{ "cells": [ { "cell_type": "markdown", "metadata": { "id": "ScitaPqhKtuW" }, "source": [ "##### Copyright 2019 The TensorFlow Hub Authors.\n", "\n", "Licensed under the Apache License, Version 2.0 (the \"License\");" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "jvztxQ6VsK2k" }, "outputs": [], "source": [ "# Copyright 2019 The TensorFlow Hub Authors. All Rights Reserved.\n", "#\n", "# Licensed under the Apache License, Version 2.0 (the \"License\");\n", "# you may not use this file except in compliance with the License.\n", "# You may obtain a copy of the License at\n", "#\n", "# http://www.apache.org/licenses/LICENSE-2.0\n", "#\n", "# Unless required by applicable law or agreed to in writing, software\n", "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", "# See the License for the specific language governing permissions and\n", "# limitations under the License.\n", "# ==============================================================================" ] }, { "cell_type": "markdown", "metadata": { "id": "oXlcl8lqBgAD" }, "source": [ "# Fast Style Transfer for Arbitrary Styles\n" ] }, { "cell_type": "markdown", "metadata": { "id": "MfBg1C5NB3X0" }, "source": [ "\n", " \n", " \n", " \n", " \n", " \n", "
\n", " View on TensorFlow.org\n", " \n", " Run in Google Colab\n", " \n", " View on GitHub\n", " \n", " Download notebook\n", " \n", " See TF Hub model\n", "
" ] }, { "cell_type": "markdown", "metadata": { "id": "YeeuYzbZcJzs" }, "source": [ "Based on the model code in [magenta](https://github.com/tensorflow/magenta/tree/master/magenta/models/arbitrary_image_stylization) and the publication:\n", "\n", "[Exploring the structure of a real-time, arbitrary neural artistic stylization\n", "network](https://arxiv.org/abs/1705.06830).\n", "*Golnaz Ghiasi, Honglak Lee,\n", "Manjunath Kudlur, Vincent Dumoulin, Jonathon Shlens*,\n", "Proceedings of the British Machine Vision Conference (BMVC), 2017.\n" ] }, { "cell_type": "markdown", "metadata": { "id": "TaM8BVxrCA2E" }, "source": [ "## Setup" ] }, { "cell_type": "markdown", "metadata": { "id": "J65jog2ncJzt" }, "source": [ "Let's start with importing TF2 and all relevant dependencies." ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "v-KXRY5XBu2u" }, "outputs": [], "source": [ "import functools\n", "import os\n", "\n", "from matplotlib import gridspec\n", "import matplotlib.pylab as plt\n", "import numpy as np\n", "import tensorflow as tf\n", "import tensorflow_hub as hub\n", "\n", "print(\"TF Version: \", tf.__version__)\n", "print(\"TF Hub version: \", hub.__version__)\n", "print(\"Eager mode enabled: \", tf.executing_eagerly())\n", "print(\"GPU available: \", tf.config.list_physical_devices('GPU'))" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "tsoDv_9geoZn" }, "outputs": [], "source": [ "# @title Define image loading and visualization functions { display-mode: \"form\" }\n", "\n", "def crop_center(image):\n", " \"\"\"Returns a cropped square image.\"\"\"\n", " shape = image.shape\n", " new_shape = min(shape[1], shape[2])\n", " offset_y = max(shape[1] - shape[2], 0) // 2\n", " offset_x = max(shape[2] - shape[1], 0) // 2\n", " image = tf.image.crop_to_bounding_box(\n", " image, offset_y, offset_x, new_shape, new_shape)\n", " return image\n", "\n", "@functools.lru_cache(maxsize=None)\n", "def load_image(image_url, image_size=(256, 256), preserve_aspect_ratio=True):\n", " \"\"\"Loads and preprocesses images.\"\"\"\n", " # Cache image file locally.\n", " image_path = tf.keras.utils.get_file(os.path.basename(image_url)[-128:], image_url)\n", " # Load and convert to float32 numpy array, add batch dimension, and normalize to range [0, 1].\n", " img = tf.io.decode_image(\n", " tf.io.read_file(image_path),\n", " channels=3, dtype=tf.float32)[tf.newaxis, ...]