vision

Autoupload

Bases: Enum

Configures the auto-submitting of images to the Learning Loop

ALL class-attribute instance-attribute

ALL = 'all'

submit all images which are run through the detector

DISABLED class-attribute instance-attribute

DISABLED = 'disabled'

no auto-submitting

FILTERED class-attribute instance-attribute

FILTERED = 'filtered'

only submit images with novel detections and in an uncertainty range (this is the default)

Events

Name	Description
NEW_DETECTIONS	detection on an image is completed (argument: image)

CameraProjector

CameraProjector(
    camera_provider: CalibratableCameraProvider,
    *,
    interval: float = 1.0
)

The camera projector computes a grid of projected image points on the ground plane.

It is mainly used for visualization purposes.

CameraProvider

CameraProvider(*, persistence_key: str | None = None)

Bases: Generic[T], PersistentModule

A camera provider holds a dictionary of cameras and manages additions and removals.

The camera dictionary should not be modified directly but by using the camera provider's methods. This way respective events are emitted and consistency can be taken care of.

The camera provider also creates an HTTP route to access camera images.

Events

Name	Description
CAMERA_ADDED	a new camera has been added (argument: camera)
CAMERA_REMOVED	a camera has been removed (argument: camera id)
NEW_IMAGE	a new image is available (argument: image)

ConfigurableCamera

ConfigurableCamera(**kwargs)

Bases: Camera

A generalized interface for adjusting camera parameters like exposure, brightness or fps.

Detector

Detector(*, name: str | None = None)

Bases: ABC

A detector allows detecting objects in images.

It also holds an upload queue for sending images with uncertain results to an active learning infrastructure like the Zauberzeug Learning Loop.

detect abstractmethod async

detect(
    image: Image,
    *,
    autoupload: Autoupload = Autoupload.FILTERED,
    tags: list[str] | None = None,
    source: str | None = None,
    creation_date: datetime | str | None = None
) -> Detections | None

Runs detections on the image and fills the image.detections property.

The parameters tags, source, and creation_date are added as metadata if the image is uploaded.

Note that the hardware detector uses a lazy strategy to schedule the inference tasks. In particular a queue with a maximum size of 1 is used. This means if the detector is busy, the image is not processed immediately, but queued up. If the detect function is called again, the queued image is dropped and the new image is queued instead. In this case this method returns None.

:return: the detections found in the image. :raises DetectorException: if the detection fails.

fetch_detector_info abstractmethod async

fetch_detector_info() -> DetectorInfo

Retrieve information about the detector.

:return: information about the detector. :raises DetectorException: if the about information cannot be retrieved.

fetch_model_version_info abstractmethod async

fetch_model_version_info() -> ModelVersioningInfo

Retrieve information about the model version and versioning mode.

:return: the information about the model versioning as data class. :raises DetectorException: if the detector is not connected or the information cannot be retrieved.

set_model_version abstractmethod async

set_model_version(
    version: Literal["follow_loop", "pause"] | str
) -> None

Set the model version or versioning mode.

Set to "follow_loop" to automatically update the model version to the latest version in the learning loop. Set to "pause" to stop automatic updates and keep the current model version. Set to a version number (e.g. "1.2") to use a specific version.

:raises DetectorException: if the version control mode is not valid or the version could not be set.

upload abstractmethod async

upload(
    image: Image,
    *,
    tags: list[str] | None = None,
    source: str | None = None,
    creation_date: datetime | str | None = None
) -> None

Uploads the image to the Learning Loop.

The parameters tags, source, and creation_date are added as metadata. If the image has detections, they are also uploaded.

:raises DetectorException: if the upload fails.

Events

Name	Description
NEW_DETECTIONS	detection on an image is completed (argument: image)

DetectorHardware

DetectorHardware(
    *,
    port: int = 8004,
    name: str | None = None,
    auto_disconnect: bool = True
)

Bases: Detector

This detector communicates with a YOLO detector via Socket.IO.

It automatically connects and reconnects, submits and receives detections and sends images that should be uploaded to the Zauberzeug Learning Loop.

Note: Images must be smaller than MAX_IMAGE_SIZE bytes (default: 10 MB).

soft_reload async

soft_reload() -> None

Trigger a soft reload of the detector.

:raises DetectorException: if the communication fails.

DetectorSimulation

DetectorSimulation(
    camera_provider: CalibratableCameraProvider,
    *,
    noise: float = 1.0,
    detection_delay: float = 0.4,
    name: str | None = None
)

Bases: Detector

This detector simulates object detection.

It requires a camera provider in order to check visibility using the cameras' calibrations. Individual camera IDs can be added to a set of blocked_cameras to simulate occlusions during pytests. A list of simulated_objects can be filled to define what can be detected. An optional noise parameter controls the spatial accuracy in pixels. An optional detection_delay parameter simulates the time it takes to process an image.

MultiCameraProvider

MultiCameraProvider(*camera_providers: CameraProvider)

Bases: CameraProvider

A multi-camera provider combines multiple camera providers into one.

This is useful if another module requires a single camera provider but the robot has multiple camera sources like USB and WiFi cameras.

RtspCameraProvider

RtspCameraProvider(
    *,
    frame_rate: int = 6,
    jovision_profile: int = 0,
    network_interface: str | None = None,
    auto_scan: bool = True
)

Bases: CameraProvider[RtspCamera], PersistentModule

This module collects and provides real RTSP streaming cameras.

SimulatedCameraProvider

SimulatedCameraProvider(
    *,
    simulate_failing: bool = False,
    auto_scan: bool = True
)

Bases: CameraProvider[SimulatedCamera], PersistentModule

This module collects and simulates cameras and generates synthetic images.

In the current implementation the images only contain the camera ID and the current time.

scan_for_cameras async

scan_for_cameras() -> AsyncGenerator[str, Any]

Simulated device discovery by returning all camera's IDs.

If simulate_device_failure is set, disconnected cameras are returned with a fixed probability.

UsbCameraProvider

UsbCameraProvider(*, auto_scan: bool = True)

Bases: CameraProvider[UsbCamera], PersistentModule

This module collects and provides real USB cameras.

Camera devices are discovered through video4linux (v4l) and accessed with openCV. Therefore the program v4l2ctl and openCV (including python bindings) must be available.

camera_objects

camera_objects(
    camera_provider: CalibratableCameraProvider,
    camera_projector: CameraProjector,
    *,
    px_per_m: float = 10000,
    debug: bool = False,
    interval: float = 1.0
)

Bases: Group

This module provides a UI element for displaying cameras in a 3D scene.

It requires a camera provider as a source of cameras as well as a camera projector to show the current images projected on the ground plane. The px_per_m argument can be used to scale the camera frustums. With debug=True camera IDs are shown (default: False).