multiobject tracking

We consider the multi-object AI problem to encompass models that infer bounding boxes, labels (from a predefined set), scores, and tracks for objects in a video.

primitives

multiobject_tracking.InputType	Role alias for model/dataset inputs in multi-object-tracking protocol signatures.
multiobject_tracking.TargetType	Role alias for model/dataset targets in multi-object-tracking protocol signatures.
multiobject_tracking.DatumMetadataType	Role alias for per-datum metadata in multi-object-tracking protocol signatures.
multiobject_tracking.VideoStream	Semantic alias for a video input stream as Iterable [VideoFrame].
multiobject_tracking.MultiobjectTrackingTarget(...)	Set of tracked objects over a sequence of frames.
multiobject_tracking.DatumMetadata	Typed dictionary containing salient characteristics of video.
multiobject_tracking.SingleFrameObjectTrackingTarget(...)	Single-frame object-tracking target.
multiobject_tracking.VideoFrame(*args, **kwargs)	Contents of a single decoded video frame.

components

multiobject_tracking.Augmentation(*args, ...)	An augmentation protocol for the multi-object tracking AI problem.
multiobject_tracking.DataLoader(*args, **kwargs)	A dataloader protocol for the multi-object tracking AI problem providing batch-level data access.
multiobject_tracking.Dataset(*args, **kwargs)	A dataset protocol for multi-object tracking AI problem providing datum-level data access.
multiobject_tracking.FieldwiseDataset(*args, ...)	A specialization of Dataset protocol (i.e., a subprotocol) that specifies additional accessor methods for getting input, target, and metadata individually.
multiobject_tracking.Metric(*args, **kwargs)	A metric protocol for the multi-object tracking AI problem.
multiobject_tracking.Model(*args, **kwargs)	A model protocol for the multi-object tracking AI problem.