Upload 42 files

.env

@@ -0,0 +1,5 @@
+DB_USER=postgres
+DB_PASSWORD=HOSAMfakher1!
+DB_HOST=db.lczwuudofcmpetlnqfwp.supabase.co
+DB_PORT=5432
+DB_NAME=postgres

Dockerfile

@@ -0,0 +1,18 @@
+FROM python:3.10-slim
+
+WORKDIR /app
+
+COPY requirements.txt .
+
+RUN apt-get update && apt-get install -y \
+    libgl1 \
+    libglib2.0-0 \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY . .
+
+EXPOSE 7860
+
+CMD ["uvicorn","app.main:app","--host","0.0.0.0","--port","7860"]

README.md

@@ -1,14 +1,91 @@
----
-title: Attendance System
-emoji: 📉
-colorFrom: purple
-colorTo: blue
-sdk: gradio
-sdk_version: 6.20.0
-python_version: '3.12'
-app_file: app.py
-pinned: false
-short_description: Attendance System
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Face Attendance API
+
+نظام حضور بالتعرف على الوجه، مبني على **FastAPI** + **PostgreSQL** + **YOLOv8 (كشف الوجه)** + **EdgeFace (embedding)**.
+
+## هيكل المشروع
+
+```
+face_attendance/
+├── app/
+│   ├── main.py                 # نقطة تشغيل التطبيق
+│   ├── config.py                # الإعدادات (DB, threshold, مسارات الموديلات)
+│   ├── api/
+│   │   ├── register.py         # POST /register  → تسجيل شخص جديد
+│   │   └── attendance.py       # POST /attendance → تسجيل الحضور
+│   ├── database/
+│   │   ├── database.py         # اتصال SQLAlchemy
+│   │   └── models.py           # جدول Person + جدول Attendance
+│   ├── services/
+│   │   ├── edgeface_model.py   # تعريف موديل EdgeFace (بدون تعديل)
+│   │   ├── embedding.py        # تحميل EdgeFace وحساب embedding
+│   │   ├── face_detector.py    # تحميل YOLO وكشف/تتبع الوجوه
+│   │   └── recognition.py      # مقارنة embeddings وتحديد الشخص
+│   ├── utils/
+│   │   └── preprocess.py       # تجهيز الصورة + cosine similarity
+│   └── weights/                # حط هنا ملفات الموديلات (.pt)
+├── requirements.txt
+└── .env.example
+```
+
+## 1) تجهيز البيئة
+
+```bash
+python -m venv venv
+source venv/bin/activate      # ويندوز: venv\Scripts\activate
+
+pip install -r requirements.txt
+```
+
+## 2) تجهيز PostgreSQL
+
+```sql
+CREATE DATABASE face_attendance;
+```
+
+انسخ `.env.example` إلى `.env` وعدّل البيانات حسب السيرفر بتاعك:
+
+```bash
+cp .env.example .env
+```
+
+الجداول (`person`, `attendance`) هتتعمل تلقائيًا أول ما السيرفر يشتغل، مفيش حاجة تعملها يدوي.
+
+## 3) حط ملفات الموديلات
+
+حط الملفين دول جوه `app/weights/`:
+- `YoloV8_Face.pt`
+- `edgeface_xxs_q.pt`
+
+(لو الاسم مختلف عندك، عدّله فى `app/config.py`)
+
+## 4) تشغيل السيرفر
+
+```bash
+uvicorn app.main:app --reload
+```
+
+هيفتح على: `http://127.0.0.1:8000/docs` (Swagger UI تلقائي لتجربة الـ APIs).
+
+## 5) استخدام الـ APIs
+
+### تسجيل شخص جديد
+```bash
+curl -X POST "http://127.0.0.1:8000/register/?name=Hossam"
+```
+هتفتح الكاميرا:
+- اضغط **c** لما وجهك يبقى واضح فى المربع الأخضر → يسجل فى الداتابيز.
+- اضغط **q** للإلغاء.
+
+### تسجيل الحضور (Real-time)
+```bash
+curl -X POST "http://127.0.0.1:8000/attendance/"
+```
+هتفتح الكاميرا وتتعرف على كل الوجوه اللي قدامها وتسجلهم فى جدول `attendance`
+(مرة واحدة بس لكل شخص فى اليوم). اضغط **q** لإنهاء الجلسة.
+
+## ملاحظات للتعديل لاحقًا
+
+- **threshold التعرف**: غيّره من `RECOGNITION_THRESHOLD` فى `app/config.py`.
+- **رقم الكاميرا**: غيّره من `CAMERA_INDEX` فى نفس الملف (لو عندك أكتر من كاميرا).
+- **تسجيل شخص من صورة بدل الكاميرا مباشرة**: عدّل `app/api/register.py` فقط، باقي الملفات مش هتتأثر.
+- **الترقية إلى pgvector** (بحث أسرع للـ embeddings عند زيادة عدد الأشخاص): غيّر نوع عمود `embedding` فى `app/database/models.py` من `ARRAY(Float)` إلى `Vector(512)` بعد تفعيل extension `pgvector` فى بوستجريس.

