Instructions to use ashuwhy/efficientnetv2s with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Keras
How to use ashuwhy/efficientnetv2s with Keras:
# !pip install -U keras tensorflow huggingface_hub # Keras needs TensorFlow installed to read "hf://" paths, so the tensorflow backend is selected here; # "jax" and "torch" also work for computation once TensorFlow is installed. import os os.environ["KERAS_BACKEND"] = "tensorflow" import keras model = keras.saving.load_model("hf://ashuwhy/efficientnetv2s") - Notebooks
- Google Colab
- Kaggle
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,156 @@
|
|
| 1 |
-
|
| 2 |
-
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Waste Classification Model
|
| 2 |
+
|
| 3 |
+
## Model Files
|
| 4 |
+
Due to their large size, the model files are hosted externally:
|
| 5 |
+
|
| 6 |
+
- [efficientnetv2s_waste_classifier.keras](https://huggingface.co/ashuwhy/waste-classification/resolve/main/efficientnetv2s_waste_classifier.keras) (86.88 MB) - Base model
|
| 7 |
+
- [efficientnetv2s_waste_classifier_fine_tuned.keras](https://huggingface.co/ashuwhy/waste-classification/resolve/main/efficientnetv2s_waste_classifier_fine_tuned.keras) (198.11 MB) - Fine-tuned model
|
| 8 |
+
- [efficientnetv2s_waste_classifier_final.keras](https://huggingface.co/ashuwhy/waste-classification/resolve/main/efficientnetv2s_waste_classifier_final.keras) (198.11 MB) - Final saved model
|
| 9 |
+
|
| 10 |
+
Please download these files and place them in this directory before using the model.
|
| 11 |
+
|
| 12 |
+
## Overview
|
| 13 |
+
This repository contains a deep learning model for waste classification, a critical component of the Smart Waste Management and Recycling Optimization Platform (SWMRO). The model is designed to classify waste items into 7 categories using EfficientNetV2S architecture, optimized for performance on Mac M2 chips.
|
| 14 |
+
|
| 15 |
+

|
| 16 |
+

|
| 17 |
+

|
| 18 |
+

|
| 19 |
+
|
| 20 |
+
## Purpose
|
| 21 |
+
The waste classification model serves as the AI backbone for automated waste sorting in the SWMRO platform. It enables:
|
| 22 |
+
|
| 23 |
+
- **Accurate Waste Categorization**: Identifies waste items across 7 different categories
|
| 24 |
+
- **Recycling Optimization**: Improves recycling efficiency through precise classification
|
| 25 |
+
- **Environmental Impact**: Reduces landfill waste by ensuring proper waste sorting
|
| 26 |
+
- **Integration with SWMRO**: Powers the platform's waste analysis capabilities
|
| 27 |
+
|
| 28 |
+
## Model Architecture
|
| 29 |
+
|
| 30 |
+
### Technical Specifications
|
| 31 |
+
- **Base Model**: EfficientNetV2S (pre-trained on ImageNet)
|
| 32 |
+
- **Input Size**: 224x224 pixels (RGB)
|
| 33 |
+
- **Classes**: 7 waste categories
|
| 34 |
+
- **Training Approach**: Transfer learning with fine-tuning
|
| 35 |
+
- **Optimization**: Adam optimizer with learning rate scheduling
|
| 36 |
+
- **Regularization**: Dropout (0.5) and BatchNormalization
|
| 37 |
+
|
| 38 |
+
### Waste Categories
|
| 39 |
+
The model classifies waste into the following categories:
|
| 40 |
+
1. **Cardboard** → RECYCLABLE
|
| 41 |
+
2. **Glass** → RECYCLABLE
|
| 42 |
+
3. **Metal** → RECYCLABLE
|
| 43 |
+
4. **Paper** → RECYCLABLE
|
| 44 |
+
5. **Plastic** → RECYCLABLE
|
| 45 |
+
6. **Compost** → ORGANIC
|
| 46 |
+
7. **Trash** → GENERAL
|
| 47 |
+
8. **Hazardous** → HAZARDOUS [not used in the current model but used in swmro, this is just for future upgrade]
|
| 48 |
+
|
| 49 |
+
## Training Process
|
| 50 |
+
|
| 51 |
+
The model was trained using a comprehensive approach:
|
| 52 |
+
|
| 53 |
+
1. **Data Preparation**:
|
| 54 |
+
- Dataset split into training (80%) and validation (20%) sets
|
| 55 |
+
- Extensive data augmentation (flips, rotations, zoom, contrast, brightness)
|
| 56 |
+
|
| 57 |
+
2. **Transfer Learning**:
|
| 58 |
+
- Initial training with frozen EfficientNetV2S base
|
| 59 |
+
- Fine-tuning of later blocks for domain adaptation
|
| 60 |
+
|
| 61 |
+
3. **Optimization Strategy**:
|
| 62 |
+
- Early stopping to prevent overfitting
|
| 63 |
+
- Learning rate reduction on plateau
|
| 64 |
+
- Model checkpointing to save best weights
|
| 65 |
+
|
| 66 |
+
## Performance
|
| 67 |
+
|
| 68 |
+
The model achieves high accuracy in waste classification tasks, making it suitable for real-world applications. Performance metrics include:
|
| 69 |
+
- High classification accuracy across all waste categories
|
| 70 |
+
- Robust performance on varied waste item appearances
|
| 71 |
+
- Efficient inference time suitable for real-time applications
|
| 72 |
+
|
| 73 |
+
## Integration with SWMRO
|
| 74 |
+
|
| 75 |
+
This model integrates with the SWMRO platform to:
|
| 76 |
+
- Power the waste analysis module
|
| 77 |
+
- Support the citizen portal for waste identification
|
| 78 |
+
- Provide data for the sustainability dashboard
|
| 79 |
+
- Enhance recycling efficiency metrics
|
| 80 |
+
|
| 81 |
+
## Usage
|
| 82 |
+
|
| 83 |
+
### Requirements
|
| 84 |
+
```
|
| 85 |
+
tensorflow>=2.8.0
|
| 86 |
+
numpy>=1.19.5
|
| 87 |
+
pillow>=8.0.0
|
| 88 |
+
matplotlib>=3.4.0
|
| 89 |
+
seaborn>=0.11.0
|
| 90 |
+
scikit-learn>=1.0.0
|
| 91 |
+
```
|
| 92 |
+
|
| 93 |
+
### Inference Example
|
| 94 |
+
```python
|
| 95 |
+
import tensorflow as tf
|
| 96 |
+
from PIL import Image
|
| 97 |
+
import numpy as np
|
| 98 |
+
|
| 99 |
+
# Load the model
|
| 100 |
+
model = tf.keras.models.load_model('efficientnetv2s_waste_classifier_final.keras')
|
| 101 |
+
|
| 102 |
+
# Class names
|
| 103 |
+
class_names = ['cardboard', 'compost', 'glass', 'metal', 'paper', 'plastic', 'trash']
|
| 104 |
+
|
| 105 |
+
# Mapping to backend categories
|
| 106 |
+
mapping = {
|
| 107 |
+
'cardboard': 'RECYCLABLE',
|
| 108 |
+
'glass': 'RECYCLABLE',
|
| 109 |
+
'metal': 'RECYCLABLE',
|
| 110 |
+
'paper': 'RECYCLABLE',
|
| 111 |
+
'plastic': 'RECYCLABLE',
|
| 112 |
+
'compost': 'ORGANIC',
|
| 113 |
+
'trash': 'GENERAL'
|
| 114 |
+
}
|
| 115 |
+
|
| 116 |
+
# Preprocess image
|
| 117 |
+
def preprocess_image(image_path):
|
| 118 |
+
img = Image.open(image_path).resize((224, 224))
|
| 119 |
+
img_array = tf.keras.preprocessing.image.img_to_array(img)
|
| 120 |
+
img_array = tf.expand_dims(img_array, 0)
|
| 121 |
+
return img_array / 255.0
|
| 122 |
+
|
| 123 |
+
# Classify waste
|
| 124 |
+
def classify_waste(image_path):
|
| 125 |
+
preprocessed_img = preprocess_image(image_path)
|
| 126 |
+
predictions = model.predict(preprocessed_img)
|
| 127 |
+
predicted_class = class_names[np.argmax(predictions[0])]
|
| 128 |
+
confidence = np.max(predictions[0])
|
| 129 |
+
backend_category = mapping[predicted_class]
|
| 130 |
+
|
| 131 |
+
return {
|
| 132 |
+
'detailed_class': predicted_class,
|
| 133 |
+
'backend_category': backend_category,
|
| 134 |
+
'confidence': float(confidence)
|
| 135 |
+
}
|
| 136 |
+
```
|
| 137 |
+
|
| 138 |
+
## Future Improvements
|
| 139 |
+
|
| 140 |
+
Planned enhancements for the model include:
|
| 141 |
+
- Expanding the dataset with more diverse waste items
|
| 142 |
+
- Implementing model quantization for edge deployment
|
| 143 |
+
- Exploring MobileNetV3 for mobile applications
|
| 144 |
+
- Adding multi-label classification for mixed waste
|
| 145 |
+
|
| 146 |
+
## License
|
| 147 |
+
|
| 148 |
+
This project is part of the SWMRO project and is available under the MIT License.
|
| 149 |
+
|
| 150 |
+
## Acknowledgments
|
| 151 |
+
|
| 152 |
+
- The model architecture is based on EfficientNetV2S by Google Research
|
| 153 |
+
- Training utilized TensorFlow and Keras frameworks
|
| 154 |
+
- Special thanks to contributors of waste classification datasets
|
| 155 |
+
|
| 156 |
+
© 2025 Ashutosh Sharma. All rights reserved.
|