检查输入时出错：期望flatten_input具有3维，但数组的形状为（无，100、100、1）

Question

我想使用TensorFlow / Keras将图片分为两类，自拍和非自拍。

我已经将样本收集到两个文件系统文件夹中，每个类别一个。

在使用从文件系统加载图片后，按照MNIST时尚的官方教程（这也是图片分类问题），我在下面进行了培训，如https://stackoverflow.com/a/52417770/226958所示 。

不幸的是，我得到一个错误：

1.10.1
Tensor("IteratorGetNext:0", shape=(?, 100, 100, 1), dtype=float32)
Tensor("IteratorGetNext:1", shape=(?,), dtype=int32)
Traceback (most recent call last):
  File "run.py", line 50, in <module>
    model.fit(images, labels, epochs=1, steps_per_epoch=60000)
  File "/home/nico/.local/lib/python2.7/site-packages/tensorflow/python/keras/engine/training.py", line 1278, in fit
    validation_split=validation_split)
  File "/home/nico/.local/lib/python2.7/site-packages/tensorflow/python/keras/engine/training.py", line 878, in _standardize_user_data
    exception_prefix='input')
  File "/home/nico/.local/lib/python2.7/site-packages/tensorflow/python/keras/engine/training_utils.py", line 182, in standardize_input_data
    'with shape ' + str(data_shape))
ValueError: Error when checking input: expected flatten_input to have 3 dimensions, but got array with shape (None, 100, 100, 1)

这是源代码：

import tensorflow as tf
print(tf.__version__)

out_shape = tf.convert_to_tensor([100, 100])
batch_size = 2

image_paths, labels = ["selfies-data/1", "selfies-data/2"], [1, 2]
epoch_size = len(image_paths)
image_paths = tf.convert_to_tensor(image_paths, dtype=tf.string)
labels = tf.convert_to_tensor(labels)

# The images loading part is from https://stackoverflow.com/a/52417770/226958
dataset = tf.data.Dataset.from_tensor_slices((image_paths, labels))
dataset = dataset.repeat().shuffle(epoch_size)

def map_fn(path, label):
    # path/label represent values for a single example
    image = tf.image.decode_jpeg(tf.read_file(path))

    # some mapping to constant size - be careful with distorting aspec ratios
    image = tf.image.resize_images(image, out_shape)
    image = tf.image.rgb_to_grayscale(image)
    # color normalization - just an example
    image = tf.to_float(image) * (2. / 255) - 1
    return image, label

# num_parallel_calls > 1 induces intra-batch shuffling
dataset = dataset.map(map_fn, num_parallel_calls=8)
dataset = dataset.batch(batch_size)
dataset = dataset.prefetch(1)

images, labels = dataset.make_one_shot_iterator().get_next()

# All of the following is from https://www.tensorflow.org/tutorials/keras/basic_classification
from tensorflow import keras

model = keras.Sequential([
    keras.layers.Flatten(input_shape=(100, 100)),
    keras.layers.Dense(128, activation=tf.nn.relu),
    keras.layers.Dense(10, activation=tf.nn.softmax)
])

model.compile(optimizer=tf.train.AdamOptimizer(),
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

print(images)
print(labels)
model.fit(images, labels, epochs=epoch_size, steps_per_epoch=60000)

尽管我已经阅读了类似的问题，但没有这个None问题。

如何使Keras适应我的输入，或转换我的输入以使Keras接受它？

Answer 1

1）图像具有一个通道，因此必须在输入shape参数中反映出来：

keras.layers.Flatten(input_shape=(100, 100, 1))

2）要使用tf.data API加载文件，您首先需要获取图片文件名及其对应的标签：

image_paths, lbls = ["selfies-data/1", "selfies-data/2"], [0., 1.]

labels = []
file_names = []
for d, l in zip(image_paths, lbls):
    # get the list all the images file names
    name = [os.path.join(d,f) for f in os.listdir(d)]
    file_names.extend(name)
    labels.extend([l] * len(name))

file_names = tf.convert_to_tensor(file_names, dtype=tf.string)
labels = tf.convert_to_tensor(labels)

dataset = tf.data.Dataset.from_tensor_slices((file_names, labels))

# the rest is the same 

您可能还需要扩展labels的尺寸，以使其具有(?, 1)的形状（而不是(?,) ）。 为此，您可以map_fn下行放在map_fn函数中：

labels = tf.expand_dims(labels, axis=-1)

3）如果您有两个班级，那么为什么最后一层有10个单元？ 这是一个二进制分类问题，因此使最后一层具有一个通过sigmoid激活的单元。 最后，将损失更改为binary_crossentropy ：

       # ... 
       keras.layers.Dense(1, activation=tf.nn.sigmoid)
])

model.compile(optimizer=tf.train.AdamOptimizer(),
              loss='binary_crossentropy',
              metrics=['accuracy'])