Module face_detector 

SCRFD Face Detection

Detects faces in images and returns bounding boxes with landmarks. Uses the SCRFD-10GF model via ONNX Runtime.

SCRFD outputs 9 tensors (3 strides x 3 types):

  Stride 8:  scores[12800,1], bboxes[12800,4], landmarks[12800,10]
  Stride 16: scores[3200,1],  bboxes[3200,4],  landmarks[3200,10]
  Stride 32: scores[800,1],   bboxes[800,4],   landmarks[800,10]

Bboxes are anchor-distance format (left, top, right, bottom distances from anchor center). Landmarks are (dx, dy) offsets from anchor center for 5 keypoints.

Structs

DetectedFace

A detected face with bounding box and landmarks

FaceDetector

SCRFD Face Detector

Functions

estimate_yaw_from_landmarks

Estimate yaw angle (degrees) from 5-point SCRFD landmarks.