Face Detection System Development

Development of Face Detection Systems

Face detection is the first stage of almost any face pipeline. Task: find all faces in an image and return bounding boxes with confidence score. Sounds simple, but real conditions add complexity: small faces at distance, profile angles, partial occlusions, poor lighting, masks.

Modern Face Detectors

SCRFD (Sample and Computation Redistribution for Face Detection, InsightFace) — current best for speed/quality ratio. SCRFD-10GF: 95.2% AP on WiderFace Hard.

RetinaFace — classic with landmark detection (5 points: eyes, nose, mouth corners). Used for alignment before face recognition.

YOLOv8 fine-tuned on WiderFace — universal variant for custom requirements.

from insightface.app import FaceAnalysis
import cv2

# InsightFace: detection + landmark detection
app = FaceAnalysis(allowed_modules=['detection'])
app.prepare(ctx_id=0, det_size=(640, 640))

def detect_faces(image_path: str) -> list[dict]:
    img = cv2.imread(image_path)
    faces = app.get(img)

    results = []
    for face in faces:
        results.append({
            'bbox': face.bbox.astype(int).tolist(),     # [x1, y1, x2, y2]
            'confidence': float(face.det_score),
            'landmarks': face.kps.astype(int).tolist()  # 5 keypoints
        })
    return results

Small Face Detection

Standard detectors lose faces smaller than 16x16 pixels. For surveillance cameras at large distances:

• Image tiling: split image into overlapping tiles, detect on each, merge results via NMS
• SAHI (Slicing Aided Hyper Inference) — automatic tiling with merge:

from sahi import AutoDetectionModel
from sahi.predict import get_sliced_prediction

model = AutoDetectionModel.from_pretrained(
    model_type='yolov8',
    model_path='face_detector.pt',
    confidence_threshold=0.3
)

result = get_sliced_prediction(
    image='crowd.jpg',
    detection_model=model,
    slice_height=512,
    slice_width=512,
    overlap_height_ratio=0.2,
    overlap_width_ratio=0.2
)

Performance on Different Hardware

Masked Face Detection

Pandemic created a separate class of tasks — masked face detection. MAFA dataset contains 35,806 annotated masked faces. Fine-tuning standard detector on MAFA+WiderFace: AP on masked faces increases from 65% to 89%.