app/api/attendance.py

@@ -0,0 +1,115 @@
+import datetime
+import cv2
+import numpy as np
+
+from fastapi import APIRouter, Depends, HTTPException, UploadFile, File
+from sqlalchemy.orm import Session
+
+from app.database.database import get_db
+from app.database.models import Attendance
+from app.services.face_detector import detect_faces
+from app.services.embedding import get_embedding
+from app.services.recognition import load_all_persons, identify_person
+
+router = APIRouter(prefix="/attendance", tags=["Attendance"])
+
+
+@router.post("/")
+async def take_attendance(
+    file: UploadFile = File(...),
+    db: Session = Depends(get_db),
+):
+    """
+    Register attendance from an uploaded image.
+    """
+
+    known_persons = load_all_persons(db)
+
+    if not known_persons:
+        raise HTTPException(
+            status_code=400,
+            detail="لا يوجد أشخاص مسجلين"
+        )
+
+    # قراءة الصورة
+    contents = await file.read()
+
+    image = cv2.imdecode(
+        np.frombuffer(contents, np.uint8),
+        cv2.IMREAD_COLOR,
+    )
+
+    if image is None:
+        raise HTTPException(
+            status_code=400,
+            detail="الصورة غير صالحة"
+        )
+
+    # كشف جميع الوجوه
+    boxes = detect_faces(image)
+
+    if len(boxes) == 0:
+        raise HTTPException(
+            status_code=400,
+            detail="لم يتم العثور على أي وجه"
+        )
+
+    present = []
+
+    today = datetime.date.today()
+
+    for box in boxes:
+
+        x1, y1, x2, y2 = box
+
+        face_crop = image[y1:y2, x1:x2]
+
+        if face_crop.size == 0:
+            continue
+
+        embedding = get_embedding(face_crop)
+
+        person_id, name, score = identify_person(
+            embedding,
+            known_persons
+        )
+
+        if person_id is None:
+            present.append({
+                "name": "Unknown",
+                "score": round(float(score), 4)
+            })
+            continue
+
+        already_exists = (
+            db.query(Attendance)
+            .filter(
+                Attendance.person_id == person_id,
+                Attendance.date == today,
+            )
+            .first()
+        )
+
+        if not already_exists:
+
+            attendance = Attendance(
+                person_id=person_id,
+                date=today,
+                time=datetime.datetime.now().time(),
+                status="Present",
+            )
+
+            db.add(attendance)
+            db.commit()
+
+        present.append({
+            "id": person_id,
+            "name": name,
+            "score": round(float(score), 4),
+            "status": "Present",
+        })
+
+    return {
+        "count": len(present),
+        "results": present,
+    }

