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

# Load your custom classification model
model_path = "kia_mlp_classification_pokemon2.weights.h5"
model_path = "kia_mlp_classification_pokemon2.keras"
model = tf.keras.models.load_model(model_path)

labels = ['Pikachu', 'Psyduck', 'Pidgey']

# Define classification function
def predict_classification(image):
    # Preprocess image
    image = Image.fromarray(image.astype('uint8'), 'RGB')  # Convert numpy array to RGB PIL image
    image = image.resize((224, 224))  # Resize the image to 224x224
    image = np.array(image) / 255.0  # Scale pixel values to [0, 1]
    
    # Predict
    prediction = model.predict(np.array([image]))  # Make sure to add a batch dimension
    confidence = {labels[i]: float(np.round(prediction[0][i], 2)) for i in range(3)}
    return confidence

# Create Gradio interface
input_image = gr.Image()
output_text = gr.Textbox(label="Predicted Value")

interface = gr.Interface(
    fn=predict_classification, 
    inputs=input_image, 
    outputs=gr.Label(),
    examples=["./pokemon/pikachu/i1.png", "./pokemon/psyduck/p2.png", "./pokemon/pidgey/pi1.png", "./pokemon/pikachu/i2.png", "./pokemon/psyduck/p1.jpg", "./pokemon/pidgey/pi2.jpg","./pokemon/psyduck/p3.jpg"], 
    description="Upload or select an image to classify as Pikachu, Psyduck, or Pidgey."
)

interface.launch()