Classification

Warning

The AI Accelerator requires ROS 2 Humble, so PolyScope X 10.12.1 is the latest compatible release. PolyScope X 10.13 and later use ROS 2 Jazzy and are not currently supported.

Load a classification model, set an image ROI, and run inference. Requires ark_init() (see General).

Typical call order: ark_init() → ark_load_classification_model → ark_set_classification_params → ark_infer_classification.

ark_load_classification_model(model)

ark_load_classification_model(model)

Loads a classification model on the AI Accelerator.

Arguments

  • model (string): Model filename or path relative to /workspaces/isaac_ros-dev/data/models/ (that prefix is prepended automatically)

Returns

  • ArkError — typically ArkError(False) (the preamble does not inspect the service response for errors)

Errors / failure modes

  • Requires ark_init() (uses /load_model client)

  • Missing or invalid model may still return ArkError(False); verify by checking inference results

ROS

  • Service: ARK_ROS2Namespace + "/load_model"

  • Type: ur_ark_ros_common_msgs/srv/LoadModel

  • Load request includes min_confidence=1.0 and default ROI struct(x=240.0, y=140.0, width=160.0, height=120.0) (override inference ROI with ark_set_classification_params)

Call order

After ark_init(), before setting ROI / inferring.

Example

ark_init()
ark_load_classification_model("classification_active/model.onnx")

ark_set_classification_params(roi=struct(x=240.0, y=140.0, width=160.0, height=120.0))

ark_set_classification_params(roi=struct(x=240.0, y=140.0, width=160.0, height=120.0))

Stores the region of interest used by the next ark_infer_classification call (global ark_classification_roi).

Arguments

  • roi (struct, default struct(x=240.0, y=140.0, width=160.0, height=120.0)): Crop rectangle in image coordinates (x, y, width, height)

Returns

  • void

Errors / failure modes

  • ROI must match what the model / training pipeline expects; a mismatched ROI is a common cause of wrong classes (see the classification example docs)

ROS

  • None directly (local URScript global only). Applied on the next /infer_classification call.

Call order

After load; before or between inference calls when the ROI changes.

Example

ark_set_classification_params(struct(x=200.0, y=120.0, width=200.0, height=160.0))

ark_infer_classification()

ark_infer_classification()

Runs classification on the current image using the stored ROI.

Arguments

  • None

Returns

  • struct(error=ArkError(...), class_name=<string>, confidence=<float>)

    • On success: class_name and confidence are the first label/confidence from the result

    • On failure: class_name="", confidence=0.0, error.error=True

Errors / failure modes

  • error.error=True if the service reports success=False, or if labels/confidences lists are empty

  • Service call timeout: 5.0 seconds

  • Requires ark_init() and a prior ark_set_classification_params (or the global ROI remains unset until set)

ROS

  • Service: ARK_ROS2Namespace + "/infer_classification"

  • Type: ur_ark_ros_common_msgs/srv/InferClassification

  • Request: roi_query=ark_classification_roi

Call order

ark_init() → load model → set ROI → infer.

Example

ark_init()
ark_set_overlay("classification")
ark_load_classification_model("classification_active/model.onnx")
ark_set_classification_params(struct(x=240.0, y=140.0, width=160.0, height=120.0))
res = ark_infer_classification()
if res.error.error == False:
  textmsg(res.class_name + " " + to_str(res.confidence))
end