import gradio as gr
import tensorflow as tf
print(tf.__version__)
import numpy as np
from PIL import Image
 
model_path = "vehicles_transferlearning_nasnetlarge.keras"
model = tf.keras.models.load_model(model_path)
 
# Define the core prediction function
def predict_vehicles(image):
    # Preprocess image
    print(type(image))
    image = Image.fromarray(image.astype('uint8'))  # Convert numpy array to PIL image
    image = image.resize((150, 150)) #resize the image to 28x28 and converts it to gray scale
    image = np.array(image)
    image = np.expand_dims(image, axis=0) # same as image[None, ...]
   
    # Predict
    prediction = model.predict(image)
 
    # No need to apply sigmoid, as the output layer already uses softmax
    # Convert the probabilities to rounded values
    prediction = np.round(prediction, 2)
 
    # Separate the probabilities for each class
    p_car = prediction[0][0]  # Probability for class 'car'
    p_motorcycle = prediction[0][1]   # Probability for class 'motorcycle'
    p_truck = prediction[0][2]    # Probability for class 'truck'
 
    return {'car': p_car, 'motorcycle': p_motorcycle, 'truck': p_truck}
 
 
# Create the Gradio interface
input_image = gr.Image()
iface = gr.Interface(
    fn=predict_vehicles,
    inputs=input_image,
    outputs=gr.Label(),
    examples=["Sample_images/sample1.png", "Sample_images/sample2.jpg", "Sample_images/sample3.jpg", "Sample_images/sample4.jpg", "Sample_images/sample5.jpg", "Sample_images/sample6.jpg"],  
    description="TEST.")
 
iface.launch()