Generate label strings for each detection.

Format single sequence metrics as a table. Args: result: Sequence result. columns: Columns to include. Returns: Form

Save Kalman filter state before track is lost (ORU mechanism).

Remove dead or immature lost tracklets and get alive trackers that are within `maximum_frames_without_update` AND (it's mature OR it was

Remove dead or immature lost tracklets and get alive trackers that are within `maximum_frames_without_update` AND (it's mature OR it was

Extract anchor points from detections based on position setting. Args: detections: Detections object with xyxy boxes. Re

Get GT and tracker file paths for a sequence. Args: gt_dir: Ground truth directory (direct parent of sequences). tracker_dir: Tra

Return video duration in seconds and frame count using OpenCV.

Initialize supervision annotators based on display options.

Load detection model via inference-models. Args: model_id: Model identifier (e.g., 'rfdetr-nano' or 'workspace/project/version').

Create tracker instance from registry. Args: tracker_id: Registered tracker name (e.g., 'bytetrack', 'sort'). **kwargs: Tracker-s

Load TrackEval-derived reference metrics for a dataset.

Parse a sequence map file to get list of sequence names.

Remove tracker detections matched to distractor ground truth regions. Uses the Hungarian algorithm to match tracker detections against ALL ground

Resolve class names to integer IDs.

Return `(icon, suffix)` for the final printed line.

Return an absolute URL for a nav item when it maps to a concrete page.

Determine the total frame count from data if not explicitly provided.

Resolve a comma-separated string of track IDs to a list of integers. Args: track_ids_arg: Comma-separated string (e.g. `"1,3,5"`). `None`

Run tracking from pre-computed detections without frame source.

Run model inference and return sv.Detections.

Run a tracker on GT-derived detections and save flat MOT output files.

Run tracking with a frame source (video, webcam, images).

Pre-vectorization HOTA reference: per-alpha Python loop + dict id-remapping. Mirrors the pre-PR second-pass logic (commit 2ca10bd^) for different

Compute normalized direction vectors from track centers to detection centers. For each (track, detection) pair, computes the unit vector pointing

Restore state and apply virtual trajectory (ORU mechanism). Generates linear interpolation between last observation and new detection

ORU interpolation for XCYCSR representation. Generates time_gap predict+update cycles with virtual observations interpolated from the

ORU interpolation for XYXY representation. Same pattern as XCYCSR: time_gap predict+update cycles with factors 0 to (time_gap-1)/time

Validate that every selected sequence has required MOT files. This performs eager filesystem validation so configuration errors are r

Ensure ``params`` keys are valid ``__init__`` parameter names.

Return points unchanged.

Add the eval subcommand to the argument parser.

Add the track subcommand to the argument parser.

Aggregate CLEAR metrics across multiple sequences. Sums integer fields and MOTP_sum, then recomputes derived ratios from the totals. Matches

Convert an `[x1, y1, x2, y2]` bbox to a measurement vector. Args: bbox: Bounding box `[x1, y1, x2, y2]`. Returns:

Clamp velocity components to prevent degenerate predictions. Called before `predict` to ensure physical plausibility (e.g. non-negati

Download benchmark tracking datasets from the official GCP bucket. Downloads ZIP files into a persistent cache directory and extracts them in

Get current filter state for saving. Returns: Dictionary with x, P, and other matrices.

Get current bounding box estimate from Kalman filter. Returns: Current bounding box estimate `[x1, y1, x2, y2]`.

Get current bounding box estimate from the filter/state.

Serialize to JSON string. Args: indent: Indentation level for formatting. Defaults to `2`. Returns: JSON str

Main entry point for the trackers CLI.

Predict next bounding box position and advance missed-frame clock. Propagates the Kalman filter and advances `time_since_update`, `age`,

Return points unchanged.

Reset tracker state by clearing all tracks and resetting ID counter. Call this method when switching to a new video or scene.

Reset tracker state by clearing all tracks and resetting ID counter. Call this method when switching to a new video or scene.

Execute the tune command.

Save to a JSON file. Args: path: Destination file path.

Restore filter state from saved dictionary. Args: state: Dictionary from get_state().

Display a frame and check for quit key (q or ESC). Returns: True if quit was requested, False otherwise.

Convert to dictionary representation. Returns: Dictionary with all metric values.

Convert to dictionary representation. Returns: Dictionary with all metric values.

Update tracker state with new detections and return tracked objects. Performs Kalman filter prediction, IoU-based association, and initializes

Create a small test video with *n* frames.

Provide back-compat access with deprecation warnings for every symbol.

Lazy imports for evaluate functions to avoid circular imports.

Initialise BIoU with a configurable buffer ratio. Args: buffer_ratio: Non-negative relative margin to expand each box

Initialise the filter with the first detection. Args: bbox: First detection `[x1, y1, x2, y2]`.

Initialize Kalman filter. Args: dim_x: Dimension of state vector. dim_z: Dimension of measurement vector.

Initialize CMC. Args: cfg: Optional configuration. If None, defaults are used. Notes: - Detector/ex

Initialize tracklet with first detection. Args: initial_bbox: Initial bounding box `[x1, y1, x2, y2]`. state_estimato