Не удается создать экземпляр модели Keras, когда используется batch_normalization

Question

Я не уверен, что я делаю неправильно, но я следую коду из книги, чтобы создать модель GAN, и во время реализации оболочка Python просто зависает. Код на самом деле является подмножеством некоторого кода из книги, но код книги также не может создать модель.

Если я закомментирую batch_norm, я могу создать экземпляр модель.

Здесь:

https://github.com/PacktPublishing/Advanced-Deep-Learning-with-Keras/blob/master/chapter4-gan/dcgan-mnist-4.2.1.py

Docs: https://keras.io/layers/normalization/

from keras.layers import Activation, Dense, Input
from keras.layers import Conv2D, Flatten
from keras.layers import Reshape, Conv2DTranspose
from keras.layers import LeakyReLU
from keras.layers import BatchNormalization
from keras.optimizers import RMSprop
from keras.models import Model
from keras.datasets import mnist
from keras.models import load_model
import keras

import numpy as np
import math
import matplotlib.pyplot as plt
import os
import argparse




def generator_model(inputs, image_size, verbose = True):
    """Generator Model

    args
    =======
    inputs = input layer
    image_size = size of image dimension (299? 480? 28?etc)

    """

    #resized dependent on how many Conv2d Transpore

    print("build generator model")

    image_resize = image_size // 4 
    kernel_size = 5
    layer_filters = [128, 64] #first two convs
    final_layer_filters = [32, 1] # last two conbs

    x= inputs
    x = Dense(image_resize * image_resize * layer_filters[0])(x)
    x = Reshape((image_resize, image_resize, layer_filters[0]))(x)
    print(x)

    for filter_ in layer_filters:
        x = BatchNormalization()(x)
        x = Activation('relu')(x)
        x = Conv2DTranspose(filters=filter_,
                            kernel_size=kernel_size,
                            strides=2,
                            padding='same')(x)


    print("built first part")
    for filter_ in final_layer_filters:
        x = BatchNormalization()(x)
        x = Activation('relu')(x)
        x = Conv2DTranspose(filters=filter_,
                            kernel_size=kernel_size,
                            strides=1,
                            padding='same')(x)

    x = Activation('sigmoid')(x)
    print("finised building")
    generator = Model(inputs, x, name='generator')
    if verbose:
        print(generator.summary())
    return generator






print(keras.__version__) #2.24
z_size = 100
img_size = 28
gen_input =  Input(shape= (z_size,), name='gen_input')
generator = generator_model(gen_input, img_size)

Оболочка выводит следующее, и пока она работает, она не завершает выполнение сценария, она просто остановлена:

2.2.4
build generator model
Tensor("reshape_1/Reshape:0", shape=(?, 7, 7, 128), dtype=float32)

Answer 1

Я попробовал ваш код в Google Colab. Следующее генерируется. Я думаю, что это не проблема кода. Вы можете проверить другую проблему, например, установка.

    Using TensorFlow backend.
    2.2.4
    build generator model
    WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/op_def_library.py:263: colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.
    Instructions for updating:
    Colocations handled automatically by placer.
    Tensor("reshape_1/Reshape:0", shape=(?, 7, 7, 128), dtype=float32)
    built first part
    finised building
    _________________________________________________________________
    Layer (type)                 Output Shape              Param #
    =================================================================
    gen_input (InputLayer)       (None, 100)               0
    _________________________________________________________________
    dense_1 (Dense)              (None, 6272)              633472
    _________________________________________________________________
    reshape_1 (Reshape)          (None, 7, 7, 128)         0
    _________________________________________________________________
    batch_normalization_1 (Batch (None, 7, 7, 128)         512
    _________________________________________________________________
    activation_1 (Activation)    (None, 7, 7, 128)         0
    _________________________________________________________________
    conv2d_transpose_1 (Conv2DTr (None, 14, 14, 128)       409728
    _________________________________________________________________
    batch_normalization_2 (Batch (None, 14, 14, 128)       512
    _________________________________________________________________
    activation_2 (Activation)    (None, 14, 14, 128)       0
    _________________________________________________________________
    conv2d_transpose_2 (Conv2DTr (None, 28, 28, 64)        204864
    _________________________________________________________________
    batch_normalization_3 (Batch (None, 28, 28, 64)        256
    _________________________________________________________________
    activation_3 (Activation)    (None, 28, 28, 64)        0
    _________________________________________________________________
    conv2d_transpose_3 (Conv2DTr (None, 28, 28, 32)        51232
    _________________________________________________________________
    batch_normalization_4 (Batch (None, 28, 28, 32)        128
    _________________________________________________________________
    activation_4 (Activation)    (None, 28, 28, 32)        0
    _________________________________________________________________
    conv2d_transpose_4 (Conv2DTr (None, 28, 28, 1)         801
    _________________________________________________________________
    activation_5 (Activation)    (None, 28, 28, 1)         0
    =================================================================
    Total params: 1,301,505
    Trainable params: 1,300,801
    Non-trainable params: 704
    _________________________________________________________________
            None