使用batch_normalization時無法實例化Keras模型

Question

我不確定自己在做什么錯，但是我正在按照本書中的代碼創建GAN模型，並且在實例化過程中，Python shell處於凍結狀態。 該代碼實際上是一本書中某些代碼的子集，但是該書中的代碼也無法創建模型。

如果我注釋掉batch_norm則可以實例化模型。

這里：

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

文件： 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)

Shell輸出以下內容，並且在仍在運行時，它還沒有完成腳本的運行，只是處於停頓狀態：

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

Answer 1

我在google colab中嘗試了您的代碼。 生成以下內容。 我認為這不是代碼問題。 您可以檢查其他問題，例如設置。

    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