AR tag bit pattern.

Defines the pattern bit size (maximal 8).

Physical size of marker in meters.

Name of the data asset that will be created.

The ID and version of the training job to save. If not set, the default package will be used (ai.intrinsic) and asset_name will be used as name.

Axis of rotation. Does not need to be normalized.

Angle of rotation specified in radians.

A unique identifier of a specific camera. This identifier determines which underlying driver is used to access a camera.

A list of camera settings. Camera settings are applied in the order in which they are defined. The specification of camera settings is optional.

A list of per-sensor configurations. These include per-sensor calibrations and poses. The specification of sensor configurations is optional.

The intrinsic camera parameters. If they are not presented, we are dealing with an uncalibrated camera.

The distortion params of the camera. If they are not present, the camera is either uncalibrated or it provides undistorted images (in all likelihood synthetic one).

The SFNC conforming feature name.

GenICam: IInteger - maps to a slider with value, min, max, and increment

GenICam: IFloat - maps to a slider with value, min, and max

GenICam: IBoolean - maps to a check box

GenICam: IString - maps to an edit box showing a string

GenICam: IEnumeration - maps to a drop down box

GenICam: ICommand

The SFNC conforming feature name.

The SFNC conforming feature name.

The configuration of the camera from which images are to be captured from.

The timeout which is used for retrieving frames from the underlying driver implementation. If this timeout is implemented by the underlying camera driver, it will not spend more than the specified time when waiting for new frames. The timeout should be greater than the combined exposure and processing time. Processing times can be roughly estimated as a value between 10 - 50 ms. The timeout just serves as an upper limit to prevent blocking calls within the camera driver. In case of intermittent network errors users can try to increase the timeout. The default timeout (if unspecified) of 500 ms works well in common setups.

Optional. Request data only for the following sensor ids (i.e. transmit mask). Empty returns all sensor images.

Additional post processing options for each sensor. If present, sensor post processing options will override the global post processing options.

If set the capture result will be stored in the specified key value store location, which will also be returned instead of the capture result itself.

Timestamp at which the images were captured. Value matches each sensor_image's acquisition_time, and is duplicated for capture result processing convenience.

Images for each of the captured sensors

Duration needed to capture all images excluding post processing like HDR. Not all camera drivers will provide it.

Number of squares of the ChArUco board in x direction.

Number of squares of the ChArUco board in y direction.

Length (width and height) of each square of the chessboard of the ChArUco board in m.

Length (width and height) of each ArUco marker on the ChArUco board in m.

Dictionary for the ArUco markers that are being used.

Pose estimation config to be trained. All file paths specified in this config must match the target paths in the file_references below.

A generic proto with user-specific custom data attached to the training job. The proto will be returned to the caller in the TrainingJobStatus message.

Name of the data asset that will be created.

The ID of the one-shot base model asset that will be used to populate the pose estimator perception model. This is only used for one-shot pose estimators creation.

Type of pose estimator to be trained. If POSE_ESTIMATOR_TYPE_UNSPECIFIED is selected, the params and inference_params fields of pose_estimation_config must be set. If any other pose estimator type is selected, the params and inference_params fields will be auto-filled with default values according to the pose estimator type.

The identifier of the camera that is to be described.

Image resolution along x axis, in pixels.

Image resolution along y axis, in pixels.

First radial distortion coefficient.

Second radial distortion coefficient.

Third radial distortion coefficient.

First tangetial distortion coefficient.

Second tangetial distortion coefficient.

Fourth radial distortion coefficient.

Fifth radial distortion coefficient.

Sixt radial distortion coefficient.

The allowed values for the enum.

Required symmetry center.

Required symmetry axis and symmetry angle represented in angle-axis representation. Currently we only support one symmetry axis.

The range of allowed floating point values.

The increment (or delta) between two neighboring allowed values.

The unit corresponding to the value. We typically default to SI units.

Recursion depth of the icosahedron. 0 means no subdivision with possible vertex indices from [0,12).

Vertex index for the respective recursion depth.

The image buffer's encoding.

Stores the underlying semantic pixel data which is stored in an image.

The image's number of channels, i.e. components per pixel.

The underlying data type of the pixels.

The width and height (cols and rows) of an image. This parameter is optional and must be specified when images are uncompressed, i.e. when encoding is ENCODING_UNSPECIFIED.

The actual underlying data.

The packing type of the pixels. If not present, implementations should default to PACKING_TYPE_INTERLEAVED.

The range of allowed integer values.

The increment (or delta) between two neighboring allowed values.

The unit corresponding to the value. We typically use SI units.

Focal length in horizontal direction. Required. The focal length is specified in pixels.

Focal length in vertical direction. Required. The focal length is specified in pixels.

Horizontal location of the principal point in the image plane. Required. The principal point is specified in pixels.

Vertical location of the principal point in the image plane. Required. The principal point is specified in pixels.

