app.py

import gradio as gr
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras.applications.vgg16 import VGG16, preprocess_input as preprocess_vgg
from tensorflow.keras.applications.efficientnet import EfficientNetB0, preprocess_input as preprocess_efficientnet
from tensorflow.keras.preprocessing import image
import numpy as np


# Define input image size (must match training settings)
image_size = (224, 224)

# Function to build the VGG16 model
def build_vgg16_model():
    base_model = VGG16(weights=None, include_top=False, input_shape=image_size + (3,))
    inputs = keras.layers.Input(shape=image_size + (3,))
    x1 = preprocess_vgg(inputs)
    x1 = base_model(x1, training=False)
    x1 = keras.layers.Flatten()(x1)  # Flatten instead of GlobalAveragePooling2D
    x1 = keras.layers.Dropout(rate=0.5)(x1)
    x1 = keras.layers.Dense(units=256, activation="relu")(x1)
    x1 = keras.layers.Dropout(rate=0.5)(x1)
    outputs = keras.layers.Dense(units=1, activation="sigmoid")(x1)

    model = keras.models.Model(inputs=[inputs], outputs=[outputs])
    
    # Compile the model
    model.compile(
        loss="binary_crossentropy",
        optimizer=keras.optimizers.Adam(learning_rate=0.001),
        metrics=["accuracy"],
    )
    
    return model

# Function to build the EfficientNetB0 model
def build_efficientnet_model():
    base_model = EfficientNetB0(input_shape=image_size + (3,), include_top=False, weights="imagenet")
    inputs = keras.layers.Input(shape=image_size + (3,))
    x = preprocess_efficientnet(inputs)  # EfficientNet includes its own preprocessing
    x = base_model(x, training=False)
    x = keras.layers.GlobalAveragePooling2D()(x)
    x = keras.layers.Dropout(rate=0.5)(x)
    x = keras.layers.Dense(units=256, activation="relu")(x)
    x = keras.layers.Dropout(rate=0.5)(x)
    outputs = keras.layers.Dense(units=1, activation="sigmoid")(x)

    model = keras.models.Model(inputs=[inputs], outputs=[outputs])

    # Compile the model
    model.compile(
        loss="binary_crossentropy",
        optimizer=keras.optimizers.Adam(learning_rate=0.001),
        metrics=["accuracy"],
    )

    return model

# Dictionary to store models
models = {
    "VGG16": build_vgg16_model(),
    "EfficientNetB0": build_efficientnet_model(),
}

# Load pre-trained weights
models["VGG16"].load_weights("VGG16_best_finetune_checkpoint.weights.h5")
models["EfficientNetB0"].load_weights("EfficientNetB0_best_finetune_checkpoint.weights.h5")

print("Models and weights loaded successfully!")

# Set the default model
current_model = models["VGG16"]

# Function to update the current model based on selection
def load_selected_model(model_name):
    global current_model
    current_model = models[model_name]
    return f"Loaded {model_name} model."

# Preprocessing function
def preprocess_img(img, model_name):
    img = img.resize(image_size)  # Resize to match training
    img_array = image.img_to_array(img)

    # Apply the correct preprocessing function based on model selection
    if model_name == "VGG16":
        img_array = preprocess_vgg(img_array)
    else:
        img_array = preprocess_efficientnet(img_array)

    img_array = np.expand_dims(img_array, axis=0)  # Add batch dimension
    return img_array

# Prediction function
def predict(img, model_name):
    img_array = preprocess_img(img, model_name)
    predictions = current_model.predict(img_array)
    
    confidence = float(predictions[0][0])  # Convert to float
    
    # Class labels
    class_labels = {0: "Defective", 1: "Good"}
    predicted_class = 1 if confidence > 0.5 else 0

    result_text = (
        f"Predicted Class: {class_labels[predicted_class]}\n"
        f"Confidence (Good): {confidence:.8f}\n"
        f"Confidence (Defective): {1 - confidence:.8f}"
    )

    return result_text

# Function to clear input and output
def clear():
    return None, ""

# Gradio Interface with Model Selection
with gr.Blocks() as interface:
    gr.Markdown(
        "## Fine-tuned Defect Tyre Classification\n\n"
        "This Gradio-based application allows users to classify tyre images as **'Good'** or **'Defective'** using a "
        "fine-tuned deep learning model. Users can select between two models (**VGG16** and **EfficientNetB0**) "
        "via a dropdown menu. The selected model is dynamically loaded and applied to the uploaded image, "
        "with predictions and confidence scores displayed as output.\n\n"
        "Upload a tyre image and select a model (**VGG16** or **EfficientNetB0**) to classify defects."
    )

    # Dropdown to select model
    model_dropdown = gr.Dropdown(
        choices=list(models.keys()), value="VGG16", label="Select Model"
    )
    model_status = gr.Textbox(label="Model Status", interactive=False)

    with gr.Row():
        input_image = gr.Image(type="pil", label="Upload Image")
        output_text = gr.Textbox(label="Classification Output", interactive=False)

    with gr.Row():
        classify_button = gr.Button("Classify")
        clear_button = gr.Button("Clear")

    # Link dropdown to model selection function
    model_dropdown.change(fn=load_selected_model, inputs=model_dropdown, outputs=model_status)
    
    # Link buttons to functions
    classify_button.click(fn=predict, inputs=[input_image, model_dropdown], outputs=output_text)
    clear_button.click(fn=clear, inputs=[], outputs=[input_image, output_text])

# Run the app
if __name__ == "__main__":
    interface.launch()