nextcv.postprocessing.boxes ⚓︎

Bounding box postprocessing functions.

FUNCTION	DESCRIPTION
`iou_np`	Calculate intersection over union of target box with all others.
`nms_cpp`	Non-Maximum-Supression (NMS) algorithm to remove overlapping bounding boxes.
`nms_np`	Non-Maximum-Supression (NMS) algorithm to remove overlapping bounding boxes.
`wbf_cpp`	Fuse detections from multiple models using Weighted Box Fusion (WBF).
`wbf_np`	Reference WBF implementation from `ensemble_boxes` package.

nextcv.postprocessing.boxes.iou_np ⚓︎

iou_np(target_box: NDArray, boxes: NDArray, target_area: NDArray, areas: NDArray, inclusive: bool = False) -> NDArray

Calculate intersection over union of target box with all others.

PARAMETER	DESCRIPTION
`target_box`	The bounding box as (top_left_x, top_left_y, bottom_right_x, bottom_right_y). TYPE: `NDArray`
`boxes`	The bounding boxes as (top_left_x, top_left_y, bottom_right_x, bottom_right_y). TYPE: `NDArray`
`target_area`	The area of the target box. TYPE: `NDArray`
`areas`	The areas of all boxes. TYPE: `NDArray`
`inclusive`	If True, uses (x2 - x1 + 1) * (y2 - y1 + 1) for area. Enable only for integer, pixel-indexed boxes (VOC/OpenCV style). TYPE: `bool` DEFAULT: `False`

RETURNS	DESCRIPTION
`NDArray`	The intersection over union of the target box with all others.

nextcv.postprocessing.boxes.nms_cpp ⚓︎

nms_cpp(bboxes: NDArray, scores: NDArray, iou_thresh: float) -> NDArray[int32]

Non-Maximum-Supression (NMS) algorithm to remove overlapping bounding boxes.

PARAMETER	DESCRIPTION
`bboxes`	The bounding boxes as (top_left_x, top_left_y, bottom_right_x, bottom_right_y). TYPE: `NDArray`
`scores`	The confidence scores for each bounding box. TYPE: `NDArray`
`iou_thresh`	The threshold for intersection over union. TYPE: `float`

nextcv.postprocessing.boxes.nms_np ⚓︎

nms_np(bboxes: NDArray, scores: NDArray, iou_thresh: float) -> NDArray

Non-Maximum-Supression (NMS) algorithm to remove overlapping bounding boxes.

PARAMETER	DESCRIPTION
`bboxes`	The bounding boxes as (top_left_x, top_left_y, bottom_right_x, bottom_right_y). TYPE: `NDArray`
`scores`	The confidence scores for each bounding box. TYPE: `NDArray`
`iou_thresh`	The threshold for intersection over union. TYPE: `float`

RETURNS	DESCRIPTION
`NDArray`	The indices of the bounding boxes to keep.

nextcv.postprocessing.boxes.wbf_cpp ⚓︎

wbf_cpp(boxes_list: list[NDArray], scores_list: list[NDArray], labels_list: list[NDArray], weights: list[float] | None = None, iou_thr: float = 0.55, skip_box_thr: float = 0.0, conf_type: str = 'avg', allows_overflow: bool = False) -> tuple[NDArray[float32], NDArray[float32], NDArray[int32]]

Fuse detections from multiple models using Weighted Box Fusion (WBF).

PARAMETER	DESCRIPTION
`boxes_list`	Per-model boxes with shape (N_i, 4) in normalized xyxy format. TYPE: `list[NDArray]`
`scores_list`	Per-model confidence scores with shape (N_i,). TYPE: `list[NDArray]`
`labels_list`	Per-model integer labels with shape (N_i,). TYPE: `list[NDArray]`
`weights`	Optional per-model weights. Defaults to equal weights. TYPE: `list[float] \| None` DEFAULT: `None`
`iou_thr`	IoU threshold for clustering boxes into the same fused box. TYPE: `float` DEFAULT: `0.55`
`skip_box_thr`	Drop boxes with score lower than this threshold. TYPE: `float` DEFAULT: `0.0`
`conf_type`	Confidence mode. One of: "avg", "max", "box_and_model_avg", "absent_model_aware_avg". TYPE: `str` DEFAULT: `'avg'`
`allows_overflow`	If True, confidence can exceed 1.0 for avg mode. TYPE: `bool` DEFAULT: `False`

RETURNS	DESCRIPTION
`tuple[NDArray[float32], NDArray[float32], NDArray[int32]]`	Tuple of (fused_boxes, fused_scores, fused_labels).

nextcv.postprocessing.boxes.wbf_np ⚓︎

wbf_np(boxes_list: list[NDArray], scores_list: list[NDArray], labels_list: list[NDArray], **kwargs: object) -> tuple[NDArray[float32], NDArray[float32], NDArray[int32]]

Reference WBF implementation from ensemble_boxes package.