74 lines
2.1 KiB
Python
74 lines
2.1 KiB
Python
"""Face detection module using UniFace (RetinaFace + ArcFace)."""
|
|
|
|
import cv2
|
|
import numpy as np
|
|
from dataclasses import dataclass, field
|
|
from pathlib import Path
|
|
from typing import List, Tuple
|
|
|
|
from uniface.detection import RetinaFace
|
|
from uniface.recognition import ArcFace
|
|
|
|
|
|
@dataclass
|
|
class FaceData:
|
|
"""Detected face with embedding."""
|
|
|
|
id: int
|
|
frame_path: Path
|
|
frame_index: int
|
|
bbox: Tuple[int, int, int, int] # (x1, y1, x2, y2)
|
|
embedding: np.ndarray
|
|
confidence: float
|
|
landmarks: np.ndarray = field(default_factory=lambda: np.empty(0))
|
|
|
|
|
|
class FaceDetector:
|
|
"""Face detector using RetinaFace + ArcFace via UniFace."""
|
|
|
|
def __init__(self, confidence_threshold: float = 0.7):
|
|
self.detector = RetinaFace(confidence_threshold=confidence_threshold)
|
|
self.recognizer = ArcFace()
|
|
|
|
def detect_faces(self, frame_path: Path, frame_index: int) -> List[FaceData]:
|
|
"""Detect faces in a frame and generate embeddings.
|
|
|
|
Args:
|
|
frame_path: Path to the frame image
|
|
frame_index: Index of the frame in the video
|
|
|
|
Returns:
|
|
List of FaceData objects with bboxes, embeddings, and confidence
|
|
"""
|
|
image = cv2.imread(str(frame_path))
|
|
if image is None:
|
|
raise ValueError(f"Could not read image: {frame_path}")
|
|
|
|
detections = self.detector.detect(image)
|
|
|
|
faces = []
|
|
for i, det in enumerate(detections):
|
|
bbox = tuple(int(v) for v in det.bbox) # (x1, y1, x2, y2)
|
|
confidence = det.confidence
|
|
landmarks = det.landmarks
|
|
|
|
embedding = self.recognizer.get_normalized_embedding(image, landmarks)
|
|
embedding = embedding.flatten()
|
|
|
|
faces.append(
|
|
FaceData(
|
|
id=frame_index * 100 + i,
|
|
frame_path=frame_path,
|
|
frame_index=frame_index,
|
|
bbox=bbox,
|
|
embedding=embedding,
|
|
confidence=confidence,
|
|
landmarks=landmarks,
|
|
)
|
|
)
|
|
|
|
return faces
|
|
|
|
def close(self):
|
|
"""Release resources."""
|
|
pass
|