app.py

import gradio as gr
import keras
from keras.models import load_model
# from keras_contrib.layers.normalization.instancenormalization import InstanceNormalization
from tensorflow_addons.layers import InstanceNormalization
import matplotlib.pyplot as plt
import numpy as np
import tensorflow as tf


cust = {'InstanceNormalization': InstanceNormalization}
model=load_model('g-cycleGAN-photo2monet-500images-epoch10_30_30_30_30.h5',cust)

path  = [['ex1.jpg'], ['ex2.jpg'], ['ex3.jpg'],['ex4.jpg'],['ex5.jpg']]

# preprocess
AUTOTUNE = tf.data.AUTOTUNE
BUFFER_SIZE = 400
BATCH_SIZE = 1
IMG_WIDTH = 256
IMG_HEIGHT = 256

def resize(image,height,width):
  '''
    Resizing the image
  '''
  resized_image = tf.image.resize(image,[height,width],method = tf.image.ResizeMethod.NEAREST_NEIGHBOR)
  return resized_image

def normalize(input_image):
  # def normalize(real_image, input_image)
  input_image = (input_image/127.5) - 1
  return input_image
#   real_image = (real_image/127.5) - 1
#   return real_image,input_image

def load(img_file):
  '''
    load the image. Since we need only the target image and a 
    gray scale version of the same, we are going to load one 
    and create the other from it
  '''
  img = tf.io.read_file(img_file)
  img = tf.io.decode_jpeg(img)

#   w = tf.shape(img)[1]
#   w = w//2

#   real_image = img[:,:w,:]
  
  real_image = tf.cast(img,tf.float32)

  return real_image

def load_image_test(image_file):
  '''
  We are not using random jitter here and thus creating
  a gray scale image after resizing.
  '''
  re = load(image_file)
  re = resize(re,IMG_HEIGHT,IMG_WIDTH)
#   inp = tf.image.rgb_to_grayscale(re)
#   re,inp = normalize(re,inp)
#   inp = re
#   re, inp = normalize(re,inp)
  re = normalize(re)
#   return inp,re
  return re


def show_preds_image(image_path):
    A = load_image_test(image_path)
    # A = (A - 127.5) / 127.5
    A = np.expand_dims(A,axis=0)
    B = model(A)
    # B = np.squeeze(B,axis=0)
    B = B[0]
    B = B * 0.5 + 0.5
    B = B.numpy()
    return B
 
inputs_image = [
    gr.components.Image(shape=(256,256),type="filepath", label="Input Image"),
]
outputs_image = [
    gr.components.Image(shape=(256,256),type="numpy", label="Output Image").style(width=256, height=256),
]
interface_image = gr.Interface(
    fn=show_preds_image,
    inputs=inputs_image,
    outputs=outputs_image,
    title="photo2monet",
    examples=path,
    cache_examples=False,
)

gr.TabbedInterface(
    [interface_image],
    tab_names=['Image inference']
).queue().launch()