Greedy non-maximum merging on oriented boxes. Mirrors :func:`_group_overlapping_boxes` but uses :func:`oriented_box_iou_batch`.

Calculate the Jaccard index (intersection over union) between two bounding boxes. If a gt object is marked as "iscrowd", a dt is allowed to m

Return a JSON-serializable equivalent of a NumPy scalar or array. Called as the ``default`` hook by :func:`json.dump`. Converts :clas

Create the small text box displaying line zone count. E.g. "out: 7". Args: text: The text to display. text_s

Random polygon mask. *num_vertices* is a direct complexity proxy: more vertices → more independent radius samples → jaggier boundary → more R

Per-mask `np.where` loop used as a ground-truth oracle.

Match predictions with target labels based on IoU levels. Args: predictions: Batch prediction. Describes a single image

Reference IoU that materialises the (N, M, H, W) overlap explicitly.

Normalise *i* to a 1-D row index for 1-D per-object fields. Handles: - Python int or np.integer scalar -> np.array([int(i)]) - boolean n

Return index pairs ``(i, j)`` whose axis-aligned envelopes overlap. Uses a fused boolean evaluation to halve peak transient memory compared to

Transform predictions into a list of predictions that can be used by the COCO evaluator.

Prepare targets and predictions for evaluation.

Scale one column's run list to a new height using a precomputed row map. Each output row is mapped to a source row via ``row_map``, which imp

Time MaskAnnotator on the compact detections (dense-skip path).

Time .area on the compact detections (used when dense timing is skipped).

Time boolean-index filtering on the compact detections (dense-skip path).

Update the attribute keeping track of which class IDs correspond to which class names. Assumes that class_names are only pro

Ensure that class_id is a 1d np.ndarray with (n, ) shape.

Ensure that confidence is a 1d np.ndarray with (n, ) shape.

Verify that for each optional field in the Detections, both instances either have the field set to None or both have it set to non-None value

Validate per-keypoint confidence: 2D ``np.ndarray`` with shape ``(n, m)``.

Validate per-keypoint visibility mask. Expects a 2D bool ``np.ndarray`` with shape ``(n, m)``.

Validates the parameters and result type for a given Vision-Language Model (VLM). Args: vlm: The VLM enum or string specifying the m

Validate that xyxy is a 2D np.ndarray with shape (N, 4). ```pycon >>> _validate_xyxy(np.array([[0, 0, 1, 1], [1, 1, 2, 2]])) ```

Start a new tracklet

Draws a table with the number of objects of each class that crossed each line. Args: frame: The image on which the table

Append detection data to the JSON file. Args: detections: The detection data. custom_data: Custom data to in

Returns the color as a BGRA tuple. Returns: BGRA tuple. Example: ```pycon >>> import su

Converts the Color instance to a hex string. Returns: The hexadecimal color string. Returns `#RRGGBBAA` if alpha is not

Returns the color as an RGBA tuple. Returns: RGBA tuple. Example: ```pycon >>> import s

Returns the point as a tuple of floats. Returns: The point as (x, y) floats.

Compute overlap metric between two bounding boxes. Supports standard IOU (intersection-over-union) and IOS (intersection-over-smaller-ar

Calculate the intersection over union (IoU) between detection bounding boxes (dt) and ground-truth bounding boxes (gt). Reference: https:

Apply greedy version of non-maximum merging per category to avoid detecting too many overlapping bounding boxes for a given object. Args

Computes a blur kernel size proportional to the shorter side of a bounding box. Args: x1: Left edge of the bounding box. y1:

Computes a pixelation size proportional to the shorter side of a bounding box. Args: x1: Left edge of the bounding box. y1:

Calculate optimal font scale based on image resolution. Adjusts font scale proportionally to the smallest dimension of the given image resolu

Actual compact footprint: peak bytes during CompactMask.from_dense().

Theoretical compact footprint: sum of all internal numpy buffer sizes.

Create predictions that test IoU+class matching behavior. For each ground truth detection, creates predictions with different patterns:

Calculate the 2D cross product (also known as the vector product or outer product) of the vector and a point, treated as vectors in 2

Theoretical dense footprint: raw numpy buffer size.

Actual dense footprint: peak bytes during (N, H, W) bool array alloc.

Draw a filled polygon on a scene. Args: scene: The scene to draw the polygon on. polygon: The polygon to be drawn, given as a lis

Draws a filled rectangle on an image. Args: scene: The scene on which the rectangle will be drawn rect: The rectangle to be

Draws a rectangle on an image. Args: scene: The scene on which the rectangle will be drawn rect: The rectangle to be drawn

Return ``np.dtype(bool)`` — always. Returns: ``np.dtype(bool)``. Examples: ```pycon >>> from sup

Start the per image evaluation on all images and keeep results in self.eval_imgs (a list of dictionaries).

Keep the largest connected component and any other components within a distance threshold. Distance can be absolute in pixels or relativ

Find all duplicate elements in the input sequence. Args: sequence: The input sequence. Returns: A list of duplicate ele

Create a Color instance from a BGRA tuple. Args: color_tuple: A tuple representing the color in BGRA format, where each

Creates a Classifications instance from a [clip](https://github.com/openai/clip) inference result. Args: clip_re

Parse bounding boxes from deepseek-vl2-formatted text, scale them to the specified resolution, and optionally filter by classes. The Dee

Creates a `sv.KeyPoints` instance from a [MediaPipe](https://github.com/google-ai-edge/mediapipe) pose landmark detection inf

Parse and scale bounding boxes from Qwen-3-VL style JSON output. Args: result: String containing the Qwen-3-VL JSON output.

Create a Color instance from an RGBA tuple. Args: color_tuple: A tuple representing the color in RGBA format, where each

Creates a Detections instance from [Segment Anything Model](https://github.com/facebookresearch/segment-anything) inference r