itasc.tracking_ultrack.retracker

Constrained greedy retracker using the shared linker similarity score.

Each unlocked target cell is matched to a reference cell by the same scoring.similarity_score (area ratio + centroid-corrected IoU - distance) used by the linker and the Extend tool, gated only by centroid distance.

Matching is greedy best-first rather than a global minimum-cost assignment: a target always takes its highest-scoring still-free reference. A global linear_sum_assignment minimises the total cost and will trade a cell’s obvious best match away to lower another cell’s cost, which in practice handed cells a far, worse reference; best-first never does that.

Functions

retrack_frame_constrained(ref_labels, ...[, ...])

Remap target IDs by linker similarity without changing locked targets.
itasc.tracking_ultrack.retracker.retrack_frame_constrained(ref_labels, target_labels, locked_target_ids, max_dist_px=50.0, reserved_ids=None, *, area_weight=1.0, iou_weight=1.0, distance_weight=0.05)[source]

Remap target IDs by linker similarity without changing locked targets.

Target cells whose ID is in locked_target_ids keep their existing IDs. Those IDs, plus any reserved_ids, are protected from assignment to unlocked target cells even if a matching reference cell exists.

Unlocked target cells are matched to available reference cells greedily in descending similarity_score order: each target takes its best still-free reference. A pair is only eligible when its centroid distance is <= max_dist_px; area ratio and centroid-corrected IoU contribute as soft score terms (no hard gate), matching the Extend tool’s behaviour.

Return type:

ndarray

Parameters: