基于Tensorflow2.0的Fashion-MNIST生成
图像来源:深度学习案例:用tensorflow2.0实现Fashion-MNIST数据集分类
一、生成预览
二、DCGAN简介
DCGAN,全称为Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks(http://arxiv.org/pdf/1511.06434),主要创新为将100维均匀分布的Z映射到小空间范围的卷积表示法中得到多个特征向量。通过一系列小数步长卷积four fractionally-strided (在最近的一些论文中,这些被错误地称为反卷积),将这个高维表示转换成一个64 x 64像素的图像。没有使用全连接层和池化层。
除输出层使用Tanh函数外,生成器还使用ReLU激活函数(Nair & Hinton, 2010)。我们观察到,使用有界激活使模型能够更快地学习,以饱和和覆盖训练分布的颜色空间。在鉴别器中,我们发现漏整流激活(Maas et al., 2013) (Xu et al., 2015)工作良好,特别是对于高分辨率建模。这与最初使用maxout激活的GAN论文形成了对比(Goodfellow et al. 2013)。
稳定的深卷积GANs架构指南
1.用strided convolutions (discriminator)和fractional-strided convolutions (generator)替换池化层。
2.在生成器和鉴别器中使用batchnorm。
3.为更深层的架构移除完全连接的隐藏层。
4.G除输出层使用Tanh外,其他层均使用ReLU激活。
5.在鉴别器中对所有层使用LeakyReLU激活
三、代码
- 构建生成器G
# tf2 生成时装
import tensorflow as tf
from tensorflow.keras import layers
def make_generator_model():
model = tf.keras.Sequential()
model.add(layers.Dense(7*7*256, use_bias=False, input_shape=(100,)))
model.add(layers.BatchNormalization())
model.add(layers.LeakyReLU())
model.add(layers.Reshape((7, 7, 256)))
assert model.output_shape == (None, 7, 7, 256) # Note: None is the batch size
model.add(layers.Conv2DTranspose(128, (5, 5), strides=(1, 1), padding='same', use_bias=False))
assert model.output_shape == (None, 7, 7, 128)
model.add(layers.BatchNormalization())
model.add(layers.LeakyReLU())
model.add(layers.Conv2DTranspose(64, (5, 5), strides=(2, 2), padding='same', use_bias=False))
assert model.output_shape == (None, 14, 14, 64)
model.add(layers.BatchNormalization())
model.add(layers.LeakyReLU())
model.add(layers.Conv2DTranspose(1, (5, 5), strides=(2, 2), padding='same', use_bias=False, activation='tanh'))
assert model.output_shape == (None, 28, 28, 1)
return model
import matplotlib.pyplot as plt
generator = make_generator_model()
noise = tf.random.normal([1, 100])
generated_image = generator(noise, training=False)
plt.imshow(generated_image[0, :, :, 0], cmap='gray')
- 构建判别器
def make_discriminator_model():
model = tf.keras.Sequential()
model.add(layers.Conv2D(64, (5, 5), strides=(2, 2), padding='same',
input_shape=[28, 28, 1]))
model.add(layers.LeakyReLU())
model.add(layers.Dropout(0.3))
model.add(layers.Conv2D(128, (5, 5), strides=(2, 2), padding='same'))
model.add(layers.LeakyReLU())
model.add(layers.Dropout(0.3))
model.add(layers.Flatten())
model.add(layers.Dense(1))
return model
discriminator = make_discriminator_model()
decision = discriminator(generated_image)
print(decision)
tf.Tensor([[0.00181567]], shape=(1, 1), dtype=float32)
- 构建损失函数
# 该方法返回计算交叉熵损失的辅助函数
cross_entropy = tf.keras.losses.BinaryCrossentropy(from_logits=True)
# 该方法量化判别器从判断真伪图片的能力。它将判别器对真实图片的预测值与值全为 1 的数组进行对比,将判别器对伪造(生成的)图片的预测值与值全为 0 的数组进行对比。
def discriminator_loss(real_output, fake_output):
real_loss = cross_entropy(tf.ones_like(real_output), real_output)
fake_loss = cross_entropy(tf.zeros_like(fake_output), fake_output)
total_loss = real_loss + fake_loss
return total_loss
# 生成器损失量化其欺骗判别器的能力。直观来讲,如果生成器表现良好,判别器将会把伪造图片判断为真实图片(或 1)。
def generator_loss(fake_output):
