neural_style.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using Theano backend.\n"
     ]
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "from __future__ import print_function\n",
    "from keras.preprocessing.image import load_img, img_to_array\n",
    "from scipy.misc import imsave\n",
    "import numpy as np\n",
    "from scipy.optimize import fmin_l_bfgs_b\n",
    "import time\n",
    "\n",
    "import vgg16\n",
    "from keras import backend as K\n",
    "from keras.applications.imagenet_utils import preprocess_input"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "base_image_path = 'tubingen.jpg'\n",
    "style_reference_image_path = 'starry_night_google.jpg'\n",
    "result_prefix = 'art'\n",
    "iterations = 5\n",
    "total_variation_weight = 1.0\n",
    "style_weight = 1.0\n",
    "content_weight = 0.075"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "width, height = load_img(base_image_path).size\n",
    "img_nrows = 400\n",
    "img_ncols = int(width * img_nrows / height)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def preprocess_image(image_path):\n",
    "    img = load_img(image_path, target_size=(img_nrows, img_ncols))\n",
    "    img = img_to_array(img)\n",
    "    img = np.expand_dims(img, axis=0)\n",
    "    img = vgg16.preprocess_input(img)\n",
    "    return img\n",
    "\n",
    "def deprocess_image(x):\n",
    "    x = x.reshape((3, img_nrows, img_ncols))\n",
    "    x = x.transpose((1, 2, 0))\n",
    "    # Remove zero-center by mean pixel\n",
    "    x[:, :, 0] += 103.939\n",
    "    x[:, :, 1] += 116.779\n",
    "    x[:, :, 2] += 123.68\n",
    "    # 'BGR'->'RGB'\n",
    "    x = x[:, :, ::-1]\n",
    "    x = np.clip(x, 0, 255).astype('uint8')\n",
    "    return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "base_image = K.variable(preprocess_image(base_image_path))\n",
    "style_reference_image = K.variable(preprocess_image(style_reference_image_path))\n",
    "combination_image = K.placeholder((1, 3, img_nrows, img_ncols))\n",
    "input_tensor = K.concatenate([base_image,\n",
    "                              style_reference_image,\n",
    "                              combination_image], axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model loaded.\n"
     ]
    }
   ],
   "source": [
    "model = vgg16.VGG16(input_tensor=input_tensor,\n",
    "                    weights='imagenet', include_top=False)\n",
    "print('Model loaded.')\n",
    "outputs_dict = dict([(layer.name, layer.output) for layer in model.layers])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def gram_matrix(x):\n",
    "    assert K.ndim(x) == 3\n",
    "    features = K.batch_flatten(x)\n",
    "    gram = K.dot(features, K.transpose(features))\n",
    "    return gram\n",
    "\n",
    "def style_loss(style, combination):\n",
    "    assert K.ndim(style) == 3\n",
    "    assert K.ndim(combination) == 3\n",
    "    S = gram_matrix(style)\n",
    "    C = gram_matrix(combination)\n",
    "    channels = 3\n",
    "    size = img_nrows * img_ncols\n",
    "    return K.sum(K.square(S - C)) / (4. * (channels ** 2) * (size ** 2))\n",
    "\n",
    "def content_loss(base, combination):\n",
    "    return K.sum(K.square(combination - base))\n",
    "\n",
    "def total_variation_loss(x):\n",
    "    assert K.ndim(x) == 4\n",
    "    a = K.square(x[:, :, :img_nrows - 1, :img_ncols - 1] - x[:, :, 1:, :img_ncols - 1])\n",
    "    b = K.square(x[:, :, :img_nrows - 1, :img_ncols - 1] - x[:, :, :img_nrows - 1, 1:])\n",
    "    return K.sum(K.pow(a + b, 1.25))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "loss = K.variable(0.)\n",
    "layer_features = outputs_dict['block5_conv2']\n",
    "base_image_features = layer_features[0, :, :, :]\n",
    "combination_features = layer_features[2, :, :, :]\n",
    "loss += content_weight * content_loss(base_image_features,\n",
    "                                      combination_features)\n",
    "\n",