app/api/register.py

@@ -0,0 +1,87 @@
+import cv2
+import numpy as np
+
+from fastapi import APIRouter, Depends, HTTPException, UploadFile, File
+from sqlalchemy.orm import Session
+
+from app.database.database import get_db
+from app.database.models import Person
+from app.services.face_detector import detect_faces
+from app.services.embedding import get_embedding
+from app.services.face_quality import check_face_quality
+
+router = APIRouter(prefix="/register", tags=["Register"])
+
+
+@router.post("/")
+async def register_person(
+    name: str,
+    file: UploadFile = File(...),
+    db: Session = Depends(get_db),
+):
+    """
+    Register a new person from an uploaded image.
+    """
+
+    # قراءة الصورة
+    contents = await file.read()
+
+    image = cv2.imdecode(
+        np.frombuffer(contents, np.uint8),
+        cv2.IMREAD_COLOR,
+    )
+
+    if image is None:
+        raise HTTPException(
+            status_code=400,
+            detail="الصورة غير صالحة"
+        )
+
+    # كشف الوجه
+    boxes = detect_faces(image)
+
+    if len(boxes) == 0:
+        raise HTTPException(
+            status_code=400,
+            detail="لم يتم العثور على أي وجه"
+        )
+
+    # أول وجه فقط
+    x1, y1, x2, y2 = boxes[0]
+
+    face_crop = image[y1:y2, x1:x2]
+
+    if face_crop.size == 0:
+        raise HTTPException(
+            status_code=400,
+            detail="فشل استخراج الوجه"
+        )
+
+    # فحص الجودة
+    quality = check_face_quality(face_crop)
+
+    if not quality["ok"]:
+        raise HTTPException(
+            status_code=400,
+            detail=quality["message"]
+        )
+
+    # استخراج الـ Embedding
+    embedding = get_embedding(face_crop)
+
+    # حفظه في قاعدة البيانات
+    person = Person(
+        name=name,
+        embedding=embedding.tolist(),
+    )
+
+    db.add(person)
+    db.commit()
+    db.refresh(person)
+
+    return {
+        "id": person.id,
+        "name": person.name,
+        "quality": quality,
+        "message": "تم التسجيل بنجاح"
+    }

app/config.py

@@ -0,0 +1,30 @@
+import os
+from dotenv import load_dotenv
+
+load_dotenv()
+
+# ---------------- Database ----------------
+DB_USER = os.getenv("DB_USER", "postgres")
+DB_PASSWORD = os.getenv("DB_PASSWORD", "postgres")
+DB_HOST = os.getenv("DB_HOST", "localhost")
+DB_PORT = os.getenv("DB_PORT", "5432")
+DB_NAME = os.getenv("DB_NAME", "face_attendance")
+
+DATABASE_URL = (
+    f"postgresql+psycopg2://{DB_USER}:{DB_PASSWORD}@{DB_HOST}:{DB_PORT}/{DB_NAME}"
+)
+
+# ---------------- Models weights ----------------
+BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+WEIGHTS_DIR = os.path.join(BASE_DIR, "app", "weights")
+
+YOLO_FACE_WEIGHTS = os.path.join(WEIGHTS_DIR, "YoloV8_Face.pt")
+EDGEFACE_WEIGHTS = os.path.join(WEIGHTS_DIR, "edgeface_xxs_q.pt")
+EDGEFACE_MODEL_NAME = "edgeface_xxs_q"
+
+# ---------------- Recognition ----------------
+# لو الـ similarity بين الوجه واللي مخزن اقل من الرقم ده -> Unknown
+RECOGNITION_THRESHOLD = 0.25
+
+# ---------------- Camera ----------------
+CAMERA_INDEX = 0

