image_utilsのimportがうまくいきません

前提・実現したいこと

「直感DeepLearning」という本に記載されているサンプルコード
を実行した際に発生したエラーを解消させたいです．

https://github.com/oreilly-japan/deep-learning-with-keras-ja/tree/master/ch04
の中の example_gan_convolutional.py
というファイルです．

発生している問題・エラーメッセージ

ImportError: cannot import name 'dim_ordering_fix' from 'image_utils' (\Users(ユーザー名)\AppData\Local\Programs\Python\Python37\Lib\site-packages\image_utils\__init__.py)

該当のソースコード

Python3
1import os
2import pandas as pd
3import numpy as np
4import matplotlib as mpl
5import keras.backend as K
6from keras.layers import Flatten, Dropout, LeakyReLU, Input, Activation, Dense, BatchNormalization
7from keras.layers.convolutional import UpSampling2D, Conv2D
8from keras.models import Model
9from keras.optimizers import Adam
10from keras.callbacks import TensorBoard
11from keras.datasets import mnist
12from keras_adversarial.image_grid_callback import ImageGridCallback
13from keras_adversarial import AdversarialModel, simple_gan, gan_targets
14from keras_adversarial import AdversarialOptimizerSimultaneous, normal_latent_sampling
15from image_utils import dim_ordering_fix, dim_ordering_input, dim_ordering_reshape, dim_ordering_unfix
16
17
18# This line allows mpl to run with no DISPLAY defined
19mpl.use("Agg")
20
21
22def model_generator():
23    nch = 256
24    g_input = Input(shape=[100])
25    H = Dense(nch * 14 * 14)(g_input)
26    H = BatchNormalization()(H)
27    H = Activation("relu")(H)
28    H = dim_ordering_reshape(nch, 14)(H)
29    H = UpSampling2D(size=(2, 2))(H)
30    H = Conv2D(int(nch / 2), (3, 3), padding="same")(H)
31    H = BatchNormalization()(H)
32    H = Activation("relu")(H)
33    H = Conv2D(int(nch / 4), (3, 3), padding="same")(H)
34    H = BatchNormalization()(H)
35    H = Activation("relu")(H)
36    H = Conv2D(1, (1, 1), padding="same")(H)
37    g_V = Activation("sigmoid")(H)
38    return Model(g_input, g_V)
39
40
41def model_discriminator(input_shape=(1, 28, 28), dropout_rate=0.5):
42    d_input = dim_ordering_input(input_shape, name="input_x")
43    nch = 512
44    # nch = 128
45    H = Conv2D(int(nch / 2), (5, 5),
46               strides=(2, 2),
47               padding="same",
48               activation="relu",
49               )(d_input)
50    H = LeakyReLU(0.2)(H)
51    H = Dropout(dropout_rate)(H)
52    H = Conv2D(nch, (5, 5),
53               strides=(2, 2),
54               padding="same",
55               activation="relu",
56               )(H)
57    H = LeakyReLU(0.2)(H)
58    H = Dropout(dropout_rate)(H)
59    H = Flatten()(H)
60    H = Dense(int(nch / 2))(H)
61    H = LeakyReLU(0.2)(H)
62    H = Dropout(dropout_rate)(H)
63    d_V = Dense(1, activation="sigmoid")(H)
64    return Model(d_input, d_V)
65
66
67def mnist_process(x):
68    x = x.astype(np.float32) / 255.0
69    return x
70
71
72def mnist_data():
73    (xtrain, ytrain), (xtest, ytest) = mnist.load_data()
74    return mnist_process(xtrain), mnist_process(xtest)
75
76
77def generator_sampler(latent_dim, generator):
78    def fun():
79        zsamples = np.random.normal(size=(10 * 10, latent_dim))
80        gen = dim_ordering_unfix(generator.predict(zsamples))
81        return gen.reshape((10, 10, 28, 28))
82
83    return fun
84
85
86if __name__ == "__main__":
87    # z \in R^100
88    latent_dim = 100
89    # x \in R^{28x28}
90    input_shape = (1, 28, 28)
91
92    # generator (z -> x)
93    generator = model_generator()
94    # discriminator (x -> y)
95    discriminator = model_discriminator(input_shape=input_shape)
96    # gan (x - > yfake, yreal), z generated on GPU
97    gan = simple_gan(generator, discriminator,
98                     normal_latent_sampling((latent_dim,)))
99
100    # print summary of models
101    generator.summary()
102    discriminator.summary()
103    gan.summary()
104
105    # build adversarial model
106    model = AdversarialModel(base_model=gan,
107                             player_params=[generator.trainable_weights, 
108                                            discriminator.trainable_weights],
109                             player_names=["generator", "discriminator"])
110    model.adversarial_compile(adversarial_optimizer=AdversarialOptimizerSimultaneous(),
111                              player_optimizers=[Adam(1e-4, decay=1e-4),
112                                                 Adam(1e-3, decay=1e-4)],
113                              loss="binary_crossentropy")
114
115    # train model
116    generator_cb = ImageGridCallback("output/gan_convolutional/epoch-{:03d}.png",
117                                     generator_sampler(latent_dim, generator))
118    callbacks = [generator_cb]
119    if K.backend() == "tensorflow":
120        callbacks.append(
121            TensorBoard(log_dir=os.path.join("output/gan_convolutional/", "logs/"),
122                        histogram_freq=0, write_graph=True, write_images=True))
123
124    xtrain, xtest = mnist_data()
125    xtrain = dim_ordering_fix(xtrain.reshape((-1, 1, 28, 28)))
126    xtest = dim_ordering_fix(xtest.reshape((-1, 1, 28, 28)))
127    y = gan_targets(xtrain.shape[0])
128    ytest = gan_targets(xtest.shape[0])
129    history = model.fit(x=xtrain, y=y, validation_data=(xtest, ytest), 
130                        callbacks=[generator_cb], epochs=100,
131                        batch_size=32)
132    df = pd.DataFrame(history.history)
133    df.to_csv("output/gan_convolutional/history.csv")
134
135    generator.save("output/gan_convolutional/generator.h5")
136    discriminator.save("output/gan_convolutional/discriminator.h5")
137