cnn_utils.py

# importing libraries and packages
import cv2
import h5py
import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
from sklearn.metrics import confusion_matrix, classification_report


def load_signs_dataset():
    """
    Loads the hand signs dataset from HDF5 files.

    Returns:
        train_set_x_orig (numpy.ndarray): Training set features (images).
        train_set_y_orig (numpy.ndarray): Training set labels, reshaped to (1, number_of_samples).
        test_set_x_orig (numpy.ndarray): Test set features (images).
        test_set_y_orig (numpy.ndarray): Test set labels, reshaped to (1, number_of_samples).
        classes (numpy.ndarray): Array containing the list of class labels.
    """
    
    # Open the HDF5 file containing the training dataset in read mode
    train_dataset = h5py.File('datasets/train_signs.h5', "r")
    
    # Extract training set features (images) as a NumPy array
    train_set_x_orig = np.array(train_dataset["train_set_x"][:])
    
    # Extract training set labels as a NumPy array
    train_set_y_orig = np.array(train_dataset["train_set_y"][:])

    # Open the HDF5 file containing the test dataset in read mode
    test_dataset = h5py.File('datasets/test_signs.h5', "r")
    
    # Extract test set features (images) as a NumPy array
    test_set_x_orig = np.array(test_dataset["test_set_x"][:])
    
    # Extract test set labels as a NumPy array
    test_set_y_orig = np.array(test_dataset["test_set_y"][:])
    
    # Extract the list of class labels (e.g., class names) as a NumPy array
    classes = np.array(test_dataset["list_classes"][:])

    # Reshape the training labels to (1, number_of_samples) for consistency
    train_set_y_orig = train_set_y_orig.reshape((1, train_set_y_orig.shape[0]))
    
    # Reshape the test labels to (1, number_of_samples) for consistency
    test_set_y_orig = test_set_y_orig.reshape((1, test_set_y_orig.shape[0]))

    # Return the training and test features, labels, and class labels
    return train_set_x_orig, train_set_y_orig, test_set_x_orig, test_set_y_orig, classes

def dataset_preprocessing(img_data, img_label):
    """
    Preprocesses image data and labels for model training or testing.

    Args:
        img_data (numpy.ndarray): The input image data array.
        img_label (numpy.ndarray): The corresponding labels for the image data.

    Returns:
        norm_data (numpy.ndarray): Normalized image data with pixel values scaled between 0 and 1.
        one_hot_label (numpy.ndarray): Labels converted to one-hot encoded vectors.
    """

    # Normalize the image data by scaling pixel values to the range [0, 1]
    norm_data = img_data / 255.0

    # Convert labels to one-hot encoding
    # Assuming there are 6 classes (from 0 to 5)
    # Reshape labels to a 1D array for correct indexing
    one_hot_label = np.eye(6)[img_label.reshape(-1)]
    #  or simply use tensorflow.keras.utils.to_categorical(integer_labels, num_classes=num_classes)

    # Return the normalized image data and one-hot encoded labels
    return norm_data, one_hot_label

def visualize_sign(data, label, max_indx):
    """
    Visualizes a random selection of hand sign images along with their predicted labels.

    Args:
        data (numpy.ndarray): The dataset containing image data. 
                              Shape is expected to be (num_samples, height, width, channels).
        label (numpy.ndarray): The predicted labels corresponding to the dataset. 
                               Shape should be (1, num_samples).
        max_indx (int): The number of images to display in the visualization.

    Returns:
        None: Displays the images and their labels using Matplotlib.
    """

    # Create a subplot with 1 row and `max_indx` columns to display multiple images
    fig, axs = plt.subplots(1, max_indx)

    # Randomly select `max_indx` indices from the dataset
    arr = list(np.random.randint(0, data.shape[0], size=max_indx))

    # Loop through the randomly selected indices
    for i, value in enumerate(arr):
        # Display the image at the current index in the dataset
        axs[i].imshow(data[value])

        # Set the title of the subplot to display the corresponding predicted label
        axs[i].set_title(f'It is {label[0, value]}')

        # Turn off the axis for better visualization
        axs[i].axis('off')

    # Show the complete visualization
    plt.show()

def visualize_metrics(history):
    """
    Visualizes the training metrics (accuracy and loss) over epochs.

    Args:
        history (tensorflow.keras.callbacks.History): The history object returned by the 
                                                      model's `fit` method. Contains training metrics.

