SingleFrameObjectTrackingTarget

class maite.protocols.multiobject_tracking.SingleFrameObjectTrackingTarget(*args, **kwargs)

Single-frame object-tracking target.

This class is used to encode frame-level detections/tracks (both predictions and ground-truth labels) in the multi-object tracking AI problem.

Implementers must populate the following attributes:

Attributes:

boxesArrayLike

An array representing object detection boxes in a single frame with x0, y0, x1, y1 format and shape (N_DETECTIONS, 4)

labelsArrayLike

An array representing the integer labels associated with each detection box of shape (N_DETECTIONS,)

scores: ArrayLike

An array representing the scores associated with each box (of shape (N_DETECTIONS,) or (N_DETECTIONS, N_CLASSES).

track_ids: ArrayLike

An array of per-detection track identifiers with shape (N_DETECTIONS,). Use non-negative integer-like values for tracked detections, and -1 for detections that are not assigned to any track. NaN values are not supported.