Last active
November 29, 2025 16:20
-
-
Save parsapoorsh/c2bc9ad9e87dc748170233bf0fd16981 to your computer and use it in GitHub Desktop.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| from dataclasses import dataclass | |
| from pathlib import Path | |
| from typing import Tuple | |
| import cv2 | |
| import numpy as np | |
| @dataclass | |
| class FaceDetection: | |
| face_xy1: Tuple[int, int] | |
| face_wh: Tuple[int, int] | |
| right_eye: Tuple[int, int] | |
| left_eye: Tuple[int, int] | |
| nose_tip: Tuple[int, int] | |
| mouth_right: Tuple[int, int] | |
| mouth_left: Tuple[int, int] | |
| confidence: float | |
| @property | |
| def face_xy2(self): | |
| fx, fy = self.face_xy1 | |
| fw, fh = self.face_wh | |
| return (fx + fw, | |
| fy + fh) | |
| @property | |
| def face_center(self): | |
| fx, fy = self.face_xy1 | |
| fw, fh = self.face_wh | |
| return (fx + (fw // 2), | |
| fy + (fh // 2)) | |
| class FaceDetector: | |
| def __init__(self, model_path: Path, score_threshold: float = 0.7): | |
| assert 0 <= score_threshold <= 1 | |
| self.model_path = Path(model_path).absolute().resolve() | |
| if not self.model_path.is_file(): | |
| raise FileNotFoundError(self.model_path) | |
| self.detector = cv2.FaceDetectorYN.create( | |
| model=self.model_path, | |
| config="", | |
| input_size=(320, 320), | |
| score_threshold=score_threshold, | |
| nms_threshold=0.4, | |
| top_k=5000, | |
| backend_id=cv2.dnn.DNN_BACKEND_OPENCV, | |
| target_id=cv2.dnn.DNN_TARGET_CPU, | |
| ) | |
| def detect(self, img: np.ndarray) -> Tuple[FaceDetection]: | |
| ih, iw, _ = img.shape | |
| self.detector.setInputSize((iw, ih)) | |
| __, faces = self.detector.detect(img) | |
| if faces is None: | |
| faces = [] | |
| result = tuple( | |
| FaceDetection( | |
| face_xy1=tuple(map(int, f[:2])), | |
| face_wh=tuple(map(int, f[2:4])), | |
| right_eye=tuple(map(int, f[4:6])), | |
| left_eye=tuple(map(int, f[6:8])), | |
| nose_tip=tuple(map(int, f[8:10])), | |
| mouth_right=tuple(map(int, f[10:12])), | |
| mouth_left=tuple(map(int, f[12:14])), | |
| confidence=float(f[14]), | |
| ) | |
| for f in faces | |
| ) | |
| return result | |
| @staticmethod | |
| def visualize( | |
| img: np.ndarray, | |
| faces: Tuple[FaceDetection], | |
| radius: int = 4, | |
| thickness: int = 4, | |
| ) -> np.ndarray: | |
| BGR_BLUE = (255, 0, 0) | |
| BGR_RED = (0, 0, 255) | |
| BGR_GREEN = (0, 255, 0) | |
| BGR_CYAN = (255, 255, 0) | |
| BGR_YELLOW = (0, 255, 255) | |
| vimg = img.copy() | |
| for f in faces: | |
| cv2.rectangle(vimg, f.face_xy1, f.face_xy2, BGR_GREEN, thickness) | |
| cv2.circle(vimg, f.face_center, radius, BGR_CYAN, thickness) | |
| cv2.circle(vimg, f.right_eye, radius, BGR_RED, thickness) | |
| cv2.circle(vimg, f.left_eye, radius, BGR_RED, thickness) | |
| cv2.circle(vimg, f.nose_tip, radius, BGR_YELLOW, thickness) | |
| cv2.line(vimg, f.mouth_left, f.mouth_right, BGR_BLUE, thickness) | |
| cv2.putText( | |
| vimg, | |
| f"{f.confidence:.2f}", | |
| (f.face_xy1[0], f.face_xy1[1] - 10), | |
| cv2.FONT_HERSHEY_SIMPLEX, | |
| 1, | |
| BGR_GREEN, | |
| thickness, | |
| ) | |
| return vimg | |
| if __name__ == "__main__": | |
| # download the model from: | |
| # https://github.com/opencv/opencv_zoo/raw/main/models/face_detection_yunet/face_detection_yunet_2023mar.onnx | |
| input_img_path = Path("test.jpg") | |
| output_image_path = input_img_path.absolute().with_stem(input_img_path.stem + "-visualized_faces") | |
| model_path = Path("face_detection_yunet_2023mar.onnx") | |
| fd = FaceDetector(model_path=model_path) | |
| if not input_img_path.is_file(): | |
| raise FileNotFoundError(input_img_path) | |
| img = cv2.imread(input_img_path) | |
| faces = fd.detect(img) | |
| print(f"Found {len(faces)} face(s)") | |
| vimg = fd.visualize(img, faces) | |
| cv2.imwrite( | |
| filename=output_image_path, | |
| img=vimg, | |
| ) |
Author
parsapoorsh
commented
Nov 27, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment