Compute pairwise similarity between two sets of bounding boxes. Handles the empty-input edge case (returns a correctly-shaped zero ma

Instantiate any registered tracker by its short ID.

Parse argv through a fresh download subparser and return the namespace.

Build a single-frame `_MOTFrameData` with xywh boxes.

Estimate global affine transform H (2x3) from previous frame to current frame. Args: frame_bgr: Current frame in BGR for

Return a Console that writes to a StringIO buffer.

Apply a 2x3 affine camera-motion transform to a list of BoT-SORT tracklets. .. note:: This method is BoT-SORT-specific. It requir

Apply a 2x3 affine camera-motion transform **in place**. Delegates to :meth:`CMC.apply_batch` with ``[self]`` as the tracklet list. S

Look up registered tracker by name. Internal method used by CLI for tracker discovery and instantiation. Args: name: Tra

Yield numbered BGR frames from video files, webcams, network streams, or image directories. Args: source: Video file path, RTSP/HTTP

Create `HOTAMetrics` from a dictionary. Args: data: Dictionary with metric values. Returns: `HOTAMetrics` in

Convert to dictionary representation. Args: include_arrays: Whether to include per-alpha arrays. Defaults to `False`.

Random-noise BGR frame with dense texture suitable for feature tracking.

Build a minimal BoTSORTTracklet with controlled state.

Build a BoTSORTTracklet whose state is XYXYStateEstimator.

Compute HOTA metrics for multi-object tracking evaluation. HOTA (Higher Order Tracking Accuracy) evaluates both detection and association qua

Write detections for a frame in MOT format.

Return the next tracker ID (zero-indexed) and advance the internal counter.

Classify a frame source and extract metadata. The function inspects *source* without consuming any frames so it can be called before the main

Execute the download subcommand.

Remove dead or immature lost tracklets and return alive ones. A tracklet is kept if it is within ``maximum_frames_without_update`` **and**

Transform four box corners via R/t and return the enclosing axis-aligned box. Takes per-axis coordinate arrays (not a packed (N, 4) array) an

Convert bounding boxes from corner to center-scale-ratio format. Args: xyxy: Bounding boxes `[x_min, y_min, x_max, y_max]` with shape `(4

Fuse IoU similarity matrix with detection confidence scores. Following the original ByteTrack implementation, the IoU similarity is multiplie

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

Create an Optuna study, run trials, and return the best parameters. Returns: Dictionary mapping ``search_space`` parameter names

Extract an `[x1, y1, x2, y2]` bbox from the current filter state. Returns: Bounding box `[x1, y1, x2, y2]`.

Convert bounding boxes from center-scale-ratio to corner format. Args: xcycsr: Bounding boxes `[x_center, y_center, scale, aspect_ratio]`

Compose the single-line progress string.

Instantiate CMC, skipping the test if the method is unavailable in OpenCV.

Prepare GT and tracker data for metric evaluation. Compute IoU similarity matrices between GT and tracker detections for each frame, and rema

List all registered tracker names. Internal method used by CLI for help text and argument validation. Returns: Alphabeti

Restart this tracker instance's ID allocation from zero.

Tune tracker hyperparameters using Optuna. Args: tracker: Tracker ID to tune (e.g. bytetrack, sort). gt_dir: Directory of ground-

Update the tracker with detections from the current frame. This is the main per-frame entry point. Args: detect

Construct a concrete tracklet class with default constructor params.

Signal that the processing loop has ended. Must be called before leaving the `with` block so that `__exit__` can render the correct f

Evaluate multiple multi-object tracking results against ground truth. Computes standard multi-object tracking metrics (CLEAR MOT, HOTA, Identity)

Draw motion-compensated trace paths on the scene. Updates internal trajectory storage with new detection positions (converted to worl

Calculate the IoU (Intersection over Union) between two sets of boxes. Compute pairwise IoU between all boxes in boxes1 and boxes2, returning a

Compute Identity metrics for multi-object tracking evaluation. Identity metrics measure global ID consistency by finding the optimal one-to-o

Evaluate a single multi-object tracking result against ground truth. Computes standard multi-object tracking metrics (CLEAR MOT, HOTA, Identity) f

Predict next bounding box position. Returns: Predicted bounding box `[x1, y1, x2, y2]`.

