove zero points Can speedup reading and filtering of the point cloud by up to 10x

Returns:

Point cloud buffer

Return type:

np.ndarray

oakutils.nodes.models.sobel module

Models for the sobel node.

Functions

create_sobel: Creates a sobel model as a node.

oakutils.nodes.models.sobel.create_sobel(pipeline: dai.Pipeline, input_link: dai.Node.Output, blur_kernel_size: int = 3, shaves: int = 1, *, use_blur: bool | None = None, grayscale_out: bool | None = None) → dai.node.NeuralNetwork

Use to create a sobel model with a specified kernel size.

Parameters:

pipeline (dai.Pipeline) – The pipeline to add the sobel to
input_link (dai.node.XLinkOut) – The input link to connect to the sobel node. Example: cam_rgb.preview.link Explicitly pass in the link as a non-called function.
blur_kernel_size (int, optional) – The size of the blur kernel, by default 3 Only used when use_blur is True Must be an odd integer Must be between 3 and 15 (since these are the compiled model sizes)
shaves (int, optional) – The number of shaves to use, by default 1 Must be between 1 and 6
use_blur (bool, optional) – Whether or not to use a blur before the sobel, by default False
grayscale_out (bool, optional) – Whether or not to use graycale output, by default False

Returns:

The sobel node

Return type:

dai.node.NeuralNetwork

Module contents

Functions for creating models for use in the OAK-D pipeline.

Submodules

gaussian: Module for creating gaussian models.
gftt: Module for creating gftt models.
harris: Module for creating harris models.
hessian: Module for creating hessian models.
laplacian: Module for creating laplacian models.
laserscan: Module for creating laserscan models.
point_cloud: Module for creating a point cloud model onboard.
sobel: Module for creating sobel models.

Functions

create_gaussian: Creates a gaussian model with a specified kernel size.
create_gftt: Creates a gftt model.
create_harris: Creates a harris model.
create_hessian: Creates a hessian model.
create_laplacian: Creates a laplacian model with a specified kernel size.
create_laserscan: Creates a laserscan model with a specified width.
create_point_cloud: Creates a point cloud model onboard.
create_sobel: Creates a sobel model with a specified kernel size.
create_xyz_matrix: Use to create a constant reprojection matrix for the given camera matrix and image size.
get_laserscan: Use to get the laserscan data from the laserscan model.
get_point_cloud_buffer: Use to get the point cloud buffer from the point cloud model.

oakutils.nodes.models.create_gaussian(pipeline: dai.Pipeline, input_link: dai.Node.Output, kernel_size: int = 3, shaves: int = 1, *, grayscale_out: bool | None = None) → dai.node.NeuralNetwork

Use to create a gaussian model with a specified kernel size.

Parameters:

pipeline (dai.Pipeline) – The pipeline to add the gaussian to
input_link (dai.node.XLinkOut) – The input link to connect to the gaussian node. Example: cam_rgb.preview.link Explicitly pass in the link as a non-called function.
kernel_size (int, optional) – The size of the gaussian kernel, by default 3 Must be an odd integer Must be between 3 and 15 (since these are the compiled model sizes)
shaves (int, optional) – The number of shaves to use, by default 1 Must be between 1 and 6
grayscale_out (bool, optional) – Whether or not to use graycale output, by default False

Returns:

The gaussian node

Return type:

dai.node.NeuralNetwork

oakutils.nodes.models.create_gftt(pipeline: dai.Pipeline, input_link: dai.Node.Output, blur_kernel_size: int = 3, shaves: int = 1, *, use_blur: bool | None = None, grayscale_out: bool | None = None) → dai.node.NeuralNetwork

Use to create a gftt model with a specified kernel size.

