import streamlit as st
from transformers import pipeline
#from datasets import load_dataset, Image
from huggingface_hub import from_pretrained_keras
import keras
import numpy as np
from PIL import Image

loaded_model = keras.saving.load_model("best_model.keras")

uploaded_img = st.file_uploader("Upload your file here...",type=['png', 'jpeg', 'jpg'])

if uploaded_img is not None:
    st.image(uploaded_img)
    img = Image.open(uploaded_img)
    print(img.mode)
    resized_img = img.resize((160,160))
    print(resized_img.mode)
    input_arr = keras.preprocessing.image.img_to_array(resized_img)
    print(input_arr.shape)
    input_arr = input_arr.astype('float32') / 255.
    print(input_arr.shape)
    result = loaded_model.predict(input_arr)
    st.write(f"Your prediction is: {result}")


#model = from_pretrained_keras("jableable/road_model")

#pipe = pipeline('sentiment-analysis')
#text = st.text_area('enter some text!')

#if text:
    #out = pipe(text)
    #st.json(out)

#loaded_model = keras.saving.load_model("jableable/road_model")

#model = from_pretrained_keras("keras-io/ocr-for-captcha")
#model.summary()
#prediction = model.predict(image)
#prediction = tf.squeeze(tf.round(prediction))
#print(f'The image is a {classes[(np.argmax(prediction))]}!')


#dataset = load_dataset("beans", split="train")

#loaded_img = dataset[0]["image"]
#print(loaded_img)