app/database/database.py

@@ -0,0 +1,18 @@
+from sqlalchemy import create_engine
+from sqlalchemy.orm import sessionmaker, declarative_base
+
+from app.config import DATABASE_URL
+
+engine = create_engine(DATABASE_URL)
+SessionLocal = sessionmaker(autocommit=False, autoflush=False, bind=engine)
+
+Base = declarative_base()
+
+
+def get_db():
+    """Dependency بتفتح جلسة DB وتقفلها لوحدها بعد كل request"""
+    db = SessionLocal()
+    try:
+        yield db
+    finally:
+        db.close()

app/database/models.py

@@ -0,0 +1,41 @@
+from sqlalchemy import (
+    Column,
+    Integer,
+    String,
+    Float,
+    Date,
+    Time,
+    DateTime,
+    ForeignKey,
+    UniqueConstraint,
+    func,
+)
+from sqlalchemy.dialects.postgresql import ARRAY
+from sqlalchemy.orm import relationship
+
+from app.database.database import Base
+
+
+class Person(Base):
+    __tablename__ = "person"
+
+    id = Column(Integer, primary_key=True, index=True)
+    name = Column(String, nullable=False, index=True)
+    embedding = Column(ARRAY(Float), nullable=False)  # 512-d embedding
+    created_at = Column(DateTime(timezone=True), server_default=func.now())
+
+    attendances = relationship("Attendance", back_populates="person")
+
+
+class Attendance(Base):
+    __tablename__ = "attendance"
+    # شخص واحد يتسجل مرة واحدة بس فى اليوم
+    __table_args__ = (UniqueConstraint("person_id", "date", name="uq_person_date"),)
+
+    id = Column(Integer, primary_key=True, index=True)
+    person_id = Column(Integer, ForeignKey("person.id"), nullable=False)
+    date = Column(Date, nullable=False)
+    time = Column(Time, nullable=False)
+    status = Column(String, default="Present")
+
+    person = relationship("Person", back_populates="attendances")

app/main.py

@@ -0,0 +1,17 @@
+from fastapi import FastAPI
+
+from app.database.database import engine, Base
+from app.api import register, attendance
+
+# ينشئ الجداول لو مش موجودة (person + attendance)
+Base.metadata.create_all(bind=engine)
+
+app = FastAPI(title="Face Attendance API")
+
+app.include_router(register.router)
+app.include_router(attendance.router)
+
+
+@app.get("/")
+def root():
+    return {"message": "Face Attendance API is running. Check /docs"}

