import gradio as gr
from PIL import Image
import numpy as np
from tensorflow.keras.preprocessing import image as keras_image
from tensorflow.keras.applications.resnet50 import preprocess_input
from tensorflow.keras.models import load_model

# Load your trained model
model = load_model('/home/user/app/mein_modell.h5')

def predict_character(img):
    img = Image.fromarray(img.astype('uint8'), 'RGB')  # Ensure the image is in RGB
    img = img.resize((224, 224))  # Resize the image to the input size of the model
    img_array = keras_image.img_to_array(img)  # Convert the image to an array
    img_array = np.expand_dims(img_array, axis=0)  # Expand dimensions to match model input
    img_array = preprocess_input(img_array)  # Preprocess the input as expected by ResNet50
    
    prediction = model.predict(img_array)  # Predict using the model
    classes = ['Chopper', 'Nami', 'Ruffy', 'Sanji', 'Usopp', 'Zoro']  # Character names as per your dataset
    return {classes[i]: float(prediction[0][i]) for i in range(len(classes))}  # Return the prediction in a dictionary format

# Define Gradio interface
interface = gr.Interface(
    fn=predict_character, 
    inputs=gr.Image(),  # Gradio handles resizing automatically based on the model input
    outputs=gr.Label(num_top_classes=6),  # Show top 3 predictions
    title="One Piece Character Classifier",
    description="Upload an image of a One Piece character and the classifier will predict which character it is."
)

# Launch the interface
interface.launch()