\n", " img = crop_center(img)\n", " img = tf.image.resize(img, image_size, preserve_aspect_ratio=True)\n", " return img\n", "\n", "def show_n(images, titles=('',)):\n", " n = len(images)\n", " image_sizes = [image.shape[1] for image in images]\n", " w = (image_sizes[0] * 6) // 320\n", " plt.figure(figsize=(w * n, w))\n", " gs = gridspec.GridSpec(1, n, width_ratios=image_sizes)\n", " for i in range(n):\n", " plt.subplot(gs[i])\n", " plt.imshow(images[i][0], aspect='equal')\n", " plt.axis('off')\n", " plt.title(titles[i] if len(titles) > i else '')\n", " plt.show()\n" ] }, { "cell_type": "markdown", "metadata": { "id": "8etHh05-CJHc" }, "source": [ "Let's get as well some images to play with." ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "dRc0vat3Alzo" }, "outputs": [], "source": [ "# @title Load example images { display-mode: \"form\" }\n", "\n", "content_image_url = 'https://upload.wikimedia.org/wikipedia/commons/thumb/f/fd/Golden_Gate_Bridge_from_Battery_Spencer.jpg/640px-Golden_Gate_Bridge_from_Battery_Spencer.jpg' # @param {type:\"string\"}\n", "style_image_url = 'https://upload.wikimedia.org/wikipedia/commons/0/0a/The_Great_Wave_off_Kanagawa.jpg' # @param {type:\"string\"}\n", "output_image_size = 384 # @param {type:\"integer\"}\n", "\n", "# The content image size can be arbitrary.\n", "content_img_size = (output_image_size, output_image_size)\n", "# The style prediction model was trained with image size 256 and it's the \n", "# recommended image size for the style image (though, other sizes work as \n", "# well but will lead to different results).\n", "style_img_size = (256, 256) # Recommended to keep it at 256.\n", "\n", "content_image = load_image(content_image_url, content_img_size)\n", "style_image = load_image(style_image_url, style_img_size)\n", "style_image = tf.nn.avg_pool(style_image, ksize=[3,3], strides=[1,1], padding='SAME')\n", "show_n([content_image, style_image], ['Content image', 'Style image'])" ] }, { "cell_type": "markdown", "metadata": { "id": "yL2Bn5ThR1nY" }, "source": [ "## Import TF Hub module" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "467AVDSuzBPc" }, "outputs": [], "source": [ "# Load TF Hub module.\n", "\n", "hub_handle = 'https://tfhub.dev/google/magenta/arbitrary-image-stylization-v1-256/2'\n", "hub_module = hub.load(hub_handle)" ] }, { "cell_type": "markdown", "metadata": { "id": "uAR70_3wLEDB" }, "source": [ "The signature of this hub module for image stylization is:\n", "```\n", "outputs = hub_module(content_image, style_image)\n", "stylized_image = outputs[0]\n", "```\n", "Where `content_image`, `style_image`, and `stylized_image` are expected to be 4-D Tensors with shapes `[batch_size, image_height, image_width, 3]`.\n", "\n", "In the current example we provide only single images and therefore the batch dimension is 1, but one can use the same module to process more images at the same time.\n", "\n", "The input and output values of the images should be in the range [0, 1].\n", "\n", "The shapes of content and style image don't have to match. Output image shape\n", "is the same as the content image shape." ] }, { "cell_type": "markdown", "metadata": { "id": "qEhYJno1R7rP" }, "source": [ "## Demonstrate image stylization" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "lnAv-F3O9fLV" }, "outputs": [], "source": [ "# Stylize content image with given style image.\n", "# This is pretty fast within a few milliseconds on a GPU.\n", "\n", "outputs = hub_module(tf.constant(content_image), tf.constant(style_image))\n", "stylized_image = outputs[0]" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "OEAPEdq698gs" }, "outputs": [], "source": [ "# Visualize input images and the generated stylized image.