app/services/edgeface_model.py

@@ -0,0 +1,90 @@
+"""
+تعريف موديل EdgeFace (بدون أي تعديل عن الكود الأصلي)
+"""
+import timm
+import torch
+import torch.nn as nn
+
+
+class LoRaLin(nn.Module):
+    def __init__(self, in_features, out_features, rank, bias=True):
+        super(LoRaLin, self).__init__()
+        self.in_features = in_features
+        self.out_features = out_features
+        self.rank = rank
+        self.linear1 = nn.Linear(in_features, rank, bias=False)
+        self.linear2 = nn.Linear(rank, out_features, bias=bias)
+
+    def forward(self, input):
+        x = self.linear1(input)
+        x = self.linear2(x)
+        return x
+
+
+def replace_linear_with_lowrank_recursive_2(model, rank_ratio=0.2):
+    for name, module in model.named_children():
+        if isinstance(module, nn.Linear) and "head" not in name:
+            in_features = module.in_features
+            out_features = module.out_features
+            rank = max(2, int(min(in_features, out_features) * rank_ratio))
+            bias = False
+            if module.bias is not None:
+                bias = True
+            lowrank_module = LoRaLin(in_features, out_features, rank, bias)
+            setattr(model, name, lowrank_module)
+        else:
+            replace_linear_with_lowrank_recursive_2(module, rank_ratio)
+
+
+def replace_linear_with_lowrank_2(model, rank_ratio=0.2):
+    replace_linear_with_lowrank_recursive_2(model, rank_ratio)
+    return model
+
+
+class TimmFRWrapperV2(nn.Module):
+    """Wraps timm model"""
+
+    def __init__(self, model_name="edgenext_x_small", featdim=512, batchnorm=False):
+        super().__init__()
+        self.featdim = featdim
+        self.model_name = model_name
+
+        self.model = timm.create_model(self.model_name)
+        self.model.reset_classifier(self.featdim)
+
+    def forward(self, x):
+        x = self.model(x)
+        return x
+
+
+def get_timmfrv2(model_name, **kwargs):
+    return TimmFRWrapperV2(model_name=model_name, **kwargs)
+
+
+def get_model(name, **kwargs):
+    if name == "edgeface_xs_gamma_06":
+        return replace_linear_with_lowrank_2(
+            get_timmfrv2("edgenext_x_small", batchnorm=False), rank_ratio=0.6
+        )
+    elif name == "edgeface_xs_q":
+        model = get_timmfrv2("edgenext_x_small", batchnorm=False)
+        model = torch.quantization.quantize_dynamic(
+            model, qconfig_spec={torch.nn.Linear}, dtype=torch.qint8
+        )
+        return model
+    elif name == "edgeface_xxs":
+        return get_timmfrv2("edgenext_xx_small", batchnorm=False)
+    elif name == "edgeface_base":
+        return get_timmfrv2("edgenext_base", batchnorm=False)
+    elif name == "edgeface_xxs_q":
+        model = get_timmfrv2("edgenext_xx_small", batchnorm=False)
+        model = torch.quantization.quantize_dynamic(
+            model, qconfig_spec={torch.nn.Linear}, dtype=torch.qint8
+        )
+        return model
+    elif name == "edgeface_s_gamma_05":
+        return replace_linear_with_lowrank_2(
+            get_timmfrv2("edgenext_small", batchnorm=False), rank_ratio=0.5
+        )
+    else:
+        raise ValueError(f"Unknown model name: {name}")

app/services/embedding.py

@@ -0,0 +1,22 @@
+import torch
+import numpy as np
+
+from app.config import EDGEFACE_WEIGHTS, EDGEFACE_MODEL_NAME
+from app.services.edgeface_model import get_model
+from app.utils.preprocess import preprocess_face
+
+
+# ---------------- تحميل الموديل مرة واحدة بس ----------------
+_net = get_model(EDGEFACE_MODEL_NAME, fp16=False)
+_net.load_state_dict(torch.load(EDGEFACE_WEIGHTS, map_location="cpu"))
+_net.eval()
+
+
+def get_embedding(face_bgr: np.ndarray) -> np.ndarray:
+    """بياخد صورة وجه (crop بصيغة BGR) ويرجع الـ embedding بتاعه (512,)"""
+    face_tensor = preprocess_face(face_bgr)
+
+
+    with torch.no_grad():
+        emb = _net(face_tensor).squeeze().numpy().astype(np.float32)
+    return emb

app/services/face_detector.py

@@ -0,0 +1,38 @@
+from ultralytics import YOLO
+
+from app.config import YOLO_FACE_WEIGHTS
+
+_TRACKER_YAML = "bytetrack.yaml"
+
+# هيتحمل مرة واحدة بس لما السيرفر يشتغل
+_yolo_model = YOLO(YOLO_FACE_WEIGHTS)
+
+
+def get_yolo_model():
+    return _yolo_model
+
+
+def detect_faces(frame, conf=0.25):
+    """كشف الوجوه من غير tracking (بتستخدم فى /register)"""
+    results = _yolo_model.predict(frame, conf=conf, verbose=False)[0]
+    boxes = []
+    if results.boxes is not None:
+        for box in results.boxes.xyxy.cpu():
+            boxes.append(tuple(map(int, box.tolist())))
+    return boxes
+
+
+def track_faces(frame, conf=0.25):
+    """كشف + تتبع الوجوه (بتستخدم فى /attendance)"""
+    results = _yolo_model.track(
+        frame, persist=True, tracker=_TRACKER_YAML, conf=conf, verbose=False
+    )[0]
+
+    tracked = []
+    if results.boxes is not None and results.boxes.id is not None:
+        boxes = results.boxes.xyxy.cpu()
+        track_ids = results.boxes.id.int().cpu().tolist()
+        for box, track_id in zip(boxes, track_ids):
+            x1, y1, x2, y2 = map(int, box.tolist())
+            tracked.append((track_id, (x1, y1, x2, y2)))
+    return tracked

