官网实例详解4.41（variational_autoencoder_deconv.py）-keras学习笔记四

使用Keras和反卷积层建立变分自编码器演示脚本

Keras实例目录

代码注释

 

'''This script demonstrates how to build a variational autoencoder
with Keras and deconvolution layers.
使用Keras和反卷积层建立变分自编码器演示脚本
# Reference

- Auto-Encoding Variational Bayes
  自动编码变分贝叶斯
  https://arxiv.org/abs/1312.6114
'''
from __future__ import print_function

import numpy as np
import matplotlib.pyplot as plt
from scipy.stats import norm

from keras.layers import Input, Dense, Lambda, Flatten, Reshape
from keras.layers import Conv2D, Conv2DTranspose
from keras.models import Model
from keras import backend as K
from keras import metrics
from keras.datasets import mnist

# input image dimensions
# 输入图像维度
img_rows, img_cols, img_chns = 28, 28, 1
# number of convolutional filters to use
# 使用的卷积过滤器数量
filters = 64
# convolution kernel size
# 卷积核大小
num_conv = 3

batch_size = 100
if K.image_data_format() == 'channels_first':
    original_img_size = (img_chns, img_rows, img_cols)
else:
    original_img_size = (img_rows, img_cols, img_chns)
latent_dim = 2
intermediate_dim = 128
epsilon_std = 1.0
epochs = 5

x = Input(shape=original_img_size)
conv_1 = Conv2D(img_chns,
                kernel_size=(2, 2),
                padding='same', activation='relu')(x)
conv_2 = Conv2D(filters,
                kernel_size=(2, 2),
                padding='same', activation='relu',
                strides=(2, 2))(conv_1)
conv_3 = Conv2D(filters,
                kernel_size=num_conv,
                padding='same', activation='relu',
                strides=1)(conv_2)
conv_4 = Conv2D(filters,
                kernel_size=num_conv,
                padding='same', activation='relu',
                strides=1)(conv_3)
flat = Flatten()(conv_4)
hidden = Dense(intermediate_dim, activation='relu')(flat)

z_mean = Dense(latent_dim)(hidden)
z_log_var = Dense(latent_dim)(hidden)


def sampling(args):
    z_mean, z_log_var = args
    epsilon = K.random_normal(shape=(K.shape(z_mean)[0], latent_dim),
                              mean=0., stddev=epsilon_std)
    return z_mean + K.exp(z_log_var) * epsilon

# note that "output_shape" isn't necessary with the TensorFlow backend
# so you could write `Lambda(sampling)([z_mean, z_log_var])`
# 注意，“output_shape”对于TensorFlow后端不是必需的。因此可以编写Lambda(sampling)([z_mean, z_log_var])`
z = Lambda(sampling, output_shape=(latent_dim,))([z_mean, z_log_var])

# we instantiate these layers separately so as to reuse them later
# 分别实例化这些层，以便在以后重用它们。
decoder_hid = Dense(intermediate_dim, activation='relu')
decoder_upsample = Dense(filters * 14 * 14, activation='relu')

if K.image_data_format() == 'channels_first':
    output_shape = (batch_size, filters, 14, 14)
else:
    output_shape = (batch_size, 14, 14, filters)

decoder_reshape = Reshape(output_shape[1:])
decoder_deconv_1 = Conv2DTranspose(filters,
                                   kernel_size=num_conv,
                                   padding='same',
                                   strides=1,
                                   activation='relu')
decoder_deconv_2 = Conv2DTranspose(filters,
                                   kernel_size=num_conv,
                                   padding='same',
                                   strides=1,
                                   activation='relu')
if K.image_data_format() == 'channels_first':
    output_shape = (batch_size, filters, 29, 29)
else:
    output_shape = (batch_size, 29, 29, filters)
decoder_deconv_3_upsamp = Conv2DTranspose(filters,
                                          kernel_size=(3, 3),
                                          strides=(2, 2),
                                          padding='valid',
                                          activation='relu')
decoder_mean_squash = Conv2D(img_chns,
                             kernel_size=2,
                             padding='valid',
                             activation='sigmoid')

hid_decoded = decoder_hid(z)
up_decoded = decoder_upsample(hid_decoded)
reshape_decoded = decoder_reshape(up_decoded)
deconv_1_decoded = decoder_deconv_1(reshape_decoded)
deconv_2_decoded = decoder_deconv_2(deconv_1_decoded)
x_decoded_relu = decoder_deconv_3_upsamp(deconv_2_decoded)
x_decoded_mean_squash = decoder_mean_squash(x_decoded_relu)

