ay8112/cleancity-waste-dataset

UP Swachhta Mitra - Training Dataset

This directory contains the training and validation datasets for the waste classification model.

📁 Directory Structure

/data/
  ├── train/              # Training data (80%)
  │   ├── plastic/
  │   ├── paper/
  │   ├── metal/
  │   ├── glass/
  │   ├── cardboard/
  │   └── organic/
  └── val/                # Validation data (20%)
      ├── plastic/
      ├── paper/
      ├── metal/
      ├── glass/
      ├── cardboard/
      └── organic/

🔽 Download Dataset

Option 1: TrashNet Dataset (Recommended)

1. Visit: https://github.com/garythung/trashnet
2. Download the dataset
3. Extract and organize images into the folders above

Option 2: TACO Dataset

1. Visit: http://tacodataset.org/
2. Download the dataset
3. Organize images by category

Option 3: Kaggle Datasets

Search for waste classification datasets on Kaggle:

• https://www.kaggle.com/datasets/mostafaabla/garbage-classification
• https://www.kaggle.com/datasets/asdasdasasdas/garbage-classification

📊 Dataset Requirements

• Minimum images per category: 100-200 images
• Recommended: 500-1000 images per category
• Image formats: JPG, JPEG, PNG
• Split ratio: 80% training, 20% validation

🎯 Organizing Your Images

Place images in their respective category folders:

Training Data (80% of images):

• data/train/plastic/ - plastic bottles, bags, containers
• data/train/paper/ - newspapers, magazines, cardboard boxes
• data/train/metal/ - cans, aluminum foil, metal containers
• data/train/glass/ - bottles, jars, broken glass
• data/train/cardboard/ - boxes, packaging
• data/train/organic/ - food waste, plant materials

Validation Data (20% of images):

• data/val/plastic/
• data/val/paper/
• data/val/metal/
• data/val/glass/
• data/val/cardboard/
• data/val/organic/

🔧 Training the Model

Once you've populated the folders with images:

1. Install Python dependencies:

   pip install -r scripts/requirements.txt
   

2. Verify dataset structure:

   python scripts/train_model.py --check-data
   

3. Train the model:

   python scripts/train_model.py
   

4. Model output:

   • Trained model will be saved as waste_classifier.h5
   • Training history and metrics will be displayed

📝 Tips for Better Results

1. Image Quality:

   • Use clear, well-lit images
   • Various angles and perspectives
   • Different backgrounds

2. Data Augmentation:

   • The training script automatically applies:
     • Rotation (up to 20°)
     • Width/height shifts (20%)
     • Zoom (20%)
     • Horizontal flips

3. Balanced Dataset:

   • Try to have similar number of images in each category
   • Remove duplicates and corrupted images

4. Testing:

   • After training, disable DEMO_MODE in your .env
   • Test with real images in the application

🚀 Next Steps

After training:

1. The model file waste_classifier.h5 will be created in the scripts/ folder
2. Remove or set DEMO_MODE=0 in your environment
3. Restart the backend server
4. Test the classification in the app!

📚 Resources

• TensorFlow Image Classification Tutorial
• Keras ImageDataGenerator
• Waste Classification Best Practices