Parameters:

pipeline (dai.Pipeline) – The pipeline to add the gftt to
input_link (dai.node.XLinkOut) – The input link to connect to the gftt node. Example: cam_rgb.preview.link Explicitly pass in the link as a non-called function.
blur_kernel_size (int, optional) – The size of the blur kernel, by default 3 Only used when use_blur is True Must be an odd integer Must be between 3 and 15 (since these are the compiled model sizes)
shaves (int, optional) – The number of shaves to use, by default 1 Must be between 1 and 6
use_blur (bool, optional) – Whether or not to use a blur before the gftt, by default False
grayscale_out (bool, optional) – Whether or not to use graycale output, by default False

Returns:

The gftt node

Return type:

dai.node.NeuralNetwork

oakutils.nodes.models.create_harris(pipeline: dai.Pipeline, input_link: dai.Node.Output, blur_kernel_size: int = 3, shaves: int = 1, *, use_blur: bool | None = None, grayscale_out: bool | None = None) → dai.node.NeuralNetwork

Use to create a harris model with a specified kernel size.

Parameters:

pipeline (dai.Pipeline) – The pipeline to add the harris to
input_link (dai.node.XLinkOut) – The input link to connect to the harris node. Example: cam_rgb.preview.link Explicitly pass in the link as a non-called function.
blur_kernel_size (int, optional) – The size of the blur kernel, by default 3 Only used when use_blur is True Must be an odd integer Must be between 3 and 15 (since these are the compiled model sizes)
shaves (int, optional) – The number of shaves to use, by default 1 Must be between 1 and 6
use_blur (bool, optional) – Whether or not to use a blur before the harris, by default False
grayscale_out (bool, optional) – Whether or not to use graycale output, by default False

Returns:

The harris node

Return type:

dai.node.NeuralNetwork

oakutils.nodes.models.create_hessian(pipeline: dai.Pipeline, input_link: dai.Node.Output, blur_kernel_size: int = 3, shaves: int = 1, *, use_blur: bool | None = None, grayscale_out: bool | None = None) → dai.node.NeuralNetwork

Use to create a hessian model with a specified kernel size.

Parameters:

pipeline (dai.Pipeline) – The pipeline to add the hessian to
input_link (dai.node.XLinkOut) – The input link to connect to the hessian node. Example: cam_rgb.preview.link Explicitly pass in the link as a non-called function.
blur_kernel_size (int, optional) – The size of the blur kernel, by default 3 Only used when use_blur is True Must be an odd integer Must be between 3 and 15 (since these are the compiled model sizes)
shaves (int, optional) – The number of shaves to use, by default 1 Must be between 1 and 6
use_blur (bool, optional) – Whether or not to use a blur before the hessian, by default False
grayscale_out (bool, optional) – Whether or not to use graycale output, by default False

Returns:

The hessian node

Return type:

dai.node.NeuralNetwork

oakutils.nodes.models.create_laplacian(pipeline: dai.Pipeline, input_link: dai.Node.Output, kernel_size: int = 3, blur_kernel_size: int = 3, shaves: int = 1, *, use_blur: bool | None = None, grayscale_out: bool | None = None) → dai.node.NeuralNetwork

Use to create a laplacian model with a specified kernel size.

Parameters:

pipeline (dai.Pipeline) – The pipeline to add the laplacian to
input_link (dai.node.XLinkOut) – The input link to connect to the laplacian node. Example: cam_rgb.preview.link Explicitly pass in the link as a non-called function.
kernel_size (int, optional) – The size of the laplacian kernel, by default 3 Must be an odd integer Must be between 3 and 15 (since these are the compiled model sizes)
blur_kernel_size (int, optional) – The size of the blur kernel, by default 3 Only used when use_blur is True Must be an odd integer Must be between 3 and 15 (since these are the compiled model sizes)
shaves (int, optional) – The number of shaves to use, by default 1 Must be between 1 and 6
use_blur (bool, optional) – Whether or not to use a blur before the laplacian, by default False
grayscale_out (bool, optional) – Whether or not to use graycale output, by default False

Returns:

The laplacian node

Return type:

dai.node.NeuralNetwork

oakutils.nodes.models.create_laserscan(pipeline: dai.Pipeline, input_link: dai.Node.Output, width: int = 10, scans: int = 1, shaves: int = 1) → dai.node.NeuralNetwork

Use to create a laserscan model with a specified width.

