API Reference (auto)

BBoxXYWHNorm

Bases: BaseModel

Canonical TL-normalized bbox: [x, y, w, h] with invariants.

to_xyxy_px(b, W, H)

Convert canonical bbox to pixel XYXY coordinates.

from_xyxy_px(x1, y1, x2, y2, W, H)

Convert pixel XYXY coordinates to canonical bbox.

iou_xyxy(a, b)

Intersection over Union for two xyxy pixel-space boxes.

Returns 0.0 when there is no overlap (including edge-touching).

area_xyxy(b)

Area of a pixel-space xyxy rectangle.

Negative extents are clamped to zero to be robust to degenerate inputs.

intersect_xyxy(a, b)

Intersection rectangle of two xyxy boxes or None if disjoint/touching.

clamp_xyxy(b, W, H)

Clamp an xyxy box to image bounds [0,W] x [0,H].

Ordering is preserved so that the output always has x1 <= x2 and y1 <= y2.

to_xywh_px(bbox_norm, W, H)

Convert normalized TL-xywh to pixel TL-xywh using image dimensions.

Uses simple scaling without rounding to preserve fractional pixel information.

from_xywh_px(x, y, w, h, W, H)

Convert pixel TL-xywh to normalized TL-xywh using image dimensions.

Values are clamped to [0,1] to avoid float precision edge cases, with a minimal positive width/height enforced by the BBoxXYWHNorm model.

xyxy_to_xywh(b)

Convert pixel-space xyxy → pixel TL-xywh (no clamping).

xywh_to_xyxy(b)

Convert pixel TL-xywh → pixel-space xyxy (no clamping).

group_detections_by_frame(dets)

Group detections by frame_number preserving per-frame order.

Returns a dict with integer keys sorted in ascending order.

detection_xyxy_px(det, W, H)

Return pixel xyxy tuple for a Detection using image dimensions.

Prefers the canonical bbox_norm if present. If only the legacy bbox (pixel xywh) is provided, falls back to converting that to xyxy.