|
--- |
|
datasets: |
|
- garythung/trashnet |
|
metrics: |
|
- accuracy |
|
- precision |
|
- recall |
|
pipeline_tag: image-classification |
|
--- |
|
### **Model Card: Trash Classification Using MobileNetV2** |
|
|
|
--- |
|
|
|
## **Model Details** |
|
|
|
- **Model Name**: Trash Classification CNN with MobileNetV2 |
|
- **Model Type**: Convolutional Neural Network (CNN) |
|
- **Architecture**: MobileNetV2 |
|
- **Dataset**: [TrashNet Dataset](https://huggingface.co/datasets/garythung/trashnet) |
|
- **Languages**: None (Image-based model) |
|
- **License**: MIT |
|
|
|
--- |
|
|
|
## **Model Description** |
|
|
|
This model classifies images of trash into six categories: |
|
- **trash** |
|
- **plastic** |
|
- **cardboard** |
|
- **metal** |
|
- **paper** |
|
- **glass** |
|
|
|
The model is designed to assist in waste segregation and recycling initiatives by automating the identification of waste types. It uses MobileNetV2, a lightweight CNN architecture pre-trained on ImageNet, fine-tuned on the TrashNet dataset for this specific task. |
|
|
|
--- |
|
|
|
## **Intended Use** |
|
|
|
### **Primary Use Cases** |
|
- Waste management systems to automate sorting. |
|
- Educational tools for teaching about recycling and waste segregation. |
|
- Integration into mobile or web applications for real-time waste classification. |
|
|
|
### **Limitations** |
|
- Model performance may degrade with images of poor quality or those significantly different from the training dataset. |
|
- Currently supports only six predefined trash categories. |
|
|
|
--- |
|
|
|
## **Performance Metrics** |
|
|
|
- **Training Accuracy**: 95% |
|
- **Testing Accuracy**: 90% |
|
- **Metrics Evaluated**: Accuracy, Precision, Recall, F1-score |
|
- **Confusion Matrix**: [Available in evaluation results] |
|
|
|
--- |
|
|
|
## **How to Use the Model** |
|
|
|
### **Input Format** |
|
- Images resized to 224x224 pixels and normalized to a range of 0-1. |
|
|
|
### **Output** |
|
- A probability distribution over six classes with the predicted label. |
|
|
|
### **Code Example** |
|
|
|
```python |
|
from transformers import pipeline |
|
from PIL import Image |
|
|
|
# Load pre-trained model |
|
classifier = pipeline("image-classification", model="your-model-id") |
|
|
|
# Load an image |
|
image = Image.open("sample_image.jpg") |
|
|
|
# Perform classification |
|
results = classifier(image) |
|
print(results) |
|
``` |
|
|
|
--- |
|
|
|
## **Training Details** |
|
|
|
- **Framework**: TensorFlow/Keras |
|
- **Optimizer**: Adam |
|
- **Learning Rate**: 0.001 |
|
- **Loss Function**: Categorical Crossentropy |
|
- **Batch Size**: 32 |
|
- **Epochs**: 20 |
|
|
|
### **Data Preprocessing** |
|
- Images were resized to 224x224 pixels and normalized. |
|
- Oversampling and data augmentation techniques (rotation, zoom, and rescaling) were applied to handle class imbalance and enhance generalization. |
|
|
|
--- |
|
|
|
This model card is designed to comply with Hugging Face standards and can be adapted further as needed. Let me know if you need any specific sections expanded! |
