png

Neural style transfer is a technique for blending two images—a content image and a style reference image (such as a famous painter’s work)—so that the output image appears like the content image but is “painted” in the manner of the style reference image. This is accomplished by adjusting the output image’s content statistics to match the content image’s content statistics and the style reference image’s style statistics. A convolutional network is used to extract these data from the images. We’ll be using TensorFlow in this project with a pretrained Neural Style Transfer Model. Make sure to have TensorFlow installed before proceeding.

Github link for project

import tensorflow_hub as hub
import tensorflow as tf
from matplotlib import pyplot as plt
import numpy as np
import cv2

Download Pretrained Model

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

Load Image

def load_image(img_path):
    img = tf.io.read_file(img_path)
    img = tf.image.decode_image(img, channels=3)
    img = tf.image.convert_image_dtype(img, tf.float32)
    img = img[tf.newaxis, :]
    return img

load_mage function will load and preprocess our image before passing it through the model

content_image = load_image('cat.jpg')
style_image = load_image('starrynight.jfif')

Preview Original Images

content_image.shape

TensorShape([1, 1440, 2560, 3])

plt.imshow(np.squeeze(content_image))
plt.show()

png

This is the base picture(my lovely cat) that we want to apply our style effect

plt.imshow(np.squeeze(style_image))
plt.show()

png

This is the image style that we will try to apply on our base picture, which in this case we’ll use the Starry Night painting by Vincent van Gogh

Stylize Image

stylized_image = model(tf.constant(content_image), tf.constant(style_image))[0]

fig = plt.figure(figsize=(8, 8))
plt.imshow(np.squeeze(stylized_image))
plt.show()

png

Voila! here we can see my cat picture but in the style of Vincent van Gogh’s Starry Night. Pretty neat huh?

Export Image

Let’s export this image so you can post it online and show how artsy you are on social media!

cv2.imwrite('generated_img.jpg', cv2.cvtColor(np.squeeze(stylized_image)*255, cv2.COLOR_BGR2RGB))