return cross_entropy(tf.ones_like(fake_output), fake_output)
# 需要分别训练两个网络,判别器和生成器的优化器是不同的。
generator_optimizer = tf.keras.optimizers.Adam(1e-4)
discriminator_optimizer = tf.keras.optimizers.Adam(1e-4)
# 注意 `tf.function` 的使用,该注解使函数被“编译”
noise_dim = 100
@tf.function
def train_step(images):
noise = tf.random.normal([BATCH_SIZE, noise_dim])
with tf.GradientTape() as gen_tape, tf.GradientTape() as disc_tape:
generated_images = generator(noise, training=True)
real_output = discriminator(images, training=True)
fake_output = discriminator(generated_images, training=True)
gen_loss = generator_loss(fake_output)
disc_loss = discriminator_loss(real_output, fake_output)
gradients_of_generator = gen_tape.gradient(gen_loss, generator.trainable_variables)
gradients_of_discriminator = disc_tape.gradient(disc_loss, discriminator.trainable_variables)
generator_optimizer.apply_gradients(zip(gradients_of_generator, generator.trainable_variables))
discriminator_optimizer.apply_gradients(zip(gradients_of_discriminator, discriminator.trainable_variables))
return gen_loss, disc_loss
- 准备Fashion-MNIST数据
# 准备数据
(train_images, train_labels), (_, _) = tf.keras.datasets.fashion_mnist.load_data()
train_images = train_images.reshape(train_images.shape[0], 28, 28, 1).astype('float32')
train_images = (train_images - 127.5) / 127.5 # Normalize the images to [-1, 1]
BUFFER_SIZE = 60000
BATCH_SIZE = 256
# Batch and shuffle the data
train_dataset = tf.data.Dataset.from_tensor_slices(train_images).shuffle(BUFFER_SIZE).batch(BATCH_SIZE)
def generate_and_save_images(model, epoch, test_input):
# 注意 training` 设定为 False
# 因此,所有层都在推理模式下运行(batchnorm)
predictions = model(test_input, training=False)
fig = plt.figure(figsize=(4,4))
for i in range(predictions.shape[0]):
plt.subplot(4, 4, i+1)
plt.imshow(predictions[i, :, :, 0] * 127.5 + 127.5, cmap='gray')
plt.axis('off')
plt.savefig('./output/fashion-mnist/image_at_epoch_{:04d}.png'.format(epoch))
plt.show()
- Training
noise_dim = 100
num_examples_to_generate = 16
seed = tf.random.normal([num_examples_to_generate, noise_dim])
def train(dataset, epochs):
for epoch in range(epochs):
for i,image_batch in enumerate(dataset):
g,d = train_step(image_batch)
print("batch %d, gen_loss %f,disc_loss %f" % (i, g.numpy(),d.numpy()))
# 每个 epoch 生成一次图片
generate_and_save_images(generator, epoch, seed)
# 保存模型
generator.save('./save/dcgan_fashion-mnist_tf2.h5')
EPOCHS = 50
train(train_dataset, EPOCHS)
- Test
# 加载模型
import tensorflow as tf
import matplotlib.pyplot as plt
model = tf.keras.models.load_model('./save/dcgan_fashion-mnist_tf2.h5')
def generate_and_save_images(model, epoch, test_input):
# 注意 training` 设定为 False
# 因此,所有层都在推理模式下运行(batchnorm)
predictions = model(test_input, training=False)
fig = plt.figure(figsize=(4,4))
for i in range(predictions.shape[0]):
plt.subplot(4, 4, i+1)
plt.imshow(predictions[i, :, :, 0] * 127.5 + 127.5, cmap='gray')
plt.axis('off')
plt.show()
test_input = tf.random.normal([16, 100])
epoch = 10
generate_and_save_images(model, epoch, test_input)
四、注意事项
- GPU训练更佳:10min,CPU训练速度:约150min
- tensorflow2版本
了解更多,关注公众号:AI预见未来