import numpy as np
import pandas as pd
import matplotlib.pylab as plt
import gradio as gr
import PIL.Image as Image
import tensorflow as tf
import tensorflow_hub as hub

TF_MODEL_URL = 'https://tfhub.dev/google/on_device_vision/classifier/landmarks_classifier_asia_V1/1'
LABEL_MAP_URL = 'https://www.gstatic.com/aihub/tfhub/labelmaps/landmarks_classifier_asia_V1_label_map.csv'
IMAGE_SHAPE = (321, 321)

classifier = tf.keras.Sequential([hub.KerasLayer(TF_MODEL_URL,
                                                 input_shape=IMAGE_SHAPE+(3,),
                                                 output_key="predictions:logits")])

df = pd.read_csv(LABEL_MAP_URL)

label_map = dict(zip(df.id, df.name))

label_map

img_loc = "image.jpeg"

img = Image.open(img_loc).resize(IMAGE_SHAPE)

img

img = np.array(img)/255.0
img.shape

img = img[np.newaxis, ...]

img.shape

result = classifier.predict(img)

result

label_map[np.argmax(result)]

class_names=list(label_map.values())

def classify_image(image):
    img = np.array(image)/255.0
    img = img[np.newaxis, ...]
    prediction = classifier.predict(img)
    return label_map[np.argmax(prediction)]

image = gr.inputs.Image(shape=(321, 321))
label = gr.outputs.Label(num_top_classes=1)

gr.Interface(    
    classify_image, 
    image, 
    label,
    capture_session=True).launch(debug=True);