File size: 3,973 Bytes
0923fff 3f92370 |
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 |
---
license: bigcode-openrail-m
language:
- fr
metrics:
- accuracy
pipeline_tag: image-classification
---
# Mushroom Classification Model - JarviSpore
This repository contains **JarviSpore**, a mushroom image classification model trained on a multi-class dataset with 23 different types of mushrooms. Developed from scratch with TensorFlow and Keras, this model aims to provide accurate mushroom identification using advanced deep learning techniques, including *Grad-CAM* for interpreting predictions. This project explores the performance of from-scratch models compared to transfer learning.
## Model Details
- **Architecture**: Custom CNN (Convolutional Neural Network)
- **Number of Classes**: 23 mushroom classes
- **Input Format**: RGB images resized to 224x224 pixels
- **Framework**: TensorFlow & Keras
- **Training**: Conducted on a machine with an i9 14900k processor, 192GB RAM, and an RTX 3090 GPU
## Key Features
1. **Multi-Class Classification**: The model can predict among 23 mushroom species.
2. **Regularization**: Includes L2 regularization and Dropout to prevent overfitting.
3. **Class Weighting**: Manages dataset imbalances by applying specific weights for each class.
4. **Grad-CAM Visualization**: Utilizes Grad-CAM to generate heatmaps, allowing visualization of the regions influencing the model's predictions.
## Model Training
The model was trained using a structured dataset directory with data split as follows:
- train: Balanced training dataset
- validation: Validation set to monitor performance
- test: Test set to evaluate final accuracy
Main training hyperparameters include:
- **Batch Size**: 32
- **Epochs**: 20 with Early Stopping
- **Learning Rate**: 0.0001
Training was tracked and logged via MLflow, including accuracy and loss curves, as well as the best model weights saved automatically.
## Model Usage
### Prerequisites
Ensure the following libraries are installed:
bash
pip install tensorflow pillow matplotlib numpy
### Loading the Model
To load and use the model for predictions:
python
import tensorflow as tf
from PIL import Image
import numpy as np
# Load the model
model = tf.keras.models.load_model("path_to_model.h5")
# Prepare an image for prediction
def prepare_image(image_path):
img = Image.open(image_path).convert("RGB")
img = img.resize((224, 224))
img_array = tf.keras.preprocessing.image.img_to_array(img)
img_array = np.expand_dims(img_array, axis=0)
return img_array
# Prediction
image_path = "path_to_image.jpg"
img_array = prepare_image(image_path)
predictions = model.predict(img_array)
predicted_class = np.argmax(predictions[0])
print(f"Predicted Class: {predicted_class}")
### Grad-CAM Visualization
The integrated *Grad-CAM* functionality allows interpretation of the model's predictions. To use it, select an image and apply the Grad-CAM function to display the heatmap overlaid on the original image, highlighting areas influencing the model.
Grad-CAM example usage:
python
# Example usage of the make_gradcam_heatmap function
heatmap = make_gradcam_heatmap(img_array, model, last_conv_layer_name="last_conv_layer_name")
# Superimpose the heatmap on the original image
superimposed_img = superimpose_heatmap(Image.open(image_path), heatmap)
superimposed_img.show()
## Evaluation
The model was evaluated on the test set with an average accuracy above random chance, showing promising results for a first from-scratch version.
## Contributing
Contributions to improve accuracy or add new features (e.g., other visualization techniques or advanced optimization) are welcome. Please submit a pull request with relevant modifications.
## License
This model is licensed under a controlled license: please refer to the LICENSE file for details. You may use this model for personal projects, but any modifications or redistribution must be approved.
|