Skip to content
BranchesTags

Repository files navigation

PlastiTrace Desktop

Aplikasi desktop Python untuk deteksi jenis plastik secara realtime dari webcam menggunakan PyTorch dan OpenCV dengan bbox tracking stabil.

Installation

  1. Create and activate a virtual environment:
# Create virtual environment
python3 -m venv venv
#ensure u use python 3.14 to avoid error 
#and make sure u have the c++ or c compiler installed for numpy 

# Activate virtual environment
# On macOS/Linux:
source venv/bin/activate
# On Windows:
# venv\Scripts\activate
  1. Install dependencies:
pip install -r requirements.txt

Usage

Desktop Application (CLI - Original)

Run the desktop application with camera loop (command line interface):

python app.py

Note: Make sure the virtual environment is activated before running the application.

Tekan ESC untuk keluar dari aplikasi.

Desktop Application (GUI - NEW)

Run the desktop application with PyQt5 GUI (realtime detection with modern interface):

python app_gui.py

Features:

  • 🎨 Modern GUI dengan PyQt5
  • 📹 Realtime video preview
  • 🤖 Live plastic classification
  • ♻️ Recycling recommendations in Indonesian
  • 📊 FPS counter
  • 🎯 Bounding box visualization

Tekan ESC atau tombol "Keluar" untuk menutup aplikasi.

Web Application

Run the web-based UI with API backend:

  1. Start the Flask API backend:
python api.py

The API will run on http://localhost:5001 (port 5001 karena 5000 conflict dengan AirPlay di macOS)

  1. Open the web interface:

Open web/index.html in your browser, or serve it with:

cd web
python -m http.server 8000

Then visit http://localhost:8000/index.html

Web Features:

  • 📸 Upload images or use webcam for classification
  • 🤖 AI classification with confidence scores
  • ♻️ Recycling recommendations
  • 🎨 Modern, responsive UI (Bahasa Indonesia)
  • 📱 Mobile-friendly

Note: Web app menggunakan "capture photo" mode, bukan realtime detection. Untuk realtime detection, gunakan Desktop GUI (app_gui.py).

Features

Desktop App (CLI) Features

  • Realtime bbox detection: Deteksi objek plastik menggunakan OpenCV contours (tanpa YOLO)
  • Stable tracking: CSRT tracker untuk bbox yang stabil dan smooth
  • ROI classification: Crop ROI dari bbox → klasifikasi dengan ResNet18
  • Inference throttling: Inference hanya setiap N frames (default: setiap 3 frames)
  • Smooth overlay: Bbox, label, confidence, dan rekomendasi ditampilkan realtime
  • FPS counter: Monitor performa aplikasi
  • CPU-friendly: Optimized untuk CPU, CUDA opsional
  • Stabil: Didesain untuk running >10 menit tanpa masalah

Desktop App (GUI) Features

  • Modern PyQt5 Interface: GUI yang clean dan professional
  • Realtime Video Preview: Live camera feed dengan bbox overlay
  • Live Classification: Deteksi dan klasifikasi plastik secara realtime
  • Information Panel: Tampilan hasil, confidence score, dan rekomendasi
  • FPS Monitor: Real-time FPS counter
  • Keyboard Shortcuts: ESC untuk keluar
  • Smooth Animations: Transisi yang halus dan responsive

Web App Features

  • Image Upload: Upload gambar plastik untuk klasifikasi
  • Webcam Capture: Ambil foto langsung dari webcam
  • AI Classification: Identifikasi jenis plastik dengan confidence score
  • Recycling Guide: Rekomendasi daur ulang untuk setiap jenis plastik
  • Responsive Design: Works on desktop and mobile devices
  • Real-time Processing: Fast classification with visual feedback

Architecture

Desktop App (CLI & GUI)

Webcam → Bbox Detection (contours) → CSRT Tracker → ROI Crop →
ResNet18 Classifier (throttled) → Overlay/Display (bbox + label + recommendation)

Web App

Browser (Upload/Camera) → Flask API → ResNet18 Classifier → JSON Response →
Web UI Display (Label + Confidence + Recommendations)

Key Components

Desktop App

  • app.py: Original CLI-based realtime detection (OpenCV display)
  • vision/bbox_detector.py: Deteksi bbox menggunakan Canny edges + contours
  • vision/bbox_tracker.py: CSRT/KCF tracker untuk stabilisasi bbox
  • vision/smoothing.py: EMA smoothing untuk bbox dan confidence
  • ml/classifier.py: ResNet18 classifier dengan FP32 enforcement
  • ui/camera_loop.py: Main camera loop logic (used by app.py)

Web App

  • api.py: Flask REST API endpoint untuk image classification
  • web/index.html: React-based web UI (single file, no build required)

API Endpoints

POST /api/classify

Classify a plastic item from an uploaded image.

Request:

  • Method: POST
  • Content-Type: multipart/form-data
  • Body: image (file)

Response:

{
  "label": "PET",
  "confidence": 0.95
}

GET /api/health

Check if the API is running.

Response:

{
  "status": "ok"
}

Model

The application uses a ResNet18 model trained for 4-class plastic classification. The model file should be located at models/plastitrace.pth.

