import streamlit as st
import tensorflow as tf
import cv2
import numpy as np
from patchify import patchify
from huggingface_hub import from_pretrained_keras

model = from_pretrained_keras('ErnestBeckham/MulticancerViT')

hp = {}
hp['image_size'] = 512
hp['num_channels'] = 3
hp['patch_size'] = 64
hp['num_patches'] = (hp['image_size']**2) // (hp["patch_size"]**2)
hp["flat_patches_shape"] = (hp["num_patches"], hp['patch_size']*hp['patch_size']*hp["num_channels"])
hp['class_names'] = ['cervix_koc',
                      'cervix_dyk',
                      'cervix_pab',
                      'cervix_sfi',
                      'cervix_mep',
                      'colon_bnt',
                      'colon_aca',
                      'lung_aca',
                      'lung_bnt',
                      'lung_scc',
                      'oral_scc',
                      'oral_normal',
                      'kidney_tumor',
                      'kidney_normal',
                      'breast_benign',
                      'breast_malignant',
                      'lymph_fl',
                      'lymph_cll',
                      'lymph_mcl',
                      'brain_tumor',
                      'brain_glioma',
                      'brain_menin']

def main():
    st.title("Multi-Cancer Classification")

    # Upload image through drag and drop
    uploaded_file = st.file_uploader("Choose an image...", type=["jpg", "jpeg", "png"])

    if uploaded_file is not None:
        # Convert the uploaded file to OpenCV format
        image = convert_to_opencv(uploaded_file)
        
        # Display the uploaded image
        st.image(image, channels="BGR", caption="Uploaded Image", use_column_width=True)

        # Display the image shape
        image_class = predict_single_image(image, model, hp)
        st.write(f"Image Class: {image_class}")

def convert_to_opencv(uploaded_file):
    # Read the uploaded file using OpenCV
    image_bytes = uploaded_file.read()
    np_arr = np.frombuffer(image_bytes, np.uint8)
    image = cv2.imdecode(np_arr, cv2.IMREAD_COLOR)
    return image

def detect_image_shape(image):
    # Get the image shape
    return image.shape

def preprocess_image(image, hp):
    # Resize the image to the expected input size
    image = cv2.resize(image, (hp['image_size'], hp['image_size']))
    # Normalize pixel values to be in the range [0, 1]
    image = image / 255.0
    # Extract patches using the same patching mechanism as during training
    patch_shape = (hp['patch_size'], hp['patch_size'], hp['num_channels'])
    patches = patchify(image, patch_shape, hp['patch_size'])
    # Flatten the patches
    patches = np.reshape(patches, hp['flat_patches_shape'])
    # Convert the flattened patches into a format suitable for prediction
    patches = patches.astype(np.float32)
    
    return patches

def predict_single_image(image, model, hp):
    # Preprocess the image
    preprocessed_image = preprocess_image(image, hp)
    # Convert the preprocessed image to a TensorFlow tensor if needed
    preprocessed_image = tf.convert_to_tensor(preprocessed_image)
    # Add an extra batch dimension (required for model.predict)
    preprocessed_image = tf.expand_dims(preprocessed_image, axis=0)
    # Make the prediction
    predictions = model.predict(preprocessed_image)

    np.around(predictions)
    y_pred_classes = np.argmax(predictions, axis=1)
    class_name = hp['class_names'][y_pred_classes[0]]
    return class_name
    
if __name__ == "__main__":
    main()