In [ ]:
#@title Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
Note: 我们的 TensorFlow 社区翻译了这些文档。因为社区翻译是尽力而为, 所以无法保证它们是最准确的,并且反映了最新的 官方英文文档。如果您有改进此翻译的建议, 请提交 pull request 到 tensorflow/docs GitHub 仓库。要志愿地撰写或者审核译文,请加入 docs-zh-cn@tensorflow.org Google Group。
本教程使用深度学习来用其他图像的风格创造一个图像(曾经你是否希望可以像毕加索或梵高一样绘画?)。 这被称为神经风格迁移,该技术概述于 A Neural Algorithm of Artistic Style (Gatys et al.).
Note: 本教程演示了原始的风格迁移算法。它将图像内容优化为特定样式。最新的一些方法训练模型以直接生成风格化图像(类似于 cyclegan)。原始的这种方法要快得多(高达 1000 倍)。TensorFlow Hub 和 TensorFlow Lite 中提供了预训练的任意图像风格化模块。
神经风格迁移是一种优化技术,用于将两个图像——一个内容图像和一个风格参考图像(如著名画家的一个作品)——混合在一起,使输出的图像看起来像内容图像, 但是用了风格参考图像的风格。
这是通过优化输出图像以匹配内容图像的内容统计数据和风格参考图像的风格统计数据来实现的。 这些统计数据可以使用卷积网络从图像中提取。
例如,我们选取这张小狗的照片和 Wassily Kandinsky 的作品 7:
黄色拉布拉多犬的凝视,来自 Wikimedia Commons
如果 Kandinsky 决定用这种风格来专门描绘这只海龟会是什么样子? 是否如下图一样?
In [ ]:
import tensorflow as tf
In [ ]:
import IPython.display as display
import matplotlib.pyplot as plt
import matplotlib as mpl
mpl.rcParams['figure.figsize'] = (12,12)
mpl.rcParams['axes.grid'] = False
import numpy as np
import PIL.Image
import time
import functools
In [ ]:
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)
下载图像并选择风格图像和内容图像:
In [ ]:
content_path = tf.keras.utils.get_file('YellowLabradorLooking_new.jpg', 'https://storage.googleapis.com/download.tensorflow.org/example_images/YellowLabradorLooking_new.jpg')
# https://commons.wikimedia.org/wiki/File:Vassily_Kandinsky,_1913_-_Composition_7.jpg
style_path = tf.keras.utils.get_file('kandinsky5.jpg','https://storage.googleapis.com/download.tensorflow.org/example_images/Vassily_Kandinsky%2C_1913_-_Composition_7.jpg')
定义一个加载图像的函数,并将其最大尺寸限制为 512 像素。
In [ ]:
def load_img(path_to_img):
max_dim = 512
img = tf.io.read_file(path_to_img)
img = tf.image.decode_image(img, channels=3)
img = tf.image.convert_image_dtype(img, tf.float32)
shape = tf.cast(tf.shape(img)[:-1], tf.float32)
long_dim = max(shape)
scale = max_dim / long_dim
new_shape = tf.cast(shape * scale, tf.int32)
img = tf.image.resize(img, new_shape)
img = img[tf.newaxis, :]
return img
创建一个简单的函数来显示图像:
In [ ]:
def imshow(image, title=None):
if len(image.shape) > 3:
image = tf.squeeze(image, axis=0)
plt.imshow(image)
if title:
plt.title(title)
In [ ]:
content_image = load_img(content_path)
style_image = load_img(style_path)
plt.subplot(1, 2, 1)
imshow(content_image, 'Content Image')
plt.subplot(1, 2, 2)
imshow(style_image, 'Style Image')
本教程演示了原始的风格迁移算法。其将图像内容优化为特定风格。在进入细节之前,让我们看一下 TensorFlow Hub 模块如何快速风格迁移:
In [ ]:
import tensorflow_hub as hub
hub_module = hub.load('https://tfhub.dev/google/magenta/arbitrary-image-stylization-v1-256/1')
stylized_image = hub_module(tf.constant(content_image), tf.constant(style_image))[0]
tensor_to_image(stylized_image)
加载 VGG19 并在我们的图像上测试它以确保正常运行:
In [ ]:
x = tf.keras.applications.vgg19.preprocess_input(content_image*255)
x = tf.image.resize(x, (224, 224))
vgg = tf.keras.applications.VGG19(include_top=True, weights='imagenet')
prediction_probabilities = vgg(x)
prediction_probabilities.shape
In [ ]:
predicted_top_5 = tf.keras.applications.vgg19.decode_predictions(prediction_probabilities.numpy())[0]
[(class_name, prob) for (number, class_name, prob) in predicted_top_5]
现在,加载没有分类部分的 VGG19
,并列出各层的名称:
In [ ]:
vgg = tf.keras.applications.VGG19(include_top=False, weights='imagenet')
print()
for layer in vgg.layers:
print(layer.name)
从网络中选择中间层的输出以表示图像的风格和内容:
In [ ]:
# 内容层将提取出我们的 feature maps (特征图)
content_layers = ['block5_conv2']
# 我们感兴趣的风格层
style_layers = ['block1_conv1',
'block2_conv1',
'block3_conv1',
'block4_conv1',
'block5_conv1']
num_content_layers = len(content_layers)
num_style_layers = len(style_layers)
那么,为什么我们预训练的图像分类网络中的这些中间层的输出允许我们定义风格和内容的表示?