Update the C-BIoU tracker with detections from the current frame. Runs the association pipeline with a distinct BIoU instance per step

Update tracks state with new detections and return tracked objects. Performs Kalman filter prediction, two-stage association (high then low

Convert bounding boxes from corner to center-width-height format. Args: xyxy: Bounding boxes `[x_min, y_min, x_max, y_max]` with shape `(

Build result dictionary from per-alpha arrays. Computes final metrics (DetA, DetRe, DetPr, HOTA, OWTA) and averages over alpha thresholds for

Introspect __init__ signature and docstring to build parameter metadata. Combines type hints, default values, and docstring descriptions into a

Format a metric value for display. Float metrics are displayed as percentages with 3 decimal places (e.g., 99.353 for MOTA=0.99353), matching

Associate detections to tracks based on IOU Args: iou_matrix: IOU cost matrix. detection_boxes: Detected bou

Get metric value from a SequenceResult. Args: result: The sequence result. col: Column name. Returns: The metric val

Add the tune subcommand to the argument parser.

Aggregate HOTA metrics across multiple sequences. Uses weighted averaging by HOTA_TP for association metrics, then recomputes detection and H

Return per-detection confidence, defaulting to all-ones when absent. When ``detections.confidence`` is ``None``, every detection is treated as

Load a MOT Challenge format file. Parse a text file in the standard MOT format where each line represents one detection with comma-separated

Return the first existing image path for ``frame_idx`` under ``frame_dir``. Args: frame_dir: Directory containing frame images (typically

Set Kalman filter parameters. Args: R: Measurement noise covariance matrix. Q: Process noise covariance matrix.

Download and cache MOT test data for a given dataset. Args: dataset_key: Key from DATASETS dict (e.g., "sportsmot_flat"). Returns:

Associate detections to tracks based on Similarity (IoU) using the Jonker-Volgenant algorithm approach with no initialization instead

Get the current accumulated transformation. Returns: The accumulated transformation, or IdentityTransformation if no

Load one MOT ``img1`` frame from ``{images_dir}/{seq_name}/img1/``.

Return a top-level parser with a subparsers group.

Resolve a comma-separated `--classes` value to a list of integer IDs. Each token is checked independently: if it parses as an `int` it is used

Resolve a comma-separated `--track-ids` value to a list of integer IDs. Args: track_ids_arg: Raw `--track-ids` string (e.g. `"1,3,5"`). `

Run a tracker on a MOT detection file and write predictions. Iterates every frame from 1 to the last frame in the detection file, feeding ``s

Advance tracker until at least one confirmed track exists.

Verify CLEAR metrics match expected values. Args: computed: CLEARMetrics object from our computation. expected: Expected metrics

Verify HOTA metrics match expected values. Args: computed: HOTAMetrics object from our computation. expected: Expected metrics di

Verify Identity metrics match expected values. Args: computed: IdentityMetrics object from our computation. expected: Expected me

Emit a UserWarning when a frame is passed to a tracker that ignores it. Subclasses that do not perform camera motion compensation should call

Aggregate Identity metrics across multiple sequences. Sums IDTP, IDFN, IDFP across sequences, then recomputes IDF1, IDR, IDP from the totals.

Get current bounding box estimate from the filter/state.

Factory for creating image directories with specified number of frames.

Load one BGR frame from a MOT ``img1`` directory. Raises: FileNotFoundError: When no file matches any supported stem/extension. O

Return filename stems to probe for ``frame_idx`` (most common layouts first).

Update tracklet state with a new bounding-box observation. Handles ORU: if the track was lost and is now observed again, generates a

Record one processed frame and refresh the display.

Convert bounding boxes from center-width-height to corner format. Args: xywh: Bounding boxes `[x_center, y_center, width, height]` with s

Shared geometry used by GIoU, DIoU, CIoU and other variants. Args: boxes_1: ``(N, 4)`` array in ``[x1, y1, x2, y2]`` format. boxe

Create an Optuna study, optionally with a seeded TPE sampler.

Build the parameter dict for the enqueued baseline trial. Uses each ``search_space`` key's ``__init__`` default. Args: tracker_class

Extract a scalar metric value from ``BenchmarkResult.aggregate``. Searches CLEAR, HOTA, and Identity metrics in order. Args: result: