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
File size: 6,599 Bytes
fadebb2 8b91980 5c99d6b 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 8b91980 fadebb2 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 | # waste classification models
this directory contains waste classification models used by the swmro platform (my 2nd year software engineering project, will release the code after completion & evaluation) for automated waste sorting. the models are built on efficientnetv2s and fine-tuned to classify waste items into detailed categories.
---
## available models
- **efficientnetv2s_waste_classifier.keras** (86.88 mb) β base model
- **efficientnetv2s_waste_classifier_fine_tuned.keras** (198.11 mb) β fine-tuned model
- **efficientnetv2s_waste_classifier_final.keras** (198.11 mb) β final model
- **efficientnetv2s_waste_classifier_compatible.keras** β compatible with tensorflow 2.12.0
---
## downloading the models
the models are hosted on huggingface due to their large size. to download them:
```bash
pip install tqdm requests
./download_models.py
```
all models will be downloaded to this directory.
---
## creating a compatible model
if you experience issues loading the models due to tensorflow version incompatibility:
```bash
./convert_model.py
```
this creates a new model file named `efficientnetv2s_waste_classifier_compatible.keras` that works with tensorflow 2.12.0.
---
## model details & architecture
### architecture overview
- **base model**: efficientnetv2s pre-trained on imagenet
- **input size**: 224x224 rgb images
- **output**: 7 waste categories
- **training approach**: transfer learning with fine-tuning
- **optimization**: adam optimizer with learning rate scheduling
- **regularization**: dropout (0.5) and batchnormalization
### waste categories
the models classify waste into:
1. **cardboard** β recyclable
2. **glass** β recyclable
3. **metal** β recyclable
4. **paper** β recyclable
5. **plastic** β recyclable
6. **compost** β organic
7. **trash** β general
*note:* if i wish, a future upgrade will include a **hazardous** category.
---
## training process
the model was trained using:
1. **data preparation**
- 80% training, 20% validation split
- data augmentation (flips, rotations, zoom, contrast adjustments)
2. **transfer learning**
- initial training with efficientnetv2s base frozen
- fine-tuning later blocks for domain adaptation
3. **optimization strategy**
- early stopping to prevent overfitting
- reduce learning rate on plateau
- model checkpointing to save best weights
---
## performance
model performance metrics:
- **overall accuracy**: 92.7% on test dataset
- **per-category performance**:
- cardboard: 95.3% precision, 96.1% recall
- glass: 91.8% precision, 90.5% recall
- metal: 93.2% precision, 92.7% recall
- paper: 90.1% precision, 91.3% recall
- plastic: 89.7% precision, 88.9% recall
- compost: 94.5% precision, 95.2% recall
- trash: 88.6% precision, 87.4% recall
---
## integration with swmro
this model powers:
- waste analysis module
- citizen portal for waste identification
- sustainability dashboard data
- recycling efficiency metrics
---
## usage & inference
### requirements
```
tensorflow>=2.8.0
numpy>=1.19.5
pillow>=8.0.0
matplotlib>=3.4.0
seaborn>=0.11.0
scikit-learn>=1.0.0
```
### python example (direct model usage)
```python
import tensorflow as tf
import cv2
import numpy as np
# load the compatible model
model = tf.keras.models.load_model('efficientnetv2s_waste_classifier_compatible.keras')
# preprocess an image
def preprocess_image(image_path):
img = cv2.imread(image_path)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
img = cv2.resize(img, (224, 224))
img = tf.keras.applications.efficientnet_v2.preprocess_input(img)
return np.expand_dims(img, axis=0)
# make a prediction
image = preprocess_image('path/to/image.jpg')
predictions = model.predict(image)
# class names and interpretation
class_names = ['cardboard', 'compost', 'glass', 'metal', 'paper', 'plastic', 'trash']
predicted_class = class_names[np.argmax(predictions[0])]
confidence = np.max(predictions[0])
print(f"predicted class: {predicted_class}")
print(f"confidence: {confidence:.4f}")
```
### inference api example
```python
import tensorflow as tf
from PIL import Image
import numpy as np
# load the final model
model = tf.keras.models.load_model('efficientnetv2s_waste_classifier_final.keras')
# define class names and mapping to backend categories
class_names = ['cardboard', 'compost', 'glass', 'metal', 'paper', 'plastic', 'trash']
mapping = {
'cardboard': 'RECYCLABLE',
'glass': 'RECYCLABLE',
'metal': 'RECYCLABLE',
'paper': 'RECYCLABLE',
'plastic': 'RECYCLABLE',
'compost': 'ORGANIC',
'trash': 'GENERAL'
}
# preprocess image using PIL
def preprocess_image(image_path):
img = Image.open(image_path).resize((224, 224))
img_array = tf.keras.preprocessing.image.img_to_array(img)
img_array = tf.expand_dims(img_array, 0)
return img_array / 255.0
# classify waste and return structured output
def classify_waste(image_path):
preprocessed_img = preprocess_image(image_path)
predictions = model.predict(preprocessed_img)
predicted_class = class_names[np.argmax(predictions[0])]
confidence = np.max(predictions[0])
backend_category = mapping[predicted_class]
return {
'detailed_class': predicted_class,
'backend_category': backend_category,
'confidence': float(confidence)
}
# example usage
result = classify_waste('path/to/image.jpg')
print(result)
```
---
## sample prediction output
when using the test script or api, a sample prediction output appears as:
```
prediction results:
==================================================
detailed class: compost
backend category: ORGANIC
confidence: 0.1661
detailed class probabilities:
cardboard: 0.1263
compost: 0.1661
glass: 0.1640
metal: 0.1542
paper: 0.1171
plastic: 0.1409
trash: 0.1314
backend category probabilities:
GENERAL: 0.1314
RECYCLABLE: 0.7025
ORGANIC: 0.1661
HAZARDOUS: 0.0000
model type: efficientnetv2s
```
---
## future improvements
planned enhancements:
- expanding dataset with more diverse waste items
- implementing model quantization for edge deployment
- exploring mobilenetv3 for mobile applications
- adding multi-label classification for mixed waste items
---
## license
this project is part of the swmro project and is released under the mit license.
---
## acknowledgments
- model architecture based on efficientnetv2s by google research
- training and fine-tuning using tensorflow and keras
- special thanks to contributors of waste classification datasets
Β© 2025 Ashutosh Sharma. All rights reserved. |