app.py

import gradio as gr
import cv2
import requests
import gdown
import tensorflow as tf
from tensorflow import keras
import numpy as np
from PIL import Image
from keras.models import load_model

url = 'https://drive.google.com/file/d/1ueGC-MBuVip9jkfLJdfeH-RBy6fjDZ5j/view?usp=sharing'
output_path = 'classes.txt'
gdown.download(url, output_path, quiet=False,fuzzy=True)

with open(output_path,'r') as file:
    LABELS = [x.strip() for x in file.readlines()]


num_classes = 200
IMG_SIZE = 128

model = load_model('model.h5')

# Preprocess image
def preprocess(image):
  image = image.convert('RGB') # To RGB
  image = image.resize((IMG_SIZE, IMG_SIZE)) # Resize the image to 128 x 128
  image = np.array(image)
  image = tf.keras.applications.mobilenet_v2.preprocess_input(image) # Rescale the pixel to [-1,1] for MobileNetV2
  return image


def predict(image):
     image = preprocess(image)
     image = np.expand_dims(image, axis=0) # Add batch dimension
     prediction = model.predict(image) # Predict the image using the trained model

     # Get the top 3 predictions
     idx = np.argsort(prediction[0])[::-1][:3]
     top3_value = np.asarray([prediction[0][i] for i in idx[0:3]])
     top3_idx = idx[0:3]

     return {LABELS[i]:str(v) for i,v in zip(top3_idx,top3_value)}

# Create Gradio interface
input_image = gr.inputs.Image(type='pil')
output_text = 'label'
app = gr.Interface(fn=predict, inputs=input_image, outputs=output_text)
app.launch()