    "feature_layers = ['block1_conv1', 'block2_conv1',\n",
    "                  'block3_conv1', 'block4_conv1',\n",
    "                  'block5_conv1']\n",
    "for layer_name in feature_layers:\n",
    "    layer_features = outputs_dict[layer_name]\n",
    "    style_reference_features = layer_features[1, :, :, :]\n",
    "    combination_features = layer_features[2, :, :, :]\n",
    "    sl = style_loss(style_reference_features, combination_features)\n",
    "    loss += (style_weight / len(feature_layers)) * sl\n",
    "loss += total_variation_weight * total_variation_loss(combination_image)\n",
    "grads = K.gradients(loss, combination_image)\n",
    "outputs = [loss]\n",
    "if isinstance(grads, (list, tuple)):\n",
    "    outputs += grads\n",
    "else:\n",
    "    outputs.append(grads)\n",
    "\n",
    "f_outputs = K.function([combination_image], outputs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def eval_loss_and_grads(x):\n",
    "    x = x.reshape((1, 3, img_nrows, img_ncols))\n",
    "    print('.', end='')\n",
    "    outs = f_outputs([x])\n",
    "    loss_value = outs[0]\n",
    "    if len(outs[1:]) == 1:\n",
    "        grad_values = outs[1].flatten().astype('float64')\n",
    "    else:\n",
    "        grad_values = np.array(outs[1:]).flatten().astype('float64')\n",
    "    return loss_value, grad_values\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "class Evaluator(object):\n",
    "\n",
    "    def __init__(self):\n",
    "        self.loss_value = None\n",
    "        self.grads_values = None\n",
    "\n",
    "    def loss(self, x):\n",
    "        assert self.loss_value is None\n",
    "        loss_value, grad_values = eval_loss_and_grads(x)\n",
    "        self.loss_value = loss_value\n",
    "        self.grad_values = grad_values\n",
    "        return self.loss_value\n",
    "\n",
    "    def grads(self, x):\n",
    "        assert self.loss_value is not None\n",
    "        grad_values = np.copy(self.grad_values)\n",
    "        self.loss_value = None\n",
    "        self.grad_values = None\n",
    "        return grad_values\n",
    "\n",
    "evaluator = Evaluator()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(1L, 3L, 400L, 533L)\n",
      "Start of iteration 0\n",
      ".....................\n",
      "Current loss value: 2857231872.0\n",
      "Image saved as art_at_iteration_0.png\n",
      "Iteration 0 completed in 223s\n",
      "Start of iteration 1\n",
      ".....................\n",
      "Current loss value: 2399473664.0\n",
      "Image saved as art_at_iteration_1.png\n",
      "Iteration 1 completed in 223s\n",
      "Start of iteration 2\n",
      ".....................\n",
      "Current loss value: 2273068544.0\n",
      "Image saved as art_at_iteration_2.png\n",
      "Iteration 2 completed in 243s\n",
      "Start of iteration 3\n",
      ".....................\n",
      "Current loss value: 2208670208.0\n",
      "Image saved as art_at_iteration_3.png\n",
      "Iteration 3 completed in 251s\n",
      "Start of iteration 4\n",
      ".....................\n",
      "Current loss value: 2167820032.0\n",
      "Image saved as art_at_iteration_4.png\n",
      "Iteration 4 completed in 260s\n"
     ]
    }
   ],
   "source": [
    "x = preprocess_image(base_image_path)\n",
    "print(x.shape)\n",
    "for i in range(iterations):\n",
    "    print('Start of iteration', i)\n",
    "    start_time = time.time()\n",
    "    x, min_val, info = fmin_l_bfgs_b(evaluator.loss, x.flatten(), fprime=evaluator.grads, maxfun=20)\n",
    "    print()\n",
    "    print('Current loss value:', min_val)\n",
    "    # save current generated image\n",
    "    img = deprocess_image(x.copy())\n",
    "    fname = result_prefix + '_at_iteration_%d.png' % i\n",
    "    imsave(fname, img)\n",
    "    end_time = time.time()\n",
    "    print('Image saved as', fname)\n",
    "    print('Iteration %d completed in %ds' % (i, end_time - start_time))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}