app/services/face_quality.py

@@ -0,0 +1,138 @@
+import cv2
+import math
+import mediapipe as mp
+
+mp_face_mesh = mp.solutions.face_mesh
+
+# إعدادات يمكن تعديلها
+MIN_FACE_SIZE = 120
+MIN_BRIGHTNESS = 60
+MAX_BRIGHTNESS = 220
+MIN_SHARPNESS = 80
+MAX_ROLL_ANGLE = 10
+
+
+def check_face_quality(face_img):
+    """
+    Parameters
+    ----------
+    face_img : BGR image (cropped face)
+
+    Returns
+    -------
+    dict
+    {
+        "ok": bool,
+        "message": "...",
+        "brightness": float,
+        "sharpness": float,
+        "roll": float
+    }
+    """
+
+    h, w = face_img.shape[:2]
+
+    # -----------------------------
+    # حجم الوجه
+    # -----------------------------
+    if h < MIN_FACE_SIZE or w < MIN_FACE_SIZE:
+        return {
+            "ok": False,
+            "message": "اقترب من الكاميرا",
+            "brightness": 0,
+            "sharpness": 0,
+            "roll": 0,
+        }
+
+    # -----------------------------
+    # الإضاءة
+    # -----------------------------
+    gray = cv2.cvtColor(face_img, cv2.COLOR_BGR2GRAY)
+
+    brightness = gray.mean()
+
+    if brightness < MIN_BRIGHTNESS:
+        return {
+            "ok": False,
+            "message": "الإضاءة ضعيفة",
+            "brightness": brightness,
+            "sharpness": 0,
+            "roll": 0,
+        }
+
+    if brightness > MAX_BRIGHTNESS:
+        return {
+            "ok": False,
+            "message": "الإضاءة قوية جدًا",
+            "brightness": brightness,
+            "sharpness": 0,
+            "roll": 0,
+        }
+
+    # -----------------------------
+    # وضوح الصورة
+    # -----------------------------
+    sharpness = cv2.Laplacian(gray, cv2.CV_64F).var()
+
+    if sharpness < MIN_SHARPNESS:
+        return {
+            "ok": False,
+            "message": "الصورة غير واضحة",
+            "brightness": brightness,
+            "sharpness": sharpness,
+            "roll": 0,
+        }
+
+    # -----------------------------
+    # Face Mesh
+    # -----------------------------
+    rgb = cv2.cvtColor(face_img, cv2.COLOR_BGR2RGB)
+
+    with mp_face_mesh.FaceMesh(
+        static_image_mode=True,
+        max_num_faces=1,
+        refine_landmarks=True
+    ) as mesh:
+
+        result = mesh.process(rgb)
+
+        if not result.multi_face_landmarks:
+            return {
+                "ok": False,
+                "message": "لم يتم اكتشاف معالم الوجه",
+                "brightness": brightness,
+                "sharpness": sharpness,
+                "roll": 0,
+            }
+
+        lm = result.multi_face_landmarks[0].landmark
+
+        left_eye = lm[33]
+        right_eye = lm[263]
+
+        left = (left_eye.x * w, left_eye.y * h)
+        right = (right_eye.x * w, right_eye.y * h)
+
+        roll = math.degrees(
+            math.atan2(
+                right[1] - left[1],
+                right[0] - left[0]
+            )
+        )
+
+        if abs(roll) > MAX_ROLL_ANGLE:
+            return {
+                "ok": False,
+                "message": "اجعل رأسك مستقيمًا",
+                "brightness": brightness,
+                "sharpness": sharpness,
+                "roll": roll,
+            }
+
+    return {
+        "ok": True,
+        "message": "Face OK",
+        "brightness": brightness,
+        "sharpness": sharpness,
+        "roll": roll,
+    }

