import tensorflow as tf
import numpy as np
import gradio as gr

# With the path to your actual .h5 file
model_path = 'model__xception.h5'

# Load the pre-trained model
model = tf.keras.models.load_model(model_path)

## Define the predict function for Gradio
def predict_gradio(image):
    # Preprocess the input image
    img_array = tf.image.resize(image, (299, 299))
    img_array = tf.keras.preprocessing.image.img_to_array(img_array)
    img_array = tf.expand_dims(img_array, 0)  # Create a batch

    # Make a prediction using the trained model
    predictions = model.predict(img_array)
    print(predictions)
    score = tf.nn.sigmoid(predictions[0])

    return ("This image most likely belongs to {} with a {:.2f} percent confidence.".format(labels[np.argmax(score)], 100 * np.max(score)))

#Create a Gradio interface
iface = gr.Interface(
    fn=predict_gradio,
    inputs=gr.Image(),
    outputs="label",
    live=True,
    cache_examples=False
)

# Launch the Gradio interface
iface.launch()