The dimensions of the image plane.

Minimum rotation of the camera along longitude (in degree).

Maximum rotation of the camera along longitude (in degree).

Minimum rotation of the camera along latitude (in degree).

Maximum rotation of the camera along latitude (in degree).

Minimum camera roll angle (in degrees).

Maximum camera roll angle (in degrees).

Optional. Reference rotation frame of the model to be trained. These parameters only influence from which angles the model is seen during training not the resulting poses during inference. The rotation is parameterized by angles around the x,y,z axis in degree.

A list of unique camera identifiers. These can be used to create a CameraConfig which is required during camera creation.

Path to a mesh. Preferrably a glb file.

Initial translation, rotation, and scaling of the mesh in target coordinates.

Pose of the object relative to the root object.

Between 0..1.

Class id of the estimate.

Visibility score for the pose. Between 0..1. The visibility score indicates the proportion of the object that is visible, signifying their foreground presence. It helps in prioritizing the selection of parts during, for example, the picking process. It is defined as: Visible pixels / (Visible pixels + Occluded pixels)

A list of targets that the pose estimator should detect.

Pose-estimator-specific training configuration.

Pose-estimator-specific inference configuration.

Intrinsic camera parameters. This parameter might be required depending on the selected pose estimator; see individual pose estimator types for details.

String indicating the source of the config, such as a GCS path or ml job name.

Minimal distance of the object's origin to the camera in meters.

Maximal distance of the object's origin to the camera in meters.

Encodes icosahedron vertices which represent individual views.

Encodes all views of a hemisphere.

View space parameterization using latitude/longitude parameterization.

The identifier of the camera from which setting access is read.

The setting name. The setting name must be defined by the Standard Feature Naming Conventions (SFNC) which is part of the GenICam standard.

The access mode of a specific setting.

The identifier of the camera from which setting properties are read.

The setting name. The setting name must be defined by the Standard Feature Naming Conventions (SFNC) which is part of the GenICam standard.

The properties of a specific setting. For numeric properties (integer and float), these include the valid range of parameters, their increment and their unit. For enumerations the properties contain the values (as strings) which can be set.

The identifier of the camera from which the setting is read.

The setting name. The setting name must be defined by the Standard Feature Naming Conventions (SFNC) which is part of the GenICam standard.

The current values of a specific setting. The following types of settings are supported int64, double, bool, string or an enum value.

Required symmetry center.

Required normalized symmetry axis.

The name of the training job to save.

The ID of the data asset that will be updated. If package is not set, the default package will be used (ai.intrinsic).

The ID of the created data asset.

ComponentIDValue or 0 for GenICam, sensor number for other cameras (e.g. IPS PU)

Transforms the sensor data into the reference coordinate system of the camera.

Sensor camera parameters.

Configuration (calibration) for the sensor

Timestamp at which the image was captured

Frame data

Specifies how the returned image should be encoded.

If true, the returned image will be distorted. If false, the returned image might be undistorted depending whether the camera has undistortion parameters defined.

Optionally crops the image. Note that cropping happens after undistortion but before resizing.

Resize dimensions.

Resize width. Height is set automatically to aspect ratio.

Resize height. Width is set automatically to aspect ratio.

ComponentIDValue or 0 for Genicam, sensor number for other cameras (e.g. IPS PU)

A human readable display name.

Sensor camera parameters from factory

Transforms the sensor data into the reference coordinate system of the camera.

Supported pixel types

Dimensions of sensor output

Is sensor currently disabled

Required symmetry center.

Id will be reported back to the client to identify this object. May encode class and object id in the following way: {class}/{object_id}, for example "chair/grey_chair0".

Restricts the pose range (orientation and position) of the object. Note: This serves as a hint during training but won't guarantee that all poses are from that pose range; poses outside the pose range may still be detected and returned to the client.

Stores symmetry type and related parameters such as symmetry axis.

Progress indicator of the training job between [0, 1]. Not all training jobs provide this information.

Custom data attached by the caller during CreateTrainingJob.

Timestamp when this job was submitted for training.

Asset ID that is associated with the training job.

The identifier of the camera on which settings are updated.

The settings values which will be used to for a parameter update.

Repeated list of potential issues with the provided training configuration.

go/unspecified-enum

uint8_t

uint16_t

uint32_t

int8_t

int16_t

int32_t

float

double

uncompressed data, go/unspecified-enum

JPG compression

PNG compression

WEBP compression

YUV420P colorspace

go/unspecified-enum

Pixel channels are stored interleaved, e.g., rgbrgbrgb… for RGB images.

Pixel channels are stored in separate planes, e.g., rrrr…rrrbbb…bbbggg…ggg

go/unspecified-enum

Used for 1- and 3-channel intensity images.

Used for depth images.

Used for 3D point clouds.

Used for normals.