Expand description
Decode YuNet-style face detection tensors into fixed detection boxes.
Runtime-agnostic: caller supplies model output floats (from ORT/tract/other). Weight file is MIT (OpenCV Zoo YuNet) — PermissiveReady, not commercial-gated.
Structs§
- FaceBox
- One face box in normalized image coordinates (0..1) or pixels if scale applied.
Functions§
- face_
nms - Greedy IoU NMS into
out(caller buffer); returns kept count. - yunet_
decode_ detections - YuNet multi-stride heads emit prior-decoded boxes; this decoder accepts a flat
list of rows
[x, y, w, h, score, ...pad]withstridefloats per row.