Model Manager (YOLO Weights)

Managers can view and switch between available YOLO model weights from the dashboard:

• Drop new .pt files into the models/ directory (no subfolders).
• The right panel in the Manager Dashboard lists all available models.
• The currently loaded model is marked as (Active).
• To activate a different model, click the Activate button next to it. You will be prompted for the manager password.
• On success, the selected model is loaded and used for all future detections. The choice is persisted in the database.

How to use:

1. Place your YOLO .pt file in the models/ folder.
2. Log in as manager and go to the dashboard.
3. Find your model in the list and click Activate.
4. Enter your password to confirm. The model will be swapped instantly.

Note:

• The system will remember your selection even after restart.
• If a model file is missing, you will see an error.

System Reset (Factory Defaults)

Managers can reset the system to factory defaults from the dashboard:

• Deletes all inspection and query logs from the database
• Removes all uploaded and processed images
• Requires manager password confirmation
• Accessible only to managers from the dashboard (red “Reset All Data” button)

How to use:

1. Log in as manager
2. Click the red “Reset All Data” button under the pie chart
3. Enter your password in the modal and confirm
4. All data and images will be deleted, and a success message will be shown

Note: This action cannot be undone.

AI-Based Visual Defect Detection System (VDDS)

Overview

A Flask + SQLite + Docker prototype for visual defect detection in manufacturing, with role-based access, inspection logging, feedback, and manager dashboard.

Features

• User authentication (officer/manager roles)
• Image inspection with YOLOv8/YOLOv5n real detection (first run may download ~50MB weights)
  • Images are now processed at their native resolution for accurate bounding boxes. If you encounter GPU RAM issues, set the environment variable YOLO_IMAGE_MAX=1024 (or another value) to downscale images consistently for detection. All bounding boxes and CSV annotations will still be reported in the original image's pixel coordinates.
• Feedback form (false alarm, missed defect, annotation, disposition)
• Inspection logging (timestamp, user, result, flags, annotation, disposition, image path)
• Manager dashboard with stats and charts (Chart.js)
• Model update stub (manager only)
• Dockerised for easy deployment

Database Migration Note (vdds >= 2025-06)

If you are upgrading from an earlier version:

• The database schema now includes orig_path and proc_path columns in the inspection_log table for robust image storage.
• You must run the migration script to add these columns to your existing database:

python migrate_add_image_paths.py

• If you encounter errors or want a clean start, you may delete data/vdds.db and restart the app to auto-create a fresh database (note: this will erase all previous inspection logs).

---

Setup

1. Local (pip)

python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt
# First run may download YOLO weights (~50MB)
python run.py

App runs at http://localhost:5000

2. Docker

docker build -t vdds .
docker run -p 5000:5000 vdds

Default Logins

• Officer: officer / officerpass
• Manager: manager / managerpass

Usage Flow

1. Officer logs in, uploads image, reviews result, submits feedback.
2. All actions are logged.
3. Manager logs in to view dashboard and trigger model update stub.

Project Structure

app/
  __init__.py  models.py  routes.py  detection.py
  templates/ base.html login.html inspect.html dashboard.html model.html
static/
  uploads/ ...
data/ (auto-created)
models/ (empty for now)
run.py
Dockerfile
requirements.txt
README.md

Testing & CI

Run tests

pytest

or for quiet output:

pytest -q

Notes

• Running YOLO locally / first-time weight download may take 50 MB.
• If model download fails (offline), app will fallback to stub detection ("Model unavailable").
• All flows & CSV export now show real detection annotation JSON.

Assumptions

• All images are stored in static/uploads/.
• Only two roles exist: QualityControlOfficer and QualityControlManager.
• No email/password reset; default users are created on first run.
• All requirements from Analysis & Design reports are mapped as above. If any ambiguity, see this README for rationale.

python src/app.py

Then open your browser at: http://127.0.0.1:5000

---

🔐 Login Credentials

• Quality Control Officer

  • Username: qofficer
  • Password: 1234

• Quality Control Manager

  • Username: qmanager
  • Password: 1234

---

🚀 Usage

1. As QC Officer

   • Select “Ürün Görüntüsü Seç”
   • Click “Analiz Et”
   • View the detection result
   • If needed, submit “Yanlış Alarm” or “Kusur Atlandı” feedback

2. As QC Manager

   • Log in with the manager account
   • Navigate to Yönetici Paneli
   • Review the summary of feedback and defect metrics

---

📷 Test Input Image

• A sample image for testing is provided as test_input.jpg in the repository root. Use this file to verify the detection pipeline.

---

📦 Packaging

All code, folder structure and this README are ready to be archived as:

vdds_prototip.rar

You can download, extract and run immediately.

---

İyi çalışmalar!