Classes: HDPE, PET, PP, PS

Project Structure

plastitrace_desktop/
├── app.py                     # Entry point (Desktop CLI)
├── app_gui.py                 # Entry point (Desktop GUI - PyQt5)
├── api.py                     # Flask API backend (Web App)
├── requirements.txt
├── models/
│   └── plastitrace.pth
├── ml/
│   ├── config.py             # Constants & recommendations
│   ├── classifier.py         # PyTorch ResNet18 classifier
│   └── preprocess.py         # Image preprocessing
├── vision/
│   ├── bbox_detector.py      # Contour-based bbox detection
│   ├── bbox_tracker.py       # CSRT/KCF tracker wrapper
│   └── smoothing.py          # EMA smoothing
├── ui/
│   └── camera_loop.py        # Camera loop logic (for app.py)
├── utils/
│   └── softmax.py            # Softmax utility
└── web/
    └── index.html            # Web UI (React + Tailwind)

Configuration

Desktop App Settings (CLI & GUI)

Default settings in app.py and app_gui.py:

  • min_area=2000: Minimum bbox area untuk detection
  • inference_interval=3: Run inference setiap 3 frames
  • redetect_interval=30: Re-detect bbox setiap 30 frames (untuk koreksi drift)

Web App Settings

Default settings in api.py:

  • Host: 0.0.0.0
  • Port: 5001 (changed from 5000 to avoid AirPlay conflict on macOS)
  • Debug: True (set to False for production)
  • CORS: Enabled for all origins (restrict in production)

Web UI (index.html):

  • API URL: http://localhost:5001/api/classify

Recommendations

Aplikasi menampilkan rekomendasi daur ulang dalam Bahasa Indonesia:

  • HDPE: Umumnya bisa didaur ulang. Bilas dan masukkan ke sampah daur ulang plastik keras.
  • PET: Botol minum plastik. Bilas, lepas label bila memungkinkan, buang ke sampah daur ulang.
  • PP: Wadah makanan/kantong tertentu. Bila bersih, daur ulang; jika tidak ada fasilitas, buang sebagai residu.
  • PS: Styrofoam/foam. Sulit didaur ulang; hindari pembakaran, buang ke sampah residu.

Troubleshooting

Desktop App (CLI & GUI)

Error: "No OpenCV tracker available"

  • Install opencv-contrib-python: pip install opencv-contrib-python

Error: "No module named 'PyQt5'" (GUI only)

  • Install PyQt5: pip install PyQt5

Low FPS or slow inference

  • Reduce inference_interval in app.py or app_gui.py
  • Use smaller input images
  • Enable CUDA if available

Camera not detected

  • Check if camera is already in use by another application
  • Try changing camera_index from 0 to 1 in the code
  • Check camera permissions in System Preferences (macOS)

Web App

Error: "Port 5001 is in use"

  • Change port in api.py: app.run(host='0.0.0.0', port=5002, debug=True)
  • Update API URL in web/index.html to match new port

Error: "Cannot access camera"

  • Grant browser permission to access camera
  • Use HTTPS (required for webcam on some browsers)
  • Check if camera is already in use

Error: "Failed to classify image"

  • Make sure Flask API is running on http://localhost:5001
  • Check API logs for errors
  • Verify model file exists at models/plastitrace.pth

CORS errors

  • Ensure flask-cors is installed: pip install flask-cors
  • Check browser console for specific CORS errors

macOS AirPlay Receiver conflict (Port 5000)

  • Disable AirPlay Receiver in System Settings → General → AirDrop & Handoff
  • OR use port 5001 (already configured in api.py)

Application Comparison

Feature Desktop CLI Desktop GUI Web App
Realtime Detection ✅ Yes ✅ Yes ❌ No (capture only)
Modern UI ❌ No ✅ Yes ✅ Yes
Installation Easy Easy Medium
Performance Excellent Excellent Good
Mobile Support ❌ No ❌ No ✅ Yes
Best For Development/Testing End Users (Desktop) Web Access/Mobile

Recommendation:

  • Development/Debugging: Use app.py (CLI)
  • Desktop Users: Use app_gui.py (PyQt5 GUI) - RECOMMENDED
  • Web/Mobile Access: Use Web App (api.py + web/index.html)

Production Deployment

Desktop App

  • Package with PyInstaller for distribution:
  pip install pyinstaller
  pyinstaller --onefile --windowed app_gui.py

Web App

For production deployment:

  1. Set Flask to production mode:
    • Change debug=True to debug=False in api.py
    • Use a production WSGI server like Gunicorn:
     pip install gunicorn
     gunicorn -w 4 -b 0.0.0.0:5001 api:app

  1. Configure CORS properly:

    • Restrict allowed origins in api.py
    • Don't use CORS(app) with no parameters in production

  2. Serve static files:

    • Use Nginx or Apache to serve web/index.html
    • Configure reverse proxy to Flask API

  3. Security considerations:

    • Add rate limiting
    • Implement file size limits for uploads
    • Validate file types
    • Use HTTPS

License

[Your License Here]

Contributors

[Your Team/Contributors Here]

Changelog

v2.0 (Latest)

  • ✨ Added PyQt5 GUI desktop application (app_gui.py)
  • 🌐 Added web interface with React UI
  • 🔧 Fixed port conflict with macOS AirPlay (port 5001)
  • 🇮🇩 Full Bahasa Indonesia support in all interfaces
  • 📝 Updated documentation

v1.0

  • 🎯 Initial release with CLI desktop app
  • 🤖 ResNet18 classification
  • 📹 CSRT bbox tracking

About

No description, website, or topics provided.

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

1 fork
Report repository

Releases

Packages

Contributors

Languages