app.py

##importing the libraries
import numpy as np
import pandas as pd
from PIL import Image
import cv2
import tensorflow as tf
import os 
import matplotlib.pyplot as plt
%matplotlib inline
from tensorflow.keras.models import load_model
import gradio as gr


# Load your trained model
model = load_model('tb_pretrained.h5')

### Preprocess the new image

def predict_image(test_image):
    # img = cv2.imread(test_image)
    img = np.array(test_image)
    
    image_1 = tf.image.resize(img, (256,256))

    image_processed = np.expand_dims(image_1/256, 0)
    
    ##prediction
    
    yhat = model.predict(image_processed)
    
    ## setting a threshold 
    if yhat[0][1] > 0.70:
        return (f'There is {round((yhat[0][1])*100,2)}% chance of the image being normal')
    elif yhat[0][0] > 0.9:
        return (f'There is {round((yhat[0][0])*100,2)}% chance of an abnormality either than TB being present')
    else:
        return (f'There is a chance of TB being present')


gr.Interface(predict_image, "image", "label").launch(debug=True, share=True)