Parameters:

pipeline (dai.Pipeline) – The pipeline to add the laserscan to
input_link (dai.node.XLinkOut) – The input link to connect to the laserscan node. Example: stereo.depth.link Explicitly pass in the link as a non-called function.
width (int, optional) – The width of the laserscan, by default 10 Options are [5, 10, 20]
scans (int, optional) – The number of scans to use, by default 1 Options are [1, 3, 5] Scans are horizontal lines of depth data, each scan is sampled from a different row of the depth image, with even spacing. A center scan is always generated and is always the middle entry in the output scans.
shaves (int, optional) – The number of shaves to use, by default 1 Must be between 1 and 6

Returns:

The laserscan node

Return type:

dai.node.NeuralNetwork

Raises:

ValueError – If the width is invalid

oakutils.nodes.models.create_point_cloud(pipeline: dai.Pipeline, depth_link: dai.Node.Output, calibration: CalibrationData, input_stream_name: str = 'xyz_to_pcl', shaves: int = 4) → tuple[dai.node.NeuralNetwork, dai.node.XLinkIn, Callable[[dai.Device], None]]

Use to create a point_cloud model with a specified kernel size.

Parameters:

pipeline (dai.Pipeline) – The pipeline to add the point_cloud to
depth_link (dai.Node.Output) – The output link of the depth node Example: stereo.depth Explicity pass the object without calling (i.e. not stereo.depth())
calibration (CalibrationData) – The calibration data for the camera
input_stream_name (str, optional) – The name of the input stream, by default “xyz_to_pcl”
shaves (int, optional) – The number of shaves to use, by default 4 Must be between 1 and 6

Returns:

dai.node.NeuralNetwork – The point_cloud node
dai.node.XLinkIn – The input link to connect to the point_cloud node.
Callable[[dai.Device], None] – Function to pass the device, which will start the point cloud generation

oakutils.nodes.models.create_sobel(pipeline: dai.Pipeline, input_link: dai.Node.Output, blur_kernel_size: int = 3, shaves: int = 1, *, use_blur: bool | None = None, grayscale_out: bool | None = None) → dai.node.NeuralNetwork

Use to create a sobel model with a specified kernel size.

Parameters:

pipeline (dai.Pipeline) – The pipeline to add the sobel to
input_link (dai.node.XLinkOut) – The input link to connect to the sobel node. Example: cam_rgb.preview.link Explicitly pass in the link as a non-called function.
blur_kernel_size (int, optional) – The size of the blur kernel, by default 3 Only used when use_blur is True Must be an odd integer Must be between 3 and 15 (since these are the compiled model sizes)
shaves (int, optional) – The number of shaves to use, by default 1 Must be between 1 and 6
use_blur (bool, optional) – Whether or not to use a blur before the sobel, by default False
grayscale_out (bool, optional) – Whether or not to use graycale output, by default False

Returns:

The sobel node

Return type:

dai.node.NeuralNetwork

oakutils.nodes.models.create_xyz_matrix(width: int, height: int, camera_matrix: ndarray) → ndarray

Use to create a constant reprojection matrix for the given camera matrix and image size.

Note: This is for generating the input to the point cloud generation model.

Parameters:

width (int) – The width of the image
height (int) – The height of the image
camera_matrix (np.ndarray) – The camera matrix to use for the reprojection This should be a 3x3 matrix

Returns:

The reprojection matrix

Return type:

np.ndarray

oakutils.nodes.models.get_laserscan(data: dai.NNData) → np.ndarray

Use to get the laserscan data from the output tensor.

Parameters:: data (dai.NNData) – The data from the output tensor
Returns:: The laserscan data
Return type:: np.ndarray

oakutils.nodes.models.get_point_cloud_buffer(data: NNData, frame_size: tuple[int, int] = (640, 400), scale: float = 1000.0, *, remove_zeros: bool | None = None) → ndarray

Use to convert the raw data output from a neural network execution and converts it to a point cloud.

Parameters:

data (dai.NNData) – Raw data output from a neural network execution.
frame_size (tuple[int, int], optional) – The size of the buffer, by default (640, 400) Usually this will be the size of the depth frame. Which inherits its shape from the MonoCamera resolutions.
scale (float, optional) – The scale to apply to the point cloud, by default 1000.0 This will convert from mm to m.
remove_zeros (bool, optional) – Whether to remove zero points, by default None If None, then True is used Recommended to set to True to remove zero points Can speedup reading and filtering of the point cloud by up to 10x

Returns:

Point cloud buffer

Return type:

np.ndarray