app.py

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






# Initial parameters for pretrained model
IMG_SIZE = 300



labelInfo = {
    'lower-gi-tract anatomical-landmarks cecum': 0,
    'lower-gi-tract anatomical-landmarks ileum': 1,
    'lower-gi-tract anatomical-landmarks retroflex-rectum': 2,
    'lower-gi-tract pathological-findings hemorrhoids': 3,
    'lower-gi-tract pathological-findings polyps': 4,
    'lower-gi-tract pathological-findings ulcerative-colitis-grade-0-1': 5,
    'lower-gi-tract pathological-findings ulcerative-colitis-grade-1': 6,
    'lower-gi-tract pathological-findings ulcerative-colitis-grade-1-2': 7,
    'lower-gi-tract pathological-findings ulcerative-colitis-grade-2': 8,
    'lower-gi-tract pathological-findings ulcerative-colitis-grade-2-3': 9,
    'lower-gi-tract pathological-findings ulcerative-colitis-grade-3': 10,
    'lower-gi-tract quality-of-mucosal-views bbps-0-1': 11,
    'lower-gi-tract quality-of-mucosal-views bbps-2-3': 12,
    'lower-gi-tract quality-of-mucosal-views impacted-stool': 13,
    'lower-gi-tract therapeutic-interventions dyed-lifted-polyps': 14,
    'lower-gi-tract therapeutic-interventions dyed-resection-margins': 15,
    'upper-gi-tract anatomical-landmarks pylorus': 16,
    'upper-gi-tract anatomical-landmarks retroflex-stomach': 17,
    'upper-gi-tract anatomical-landmarks z-line': 18,
    'upper-gi-tract pathological-findings barretts': 19,
    'upper-gi-tract pathological-findings barretts-short-segment': 20,
    'upper-gi-tract pathological-findings esophagitis-a': 21,
    'upper-gi-tract pathological-findings esophagitis-b-d': 22
}
# Load the model from the H5 file
model = tf.keras.models.load_model('model/Hyper.h5')

# Define the prediction function
def predict(img):
    img_height = 300
    img_width = 300

    # Convert the NumPy array to a PIL Image object
    pil_img = Image.fromarray(img)

    # Resize the image using the PIL Image object
    pil_img = pil_img.resize((img_height, img_width))

    # Convert the PIL Image object to a NumPy array
    x = tf.keras.preprocessing.image.img_to_array(pil_img)

    x = x.reshape(1, img_height, img_width, 3)
    np.set_printoptions(formatter={'float': '{: 0.3f}'.format})


    predi = model.predict(x)
    accuracy_of_class = '{:.1f}'.format(predi[0][np.argmax(predi)] * 100) + "%"
    classes = list(labelInfo.keys())[np.argmax(predi)]
    context = {
        'predictedLabel': classes,
        # 'y_class': y_class,
        # 'z_class': z_class,
        'accuracy_of_class': accuracy_of_class
    }


   
    return context



demo = gr.Interface(fn=predict, inputs="image", outputs="text" , examples=[["O1.jpg"],["O2.jpg"],["O3.jpg"]],)

demo.launch()