从高层理解,为了使网络能够实现图像分类(该网络已被训练过),它必须理解图像。 这需要将原始图像作为输入像素并构建内部表示,这个内部表示将原始图像像素转换为对图像中存在的 feature (特征)的复杂理解。
这也是卷积神经网络能够很好地推广的一个原因:它们能够捕获不变性并定义类别(例如猫与狗)之间的 feature (特征),这些 feature (特征)与背景噪声和其他干扰无关。 因此,将原始图像传递到模型输入和分类标签输出之间的某处的这一过程,可以视作复杂的 feature (特征)提取器。通过这些模型的中间层,我们就可以描述输入图像的内容和风格。
In [ ]:
def vgg_layers(layer_names):
""" Creates a vgg model that returns a list of intermediate output values."""
# 加载我们的模型。 加载已经在 imagenet 数据上预训练的 VGG
vgg = tf.keras.applications.VGG19(include_top=False, weights='imagenet')
vgg.trainable = False
outputs = [vgg.get_layer(name).output for name in layer_names]
model = tf.keras.Model([vgg.input], outputs)
return model
然后建立模型:
In [ ]:
style_extractor = vgg_layers(style_layers)
style_outputs = style_extractor(style_image*255)
#查看每层输出的统计信息
for name, output in zip(style_layers, style_outputs):
print(name)
print(" shape: ", output.numpy().shape)
print(" min: ", output.numpy().min())
print(" max: ", output.numpy().max())
print(" mean: ", output.numpy().mean())
print()
In [ ]:
def gram_matrix(input_tensor):
result = tf.linalg.einsum('bijc,bijd->bcd', input_tensor, input_tensor)
input_shape = tf.shape(input_tensor)
num_locations = tf.cast(input_shape[1]*input_shape[2], tf.float32)
return result/(num_locations)
构建一个返回风格和内容张量的模型。
In [ ]:
class StyleContentModel(tf.keras.models.Model):
def __init__(self, style_layers, content_layers):
super(StyleContentModel, self).__init__()
self.vgg = vgg_layers(style_layers + content_layers)
self.style_layers = style_layers
self.content_layers = content_layers
self.num_style_layers = len(style_layers)
self.vgg.trainable = False
def call(self, inputs):
"Expects float input in [0,1]"
inputs = inputs*255.0
preprocessed_input = tf.keras.applications.vgg19.preprocess_input(inputs)
outputs = self.vgg(preprocessed_input)
style_outputs, content_outputs = (outputs[:self.num_style_layers],
outputs[self.num_style_layers:])
style_outputs = [gram_matrix(style_output)
for style_output in style_outputs]
content_dict = {content_name:value
for content_name, value
in zip(self.content_layers, content_outputs)}
style_dict = {style_name:value
for style_name, value
in zip(self.style_layers, style_outputs)}
return {'content':content_dict, 'style':style_dict}
在图像上调用此模型,可以返回 style_layers 的 gram 矩阵(风格)和 content_layers 的内容:
In [ ]:
extractor = StyleContentModel(style_layers, content_layers)
results = extractor(tf.constant(content_image))
style_results = results['style']
print('Styles:')
for name, output in sorted(results['style'].items()):
print(" ", name)
print(" shape: ", output.numpy().shape)
print(" min: ", output.numpy().min())
print(" max: ", output.numpy().max())
print(" mean: ", output.numpy().mean())
print()
print("Contents:")
for name, output in sorted(results['content'].items()):
print(" ", name)
print(" shape: ", output.numpy().shape)
print(" min: ", output.numpy().min())
print(" max: ", output.numpy().max())
print(" mean: ", output.numpy().mean())
In [ ]:
style_targets = extractor(style_image)['style']
content_targets = extractor(content_image)['content']
定义一个 tf.Variable
来表示要优化的图像。 为了快速实现这一点,使用内容图像对其进行初始化( tf.Variable
必须与内容图像的形状相同)
In [ ]:
image = tf.Variable(content_image)
由于这是一个浮点图像,因此我们定义一个函数来保持像素值在 0 和 1 之间:
In [ ]:
def clip_0_1(image):
return tf.clip_by_value(image, clip_value_min=0.0, clip_value_max=1.0)
创建一个 optimizer 。 本教程推荐 LBFGS,但 Adam
也可以正常工作:
In [ ]:
opt = tf.optimizers.Adam(learning_rate=0.02, beta_1=0.99, epsilon=1e-1)
为了优化它,我们使用两个损失的加权组合来获得总损失:
In [ ]:
