ML Vision reimagines classroom attendance using AI and computer vision. Instead of roll calls or manual check-ins, the system continuously analyzes live video, recognizes students, and records attendance automatically, down to the timestamp and confidence level. The platform brings together a modern React interface, a robust Spring Boot backend, and a real-time Python vision service, delivering an intuitive and automated attendance experience for both instructors and students.

• ml-vision-frontend/ — React + Vite UI (Shadcn UI, TailwindCSS)
• ml-vision-backend/ — Spring Boot REST API + PostgreSQL
• python-vision/ — Python Flask recognizer + agent that syncs student photos and performs face matching

cd ml-vision-backend
# configure src/main/resources/application.yaml
mvn spring-boot:run

cd ml-vision-frontend
npm install
npm run dev
# requires VITE_API_BASE_URL → e.g. http://localhost:8080/api

cd python-vision
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp example.env .env  # set API_BASE_URL, SESSION_ID, FRAME_SOURCE, etc.

# Flask recognizer server (frontend sends image frames)
python -m python_vision.server   # http://localhost:5001/recognize

# Agent mode (continuous camera + direct attendance upload)
python -m python_vision.main --source 0 --session-id <uuid>

React Frontend
     |
     v
Spring Boot API  <----->  Postgres
     |
     v
Python Flask /recognize  (face detection + matching)

Core flow

1. Frontend captures a video frame every second.
2. Frame → Python /recognize → returns { externalId, confidence }.
3. Frontend sends batch to Spring /api/attendance/batch.
4. On session end, frontend triggers /api/attendance/mark-absent to finalize the report.

Python vision engine

• Syncs all student photos from Spring
• Generates face embeddings with face_recognition
• Matches faces and returns confidence scores

• POST /api/attendance/batch?classId&sessionId&sessionStartedAt
• POST /api/attendance/mark-absent?classId&sessionId&sessionStartedAt
• GET /api/attendance/class/{classId}
• GET /api/attendance/class/{classId}/today

• GET/POST/PUT/DELETE /api/classes
• GET/POST/DELETE /api/classes/{classId}/roster/{externalId}

• GET /api/students
• POST /api/students (multipart; includes photo upload)
• PUT /api/students/{externalId}
• DELETE /api/students/{externalId}

• Live Attendance: real-time camera feed + continuous face recognition
• Students Module: add/edit students, upload or capture live photos
• Classes Module: manage rosters, track present/absent counts
• Reports: session-based summaries with CSV export
• Dashboard: today’s attendance rate + latest session activity

• Automatically syncs roster photos from Spring
• Stores images as {externalId}.jpg
• /reload rebuilds embeddings after a student is added or updated
• Uses FaceRecognizer and DeduplicatingRecognizer to avoid duplicate detections

• DB URL, username, password
• spring.jackson.time-zone: America/New_York
• Optional: python.reload.url

VITE_API_BASE_URL=http://localhost:8080/api
VITE_VISION_BASE_URL=http://localhost:5001   # optional override

API_BASE_URL=http://localhost:8080/api
SESSION_ID=<uuid>
FRAME_SOURCE=0
ROSTER_DIR=python-vision/roster

• Restart the Spring server after schema or entity changes.
• Restart Python or call /reload after updating student photos.
• Use LAN IPs (e.g. http://192.168.x.x) for mobile testing.
• Keep student photo filenames consistent with their externalId.