# instantiate VAE model
# 实例化VAE模型
vae = Model(x, x_decoded_mean_squash)

# Compute VAE loss
# 计算VAE损失
xent_loss = img_rows * img_cols * metrics.binary_crossentropy(
    K.flatten(x),
    K.flatten(x_decoded_mean_squash))
kl_loss = - 0.5 * K.sum(1 + z_log_var - K.square(z_mean) - K.exp(z_log_var), axis=-1)
vae_loss = K.mean(xent_loss + kl_loss)
vae.add_loss(vae_loss)

vae.compile(optimizer='rmsprop')
vae.summary()

# train the VAE on MNIST digits
# 基于MNIST数字训练VAE
(x_train, _), (x_test, y_test) = mnist.load_data()

x_train = x_train.astype('float32') / 255.
x_train = x_train.reshape((x_train.shape[0],) + original_img_size)
x_test = x_test.astype('float32') / 255.
x_test = x_test.reshape((x_test.shape[0],) + original_img_size)

print('x_train.shape:', x_train.shape)

vae.fit(x_train,
        shuffle=True,
        epochs=epochs,
        batch_size=batch_size,
        validation_data=(x_test, None))

# build a model to project inputs on the latent space
# 建立一个潜在空间输入模型
encoder = Model(x, z_mean)

# display a 2D plot of the digit classes in the latent space
# 在潜在空间中显示数字类的2D图
x_test_encoded = encoder.predict(x_test, batch_size=batch_size)
plt.figure(figsize=(6, 6))
plt.scatter(x_test_encoded[:, 0], x_test_encoded[:, 1], c=y_test)
plt.colorbar()
plt.show()

# build a digit generator that can sample from the learned distribution
# 建立一个数字生成器，可以从学习的分布中取样
decoder_input = Input(shape=(latent_dim,))
_hid_decoded = decoder_hid(decoder_input)
_up_decoded = decoder_upsample(_hid_decoded)
_reshape_decoded = decoder_reshape(_up_decoded)
_deconv_1_decoded = decoder_deconv_1(_reshape_decoded)
_deconv_2_decoded = decoder_deconv_2(_deconv_1_decoded)
_x_decoded_relu = decoder_deconv_3_upsamp(_deconv_2_decoded)
_x_decoded_mean_squash = decoder_mean_squash(_x_decoded_relu)
generator = Model(decoder_input, _x_decoded_mean_squash)

# display a 2D manifold of the digits
# 显示数字的二维形状
n = 15  # figure with 15x15 digits # 15X15数字图形
digit_size = 28
figure = np.zeros((digit_size * n, digit_size * n))
# linearly spaced coordinates on the unit square were transformed through the inverse CDF (ppf) of the Gaussian
# 单位平方的线性间隔坐标通过高斯的逆CDF（ppf）变换。
# to produce values of the latent variables z, since the prior of the latent space is Gaussian
# 产生潜在变量Z的值，因为潜在空间的先验是高斯
grid_x = norm.ppf(np.linspace(0.05, 0.95, n))
grid_y = norm.ppf(np.linspace(0.05, 0.95, n))

for i, yi in enumerate(grid_x):
    for j, xi in enumerate(grid_y):
        z_sample = np.array([[xi, yi]])
        z_sample = np.tile(z_sample, batch_size).reshape(batch_size, 2)
        x_decoded = generator.predict(z_sample, batch_size=batch_size)
        digit = x_decoded[0].reshape(digit_size, digit_size)
        figure[i * digit_size: (i + 1) * digit_size,
               j * digit_size: (j + 1) * digit_size] = digit
plt.figure(figsize=(10, 10))
plt.imshow(figure, cmap='Greys_r')
plt.show()

代码执行

 

Keras详细介绍

英文：https://keras.io/

中文：http://keras-cn.readthedocs.io/en/latest/

实例下载

https://github.com/keras-team/keras

https://github.com/keras-team/keras/tree/master/examples

完整项目下载

方便没积分童鞋，请加企鹅452205574，共享文件夹。

包括：代码、数据集合（图片）、已生成model、安装库文件等。