style_weight=1e-2
content_weight=1e4
In [ ]:
def style_content_loss(outputs):
style_outputs = outputs['style']
content_outputs = outputs['content']
style_loss = tf.add_n([tf.reduce_mean((style_outputs[name]-style_targets[name])**2)
for name in style_outputs.keys()])
style_loss *= style_weight / num_style_layers
content_loss = tf.add_n([tf.reduce_mean((content_outputs[name]-content_targets[name])**2)
for name in content_outputs.keys()])
content_loss *= content_weight / num_content_layers
loss = style_loss + content_loss
return loss
使用 tf.GradientTape
来更新图像。
In [ ]:
@tf.function()
def train_step(image):
with tf.GradientTape() as tape:
outputs = extractor(image)
loss = style_content_loss(outputs)
grad = tape.gradient(loss, image)
opt.apply_gradients([(grad, image)])
image.assign(clip_0_1(image))
现在,我们运行几个步来测试一下:
In [ ]:
train_step(image)
train_step(image)
train_step(image)
tensor_to_image(image)
运行正常,我们来执行一个更长的优化:
In [ ]:
import time
start = time.time()
epochs = 10
steps_per_epoch = 100
step = 0
for n in range(epochs):
for m in range(steps_per_epoch):
step += 1
train_step(image)
print(".", end='')
display.clear_output(wait=True)
display.display(tensor_to_image(image))
print("Train step: {}".format(step))
end = time.time()
print("Total time: {:.1f}".format(end-start))
In [ ]:
def high_pass_x_y(image):
x_var = image[:,:,1:,:] - image[:,:,:-1,:]
y_var = image[:,1:,:,:] - image[:,:-1,:,:]
return x_var, y_var
In [ ]:
x_deltas, y_deltas = high_pass_x_y(content_image)
plt.figure(figsize=(14,10))
plt.subplot(2,2,1)
imshow(clip_0_1(2*y_deltas+0.5), "Horizontal Deltas: Original")
plt.subplot(2,2,2)
imshow(clip_0_1(2*x_deltas+0.5), "Vertical Deltas: Original")
x_deltas, y_deltas = high_pass_x_y(image)
plt.subplot(2,2,3)
imshow(clip_0_1(2*y_deltas+0.5), "Horizontal Deltas: Styled")
plt.subplot(2,2,4)
imshow(clip_0_1(2*x_deltas+0.5), "Vertical Deltas: Styled")
这显示了高频分量如何增加。
而且,本质上高频分量是一个边缘检测器。 我们可以从 Sobel 边缘检测器获得类似的输出,例如:
In [ ]:
plt.figure(figsize=(14,10))
sobel = tf.image.sobel_edges(content_image)
plt.subplot(1,2,1)
imshow(clip_0_1(sobel[...,0]/4+0.5), "Horizontal Sobel-edges")
plt.subplot(1,2,2)
imshow(clip_0_1(sobel[...,1]/4+0.5), "Vertical Sobel-edges")
与此相关的正则化损失是这些值的平方和:
In [ ]:
def total_variation_loss(image):
x_deltas, y_deltas = high_pass_x_y(image)
return tf.reduce_sum(tf.abs(x_deltas)) + tf.reduce_sum(tf.abs(y_deltas))
In [ ]:
total_variation_loss(image).numpy()
以上说明了总变分损失的用途。但是无需自己实现,因为 TensorFlow 包含了一个标准实现:
In [ ]:
tf.image.total_variation(image).numpy()
In [ ]:
total_variation_weight=30
现在,将它加入 train_step
函数中:
In [ ]:
@tf.function()
def train_step(image):
with tf.GradientTape() as tape:
outputs = extractor(image)
loss = style_content_loss(outputs)
loss += total_variation_weight*tf.image.total_variation(image)
grad = tape.gradient(loss, image)
opt.apply_gradients([(grad, image)])
image.assign(clip_0_1(image))
重新初始化优化的变量:
In [ ]:
image = tf.Variable(content_image)
并进行优化:
In [ ]:
import time
start = time.time()
epochs = 10
steps_per_epoch = 100
step = 0
for n in range(epochs):
for m in range(steps_per_epoch):
step += 1
train_step(image)
print(".", end='')
display.clear_output(wait=True)
display.display(tensor_to_image(image))
print("Train step: {}".format(step))
end = time.time()
print("Total time: {:.1f}".format(end-start))
最后,保存结果:
In [ ]:
file_name = 'stylized-image.png'
tensor_to_image(image).save(file_name)
try:
from google.colab import files
except ImportError:
pass
else:
files.download(file_name)