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

CLASS_NAMES = ["Oral Homogenous Leukoplakia", "Oral Non-Homogenous Leukoplakia", "Other Oral White Lesions"]
model = tf.keras.models.load_model('./model.keras')

def predict(image):
    img = image.convert("RGB").resize((224, 224))
    arr = (np.array(img, dtype=np.float32) / 127.5) - 1.0
    preds = model.predict(np.expand_dims(arr, 0))[0]
    idx = int(np.argmax(preds))
    return {"className": CLASS_NAMES[idx], "confidence": round(float(preds[idx]) * 100, 1)}

demo = gr.Interface(fn=predict, inputs=gr.Image(type="pil"), outputs="json")
demo.launch()