BoxData Struct

Human-readable class label for the detection (for example, “person” or “cup”).

This is resolved from the model’s class index by the provider and is appended with a score so UI elements can show both category and confidence to the user.

World-space position of the visualized detection’s quad center.

This is computed downstream from 2D box coordinates and the environment depth, and is used by ObjectDetectionVisualizer to place the label and the bounding-box mesh in 3D.

World-space rotation applied to the rendered detection quad so it faces the user’s camera.

The visualizer derives this from the current passthrough pose and is helpful for billboarding labels and quads without extra per-frame math in user scripts.

World-space scale of the rendered quad that represents the detection bounds.

The visualizer derives width/height from camera intrinsics and depth so box size roughly matches the real object footprint at the estimated distance from the headset.