试图修复神经网络的构造（错误信息：负维度？）

Question

这是一个 model，基于论文https://arxiv.org/pdf/1609.04836.Z437175BA4191210EE004ZE1D93的第 12 页 B.3 节的描述

from keras.models import Sequential
from keras.layers.core import Dense, Dropout, Activation, Flatten
from keras.layers.convolutional import Convolution2D, MaxPooling2D, AveragePooling2D
from keras.layers.normalization import BatchNormalization
img_size = (3, 32, 32)

...

def shallownet(nb_classes):
    global img_size
    model = Sequential()
    model.add(Conv2D(64, (5, 5), input_shape=img_size, data_format='channels_first'))
    model.add(BatchNormalization(axis=1))
    model.add(Activation('relu'))
    model.add(MaxPooling2D(pool_size=(3,3), strides=(2,2), padding='same'))

    model.add(Conv2D(64, (5, 5), padding='same', data_format='channels_first'))
    model.add(BatchNormalization(axis=1))
    model.add(Activation('relu'))
    model.add(MaxPooling2D(pool_size=(3,3), strides=(2,2), padding='same'))

    model.add(Flatten())
    model.add(Dense(384))
    model.add(BatchNormalization())
    model.add(Activation('relu'))
    model.add(Dropout(0.5))
    model.add(Dense(192))
    model.add(BatchNormalization())
    model.add(Activation('relu'))
    model.add(Dropout(0.5))
    model.add(Dense(nb_classes, activation='softmax'))
    return model

由于 Keras 2.0，我尝试对其进行转换。 这是原始代码： https://github.com/keskarnitish/large-batch-training/blob/master/network_zoo.py

我尝试运行以下代码：

parser = argparse.ArgumentParser(description=
                '''This code first trains the user-specific network (C[1-4])
                using small-batch ADAM and large-batch ADAM, and then plots
                the parametric plot connecting the two minimizers
                illustrating the sharpness difference.''')
parser.add_argument('-n', '--network', help='''Selects which network
                    to plot the parametric plots for.
                    Choices are C1, C2, C3 and C4.''', required=True)
network_choice = vars(parser.parse_args())['network']


nb_epoch = 20

# the data, shuffled and split between train and test sets
if network_choice in ['C1', 'C2']:
    (X_train, y_train), (X_test, y_test) = cifar10.load_data()
    nb_classes = 10
elif network_choice in ['C3', 'C4']:
    (X_train, y_train), (X_test, y_test) = cifar100.load_data()
    nb_classes = 100
else:
    raise ValueError('''Invalid choice of network.
                     Please choose one of C1, C2, C3 or C4.
                     Refer to the paper for details regarding these networks''')

X_train = X_train.astype('float32')
X_test = X_test.astype('float32')
X_train /= 255
X_test /= 255

# convert class vectors to binary class matrices
Y_train = np_utils.to_categorical(y_train, nb_classes)
Y_test = np_utils.to_categorical(y_test, nb_classes)

# build the network
if network_choice in ['C1', 'C3']:
    model = network_zoo.shallownet(nb_classes)
elif network_choice in ['C2', 'C4']:
    model = network_zoo.deepnet(nb_classes)
# let's train the model using Adam
model.compile(loss='categorical_crossentropy',
              optimizer='adam',
              metrics=['accuracy'])
model.save_weights('x0.h5')

# let's first find the small-batch solution
model.fit(X_train, Y_train,
          batch_size=256,
          nb_epoch=nb_epoch,
          validation_data=(X_test, Y_test),
          shuffle=True)

并收到以下错误消息：

Traceback (most recent call last):
  File "plot_parametric_plot.py", line 64, in <module>
    model = network_zoo.shallownet(nb_classes)
  File "/mnt_home/klee/LBSBGenGapSharpnessResearch/network_zoo.py", line 36, in shallownet
    model.add(Conv2D(64, (5, 5), input_shape=img_size))
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/keras/engine/sequential.py", line 166, in add
    layer(x)
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/keras/backend/tensorflow_backend.py", line 75, in symbolic_fn_wrapper
    return func(*args, **kwargs)
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/keras/engine/base_layer.py", line 489, in __call__
    output = self.call(inputs, **kwargs)
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/keras/layers/convolutional.py", line 171, in call
    dilation_rate=self.dilation_rate)
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/keras/backend/tensorflow_backend.py", line 3717, in conv2d
    **kwargs)
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/tensorflow_core/python/ops/nn_ops.py", line 918, in convolution_v2
    name=name)
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/tensorflow_core/python/ops/nn_ops.py", line 1010, in convolution_internal
    name=name)
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/tensorflow_core/python/ops/gen_nn_ops.py", line 969, in conv2d
    data_format=data_format, dilations=dilations, name=name)
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/tensorflow_core/python/framework/op_def_library.py", line 742, in _apply_op_helper
    attrs=attr_protos, op_def=op_def)
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/tensorflow_core/python/framework/func_graph.py", line 595, in _create_op_internal
    compute_device)
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/tensorflow_core/python/framework/ops.py", line 3322, in _create_op_internal
    op_def=op_def)
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/tensorflow_core/python/framework/ops.py", line 1786, in __init__
    control_input_ops)
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/tensorflow_core/python/framework/ops.py", line 1622, in _create_c_op
    raise ValueError(str(e))
ValueError: Negative dimension size caused by subtracting 5 from 3 for 'conv2d_1/convolution' (op: 'Conv2D') with input shapes: [?,3,32,32], [5,5,32,64].

虽然我不确定这意味着什么……

更新：在写完 Marco Celani 的评论后，这是我在运行主代码后收到的新错误消息：

Traceback (most recent call last):
  File "plot_parametric_plot.py", line 64, in <module>
    model = network_zoo.shallownet(nb_classes)
  File "/mnt_home/klee/LBSBGenGapSharpnessResearch/network_zoo.py", line 36, in shallownet
    model.add(Conv2D(64, (5, 5), input_shape=img_size, data_format='channels_first'))
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/keras/engine/sequential.py", line 166, in add
    layer(x)
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/keras/backend/tensorflow_backend.py", line 75, in symbolic_fn_wrapper
    return func(*args, **kwargs)
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/keras/engine/base_layer.py", line 489, in __call__
    output = self.call(inputs, **kwargs)
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/keras/layers/convolutional.py", line 171, in call
    dilation_rate=self.dilation_rate)
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/keras/backend/tensorflow_backend.py", line 3701, in conv2d
    x, tf_data_format = _preprocess_conv2d_input(x, data_format)
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/keras/backend/tensorflow_backend.py", line 3574, in _preprocess_conv2d_input
    if not _has_nchw_support() or force_transpose:
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/keras/backend/tensorflow_backend.py", line 522, in _has_nchw_support
    gpus_available = len(_get_available_gpus()) > 0
  File "/home/klee/anaconda3/envs/sharpenv/lib/python3.7/site-packages/keras/backend/tensorflow_backend.py", line 506, in _get_available_gpus
    _LOCAL_DEVICES = tf.config.experimental_list_devices()
AttributeError: module 'tensorflow_core._api.v2.config' has no attribute 'experimental_list_devices'

Answer 1

这意味着您不能应用任何操作，因为您减少了网络内部的太多维度（它低于 0）。

查看您的数据格式似乎您的图像是 (3, 32, 32)，因此通道是第一维。 Keras 默认情况下对最后一维的通道应用卷积。 要覆盖错误，请尝试在卷积层和池化层中定义data_format='channels_first'