app.py

'''NEURAL STYLE TRANSFER '''

import gradio as gr
import tensorflow as tf
import tensorflow_hub as hub
import PIL
from PIL import Image
import numpy as np
# import time
# import requests   
#import cv2

# !mkdir nstmodel 
# !wget -c https://storage.googleapis.com/tfhub-modules/google/magenta/arbitrary-image-stylization-v1-256/2.tar.gz -O - | tar -xz -C /nstmodel
# import tensorflow.keras

# from PIL import Image, ImageOps
#import requests
#import tarfile

#MODEL_PATH='Nst_model'

# Disable scientific notation for clarity
np.set_printoptions(suppress=True)

# Load model from TF-Hub
model = hub.load('https://tfhub.dev/google/magenta/arbitrary-image-stylization-v1-256/2') 

# Load the model
#model = tf.keras.models.load_model(MODEL_PATH)

def tensor_to_image(tensor):
  tensor = tensor*255
  tensor = np.array(tensor, dtype=np.uint8)
  if np.ndim(tensor)>3:
    assert tensor.shape[0] == 1
    tensor = tensor[0]
  return PIL.Image.fromarray(tensor)



"""## Grayscaling image for testing purpose to check if we could get better results.
def gray_scaled(inp_img):
  gray = cv2.cvtColor(inp_img, cv2.COLOR_BGR2GRAY)
  gray_img = np.zeros_like(inp_img)
  gray_img[:,:,0] = gray
  gray_img[:,:,1] = gray
  gray_img[:,:,2] = gray
  return gray_img
 """

##Transformation 
def transform_my_model(content_image,style_image):
  # Convert to float32 numpy array, add batch dimension, and normalize to range [0, 1]
  #content_image=gray_scaled(content_image)
  content_image = content_image.astype(np.float32)[np.newaxis, ...] / 255.
  style_image = style_image.astype(np.float32)[np.newaxis, ...] / 255.
 
  #Resizing image
  #style_image = tf.image.resize(style_image, (256, 256))

  # Stylize image
  outputs = model(tf.constant(content_image), tf.constant(style_image))
  stylized_image = outputs[0]

  # stylized = tf.image.resize(stylized_image, (356, 356))
  stylized_image =tensor_to_image(stylized_image)
  return stylized_image


image1 = gr.inputs.Image(label="Content Image") #CONTENT IMAGE
image2 = gr.inputs.Image(label="Style Image") #STYLE IMAGE
stylizedimg=gr.outputs.Image(label="Result") 
gr.Interface(fn=transform_my_model, inputs= [image1,image2] , outputs= stylizedimg,title='Style Transfer',theme='seafoam',examples=[['Content_Images/contnt12.jpg','VG516.jpg']],article="References-\n\nExploring the structure of a real-time, arbitrary neural artistic stylization network. Golnaz Ghiasi, Honglak Lee, Manjunath Kudlur, Vincent Dumoulin.").launch(debug=True)