    Returns:
        None: Displays the accuracy and loss plots.
    """
    # Convert the history object to a DataFrame for easier visualization
    df = pd.DataFrame(history.history)

    # Create subplots for accuracy and loss
    fig, axs = plt.subplots(1, 2, figsize=(12, 4))  # 1 row, 2 columns, set figure size for better readability

    # Plot accuracy over epochs
    axs[0].plot(df['accuracy'], label='Training Accuracy')  # Training accuracy
    if 'val_accuracy' in df.columns:  # Check for validation accuracy
        axs[0].plot(df['val_accuracy'], label='Validation Accuracy', linestyle='--')
    axs[0].set_title('Accuracy')  # Set title for the plot
    axs[0].set_xlabel('Epochs')  # Label x-axis
    axs[0].set_ylabel('Accuracy')  # Label y-axis
    axs[0].legend()  # Add a legend

    # Plot loss over epochs
    axs[1].plot(df['loss'], label='Training Loss')  # Training loss
    if 'val_loss' in df.columns:  # Check for validation loss
        axs[1].plot(df['val_loss'], label='Validation Loss', linestyle='--')
    axs[1].set_title('Loss')  # Set title for the plot
    axs[1].set_xlabel('Epochs')  # Label x-axis
    axs[1].set_ylabel('Loss')  # Label y-axis
    axs[1].legend()  # Add a legend

    # Display the plots
    plt.tight_layout()  # Adjust subplot spacing for a clean layout
    plt.show()


def evaluate_model(model, test_data, test_labels, class_names):
    """
    Evaluates a trained model on test data and visualizes metrics.

    Args:
        model: The trained model to evaluate.
        test_data (numpy.ndarray): Normalized test data used for predictions.
        test_labels (numpy.ndarray): One-hot encoded true labels of the test data.
        class_names (list of str): List of class names corresponding to the classes.

    Returns:
        None: Displays a confusion matrix and classification report.
    """
    class_names = list(class_names.astype(str))
    # Make predictions on the test data
    y_pred = model.predict(test_data)  # Predicted probabilities
    y_pred_classes = np.argmax(y_pred, axis=1)  # Convert probabilities to class indices

    # Convert one-hot encoded labels to class indices
    test_labels_classes = np.argmax(test_labels, axis=1)

    # Generate a confusion matrix
    conf_matrix = confusion_matrix(test_labels_classes, y_pred_classes)

    # Print a classification report
    print("\nClassification Report:")
    print(classification_report(test_labels_classes, y_pred_classes, target_names=class_names))

    # Visualize the confusion matrix as a heatmap
    plt.figure(figsize=(8, 6))
    sns.heatmap(conf_matrix, annot=True, fmt='d', cmap='Blues',
                xticklabels=class_names, yticklabels=class_names)
    plt.title("Confusion Matrix")
    plt.xlabel("Predicted Labels")
    plt.ylabel("True Labels")
    plt.show()



def predictor(model, image_array, input_size=(64, 64), class_names=None):
    """
    Makes a prediction on a single image using the given model and visualizes the result.

    Args:
        model: The trained model to use for prediction.
        image_array (numpy.ndarray): The input image array (e.g., loaded using OpenCV).
        input_size (tuple): The expected input size of the model (default is (64, 64)).
        class_names (list of str, optional): List of class names corresponding to the model's output.

    Returns:
        None: Displays the input image with the predicted class as the title.
    """
    # Validate and convert class_names to a list of strings if provided
    if class_names is not None:
        class_names = [str(cls) for cls in class_names]
    
    # Resize the image to match the model's input size
    img = cv2.resize(image_array, input_size)

    # Normalize the pixel values to [0, 1]
    img = img / 255.0

    # Add a batch dimension (1, height, width, channels)
    img = np.expand_dims(img, axis=0)

    # Make a prediction
    pred = model.predict(img)  # Returns probabilities for each class

    # Squeeze the prediction array to remove unnecessary dimensions
    pred = np.squeeze(pred)

    # Print the predicted probabilities
    print("Predicted Probabilities:", pred)

    # Get the index of the class with the highest probability
    predicted_class_idx = np.argmax(pred, axis=0)

    # Print the predicted class
    print("Predicted Class Index:", predicted_class_idx)

    # Get the class name if provided, otherwise use the index
    predicted_class_name = class_names[predicted_class_idx] if class_names else str(predicted_class_idx)

    # Convert the image from BGR to RGB for Matplotlib visualization
    rgb_image = cv2.cvtColor(image_array, cv2.COLOR_BGR2RGB)

    # Display the image with the predicted class as the title
    plt.imshow(rgb_image)  # Use the RGB image for Matplotlib
    plt.title(f'Predicted: {predicted_class_name}')
    plt.axis('off')  # Turn off axis for better visualization
    plt.show()