app.py

import gradio as gr
import tensorflow as tf
import numpy as np
from PIL import Image

# Load model
model = tf.keras.models.load_model("model/car_model.h5")
class_names = ['Audi A4', 'Toyota Corolla', 'BMW X5', 'Ford Focus', 'Honda Civic', 
               'Hyundai Elantra', 'Mercedes C Class', 'Kia Sportage', 'Chevrolet Cruze', 'Mazda 3']  # Ganti sesuai dataset

def classify_car(image):
    image = image.resize((224, 224))
    img_array = tf.keras.utils.img_to_array(image) / 255.0
    img_array = np.expand_dims(img_array, axis=0)
    predictions = model.predict(img_array)[0]
    top_3 = np.argsort(predictions)[-3:][::-1]
    return {class_names[i]: float(predictions[i]) for i in top_3}

interface = gr.Interface(fn=classify_car, 
                         inputs=gr.Image(type="pil"),
                         outputs=gr.Label(num_top_classes=3),
                         title="Car Brand & Model Classifier",
                         description="Upload a car image to predict the brand and model.")

if __name__ == "__main__":
    interface.launch()