import tensorflow as tf
import requests
import gradio as gr

inception_net = tf.keras.applications.MobileNetV2()

# Obteniendo las labels de "https://git.io/JJkYN"
respuesta = requests.get("https://git.io/JJkYN")
etiquetas = respuesta.text.split("\n")

def clasifica_imagen(inp):
  inp = inp.reshape((-1, 224, 224, 3))
  inp = tf.keras.applications.mobilenet_v2.preprocess_input(inp)
  prediction = inception_net.predict(inp).flatten()
  confidences = {etiquetas[i]: float(prediction[i]) for i in range(1000)}
  return confidences
 
demo = gr.Interface(fn=clasifica_imagen, 
             inputs=gr.Image(shape=(224, 224)),
             outputs=gr.Label(num_top_classes=3)
             )

demo.launch()