\n", "\n", "show_n([content_image, style_image, stylized_image], titles=['Original content image', 'Style image', 'Stylized image'])" ] }, { "cell_type": "markdown", "metadata": { "id": "v-gYvjTWK-lx" }, "source": [ "## Let's try it on more images" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "WSMaY0YBNfkK" }, "outputs": [], "source": [ "# @title To Run: Load more images { display-mode: \"form\" }\n", "\n", "content_urls = dict(\n", " sea_turtle='https://upload.wikimedia.org/wikipedia/commons/d/d7/Green_Sea_Turtle_grazing_seagrass.jpg',\n", " tuebingen='https://upload.wikimedia.org/wikipedia/commons/0/00/Tuebingen_Neckarfront.jpg',\n", " grace_hopper='https://storage.googleapis.com/download.tensorflow.org/example_images/grace_hopper.jpg',\n", " )\n", "style_urls = dict(\n", " kanagawa_great_wave='https://upload.wikimedia.org/wikipedia/commons/0/0a/The_Great_Wave_off_Kanagawa.jpg',\n", " kandinsky_composition_7='https://upload.wikimedia.org/wikipedia/commons/b/b4/Vassily_Kandinsky%2C_1913_-_Composition_7.jpg',\n", " hubble_pillars_of_creation='https://upload.wikimedia.org/wikipedia/commons/6/68/Pillars_of_creation_2014_HST_WFC3-UVIS_full-res_denoised.jpg',\n", " van_gogh_starry_night='https://upload.wikimedia.org/wikipedia/commons/thumb/e/ea/Van_Gogh_-_Starry_Night_-_Google_Art_Project.jpg/1024px-Van_Gogh_-_Starry_Night_-_Google_Art_Project.jpg',\n", " turner_nantes='https://upload.wikimedia.org/wikipedia/commons/b/b7/JMW_Turner_-_Nantes_from_the_Ile_Feydeau.jpg',\n", " munch_scream='https://upload.wikimedia.org/wikipedia/commons/c/c5/Edvard_Munch%2C_1893%2C_The_Scream%2C_oil%2C_tempera_and_pastel_on_cardboard%2C_91_x_73_cm%2C_National_Gallery_of_Norway.jpg',\n", " picasso_demoiselles_avignon='https://upload.wikimedia.org/wikipedia/en/4/4c/Les_Demoiselles_d%27Avignon.jpg',\n", " picasso_violin='https://upload.wikimedia.org/wikipedia/en/3/3c/Pablo_Picasso%2C_1911-12%2C_Violon_%28Violin%29%2C_oil_on_canvas%2C_Kr%C3%B6ller-M%C3%BCller_Museum%2C_Otterlo%2C_Netherlands.jpg',\n", " picasso_bottle_of_rum='https://upload.wikimedia.org/wikipedia/en/7/7f/Pablo_Picasso%2C_1911%2C_Still_Life_with_a_Bottle_of_Rum%2C_oil_on_canvas%2C_61.3_x_50.5_cm%2C_Metropolitan_Museum_of_Art%2C_New_York.jpg',\n", " fire='https://upload.wikimedia.org/wikipedia/commons/3/36/Large_bonfire.jpg',\n", " derkovits_woman_head='https://upload.wikimedia.org/wikipedia/commons/0/0d/Derkovits_Gyula_Woman_head_1922.jpg',\n", " amadeo_style_life='https://upload.wikimedia.org/wikipedia/commons/8/8e/Untitled_%28Still_life%29_%281913%29_-_Amadeo_Souza-Cardoso_%281887-1918%29_%2817385824283%29.jpg',\n", " derkovtis_talig='https://upload.wikimedia.org/wikipedia/commons/3/37/Derkovits_Gyula_Talig%C3%A1s_1920.jpg',\n", " amadeo_cardoso='https://upload.wikimedia.org/wikipedia/commons/7/7d/Amadeo_de_Souza-Cardoso%2C_1915_-_Landscape_with_black_figure.jpg'\n", ")\n", "\n", "content_image_size = 384\n", "style_image_size = 256\n", "content_images = {k: load_image(v, (content_image_size, content_image_size)) for k, v in content_urls.items()}\n", "style_images = {k: load_image(v, (style_image_size, style_image_size)) for k, v in style_urls.items()}\n", "style_images = {k: tf.nn.avg_pool(style_image, ksize=[3,3], strides=[1,1], padding='SAME') for k, style_image in style_images.items()}\n" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "dqB6aNTLNVkK" }, "outputs": [], "source": [ "#@title Specify the main content image and the style you want to use. { display-mode: \"form\" }\n", "\n", "content_name = 'sea_turtle' # @param ['sea_turtle', 'tuebingen', 'grace_hopper']\n", "style_name = 'munch_scream' # @param ['kanagawa_great_wave', 'kandinsky_composition_7', 'hubble_pillars_of_creation', 'van_gogh_starry_night', 'turner_nantes', 'munch_scream', 'picasso_demoiselles_avignon', 'picasso_violin', 'picasso_bottle_of_rum', 'fire', 'derkovits_woman_head', 'amadeo_style_life', 'derkovtis_talig', 'amadeo_cardoso']\n", "\n", "stylized_image = hub_module(tf.constant(content_images[content_name]),\n", " tf.constant(style_images[style_name]))[0]\n", "\n", "show_n([content_images[content_name], style_images[style_name], stylized_image],\n", " titles=['Original content image', 'Style image', 'Stylized image'])" ] } ], "metadata": { "accelerator": "GPU", "colab": { "collapsed_sections": [], "name": "TF-Hub: Fast Style Transfer for Arbitrary Styles.ipynb", "private_outputs": true, "provenance": [], "toc_visible": true }, "kernelspec": { "display_name": "Python 3", "name": "python3" } }, "nbformat": 4, "nbformat_minor": 0 }