app/services/recognition.py

@@ -0,0 +1,34 @@
+import numpy as np
+from sqlalchemy.orm import Session
+
+from app.database.models import Person
+from app.utils.preprocess import cosine_similarity
+from app.config import RECOGNITION_THRESHOLD
+
+
+def load_all_persons(db: Session):
+    """بترجع كل الأشخاص المسجلين مع الـ embeddings بتاعتهم"""
+    persons = db.query(Person).all()
+    return [(p.id, p.name, np.array(p.embedding, dtype=np.float32)) for p in persons]
+
+
+def identify_person(embedding: np.ndarray, known_persons: list):
+    """
+    بتقارن embedding بكل الأشخاص المعروفين وترجع أقرب واحد
+    known_persons: list of (id, name, embedding)
+    """
+    best_id, best_name, best_score = None, "Unknown", -1.0
+
+    for person_id, name, known_emb in known_persons:
+        score = cosine_similarity(embedding, known_emb)
+        print(f"Comparing with {name}: score={score:.4f}")
+        if score > best_score:
+            best_score = score
+            best_name = name
+            best_id = person_id
+
+    print(f"Best match: {best_name} with score={best_score:.4f}")
+    if best_score < RECOGNITION_THRESHOLD:
+        return None, "Unknown", best_score
+
+    return best_id, best_name, best_score

app/utils/preprocess.py

@@ -0,0 +1,23 @@
+import cv2
+import numpy as np
+import torch
+
+
+def preprocess_face(face_bgr: np.ndarray) -> torch.Tensor:
+    """
+    بياخد صورة وجه (BGR من OpenCV) ويرجعها Tensor جاهز للموديل
+    """
+    face = cv2.resize(face_bgr, (112, 112))
+    face = cv2.cvtColor(face, cv2.COLOR_BGR2RGB)
+    face = np.transpose(face, (2, 0, 1))
+    face_tensor = torch.from_numpy(face).unsqueeze(0).float()
+    face_tensor.div_(255).sub_(0.5).div_(0.5)
+    return face_tensor
+
+
+def cosine_similarity(a: np.ndarray, b: np.ndarray) -> float:
+    """Cosine similarity بين embedding واحد وembedding تاني"""
+    a = a.flatten()
+    b = b.flatten()
+    denom = (np.linalg.norm(a) * np.linalg.norm(b)) + 1e-8
+    return float(np.dot(a, b) / denom)

app/weights/PUT_WEIGHTS_HERE.txt

@@ -0,0 +1,3 @@
+حط هنا ملفات الموديلات:
+- YoloV8_Face.pt
+- edgeface_xxs_q.pt

app/weights/YoloV8_Face.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d545bf1add5aa736a4febac4f4f9245a6d596cd0fe70d5d57989fe0cb9e626ca
+size 6389512

app/weights/edgeface_xxs_q.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46be7cc7e8d6e23501c0627fe5fbef101922ee9b85ccbef476987ab1a320b053
+size 1734661

dockerignore

@@ -0,0 +1,4 @@
+venv
+.git
+__pycache__
+*.pyc

requirements.txt

@@ -0,0 +1,12 @@
+fastapi
+uvicorn[standard]
+sqlalchemy
+psycopg2-binary
+python-dotenv
+numpy
+opencv-python-headless
+torch
+timm
+ultralytics
+pillow
+mediapipe==0.10.21