How hetsonii/smilestamp-attendance_system works
Main Components
a. Frontend UI
• React app (src/, public/)
• Pages/Components: Home, UploadPage, ReviewAttendance, Sidebar, Topbar
b. Backend API
• Flask app (src/Backend/app.py) exposing REST endpoints
c. Liveness‐Detection Pipeline (ML Service)
• Python modules under src/Backend/livenessDetection/ (dataCollection, splitData, train.py, main.py)
• Model files: yolov8n.pt, version2_best.pt, version3_best.pt
d. Data Store
• Attendance data in src/data/attendance.csv & .json
• Students list in src/data/students.js
• Archive folder for historical data
e. Static Assets & Model Storage
• public/classifier folder (trained_knn_model.clf, version3_best.pt)
f. External Dependencies
• YoloV8, PyTorch (for .pt files), scikit‐learn (knn), Flask, React
Relationships and Data Flows
• User interacts with React UI → triggers API calls (HTTP POST/GET) to Flask backend
• Frontend “UploadPage” component sends image/video frames to /detect_liveness endpoint
• Flask app routes request → invokes livenessDetection/main.py → loads model files → runs inference
• ML pipeline returns liveness verdict + face metadata back to Flask
• Flask persists attendance records to CSV/JSON and returns success payload
• Frontend “ReviewAttendance” component fetches attendance JSON via /attendance endpoint
• Data is displayed in UI; users may archive old data, triggering backend to move files under src/data/archive
Architectural Patterns
• Separation of concerns:
– Presentation layer (React)
– API layer (Flask)
– ML/inference layer (Python modules)
– Data persistence (File‐based CSV/JSON)
• Modular ML training vs. inference: training code under runs/…/train, separate from main.py serving model
• Build & deployment via Create React App + Python requirements
Detailed Elements
– Component names matching code (UploadPage.js, app.py, dataCollection.py)
– API endpoints (e.g., POST /detect_liveness, GET /attendance)
– Model file names (yolov8n.pt, version2_best.pt)
– Data flow annotations (e.g., “Image frames →❯ LivenessInference →❯ Verdict + FaceID”)
– Optional: note training flow (train.py → runs/detect/train/weights/last.pt) off to one side
Putting It All Together
[User Browser]
↓ HTTP
[React Frontend (UploadPage, ReviewAttendance, Sidebar)]
↓ HTTP
[Flask Backend API]
↙︎ ↘︎
ML Service Data Store
(main.py + model files) (attendance.csv/.json, archive/)
↑ ↓
External Libraries & Frameworks