无效参数：您必须为占位符张量“conv2d_transpose_1_target”提供一个值，其 dtype 为 float 和 shape [?,?,?,?]答案

【问题标题】：Invalid argument: You must feed a value for placeholder tensor 'conv2d_transpose_1_target' with dtype float and shape [?,?,?,?]无效参数：您必须为占位符张量“conv2d_transpose_1_target”提供一个值，其 dtype 为 float 和 shape [?,?,?,?]
【发布时间】：2023-03-14 14:51:01
【问题描述】：

我正在尝试在 Google Colab 中训练自动编码器。使用 ImageDataGenerator。我用这个运行了这段代码

from keras.preprocessing.image import ImageDataGenerator

batch_size = 128

train_datagen = ImageDataGenerator(rescale=1./255, validation_split = 0.2)

training_generator = train_datagen.flow_from_directory(train_dir,
                                                     target_size=(105, 105),
                                                     color_mode='grayscale',
                                                     batch_size = batch_size,
                                                     class_mode=None,
                                                     subset='training')
validation_generator = train_datagen.flow_from_directory(train_dir,
                                                     target_size=(105, 105),
                                                     color_mode='grayscale',
                                                     batch_size = batch_size,
                                                     class_mode=None,
                                                     subset='validation')

history = autoencoder.fit_generator(generator=training_generator, 
                                    epochs=5, 
                                    steps_per_epoch=training_generator.samples // batch_size,
                                    validation_data=validation_generator,
                                    validation_steps = validation_generator.samples // batch_size,
                                    use_multiprocessing=False)

它一直运行到它抛出此错误的第一个 epoch 步骤：

/usr/local/lib/python3.6/dist-packages/keras_preprocessing/image/utils.py:173: UserWarning: Using ".tiff" files with multiple bands will cause distortion. Please verify your output.
  warnings.warn('Using ".tiff" files with multiple bands '
Found 1375004 images belonging to 1 classes.
Found 343750 images belonging to 1 classes.
Epoch 1/5
10741/10742 [============================>.] - ETA: 0s - loss: 0.0052Epoch 1/5
---------------------------------------------------------------------------
InvalidArgumentError                      Traceback (most recent call last)
<ipython-input-25-c39bd284b251> in <module>()
     23                                     validation_data=validation_generator,
     24                                     validation_steps = validation_generator.samples // batch_size,
---> 25                                     use_multiprocessing=False)

5 frames
/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/client/session.py in __call__(self, *args, **kwargs)
   1470         ret = tf_session.TF_SessionRunCallable(self._session._session,
   1471                                                self._handle, args,
-> 1472                                                run_metadata_ptr)
   1473         if run_metadata:
   1474           proto_data = tf_session.TF_GetBuffer(run_metadata_ptr)

InvalidArgumentError: 2 root error(s) found.
  (0) Invalid argument: You must feed a value for placeholder tensor 'conv2d_transpose_1_target' with dtype float and shape [?,?,?,?]
     [[{{node conv2d_transpose_1_target}}]]
     [[loss/mul/_83]]
  (1) Invalid argument: You must feed a value for placeholder tensor 'conv2d_transpose_1_target' with dtype float and shape [?,?,?,?]
     [[{{node conv2d_transpose_1_target}}]]
0 successful operations.
0 derived errors ignored.

我使用了K.clear_session()，但它不起作用。而且我仍然不知道我做错了什么，因为有一个thread 已经解释了这个问题。我是 keras 的新手，任何帮助和建议都会是一种祝福！

这是自动编码器摘要和代码：

def create_model():
  model = Sequential()
  model.add(Conv2D(64, kernel_size=(11,11), strides=2, activation='relu', padding='valid', input_shape=(105,105,1)))
  model.add(MaxPooling2D(pool_size=(2,2), strides=2, padding='valid'))
  model.add(Conv2D(128, kernel_size=(1,1), strides=1, activation='relu', padding='valid'))
  model.add(Conv2DTranspose(64, kernel_size=(1,1), strides=1, activation='relu', padding='valid'))
  model.add(UpSampling2D(size=(2,2)))
  model.add(Conv2DTranspose(1, kernel_size=(11,11), strides=2, activation='relu', padding='valid'))

  adam = Adam(lr=0.01)
  model.compile(optimizer=adam , loss='mean_squared_error')
  return model

with tpu_strategy.scope(): # creating the model in the TPUStrategy scope means we will train the model on the TPU
  autoencoder = create_model()
autoencoder.summary()

Model: "sequential_2"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_4 (Conv2D)            (None, 48, 48, 64)        7808      
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 24, 24, 64)        0         
_________________________________________________________________
conv2d_5 (Conv2D)            (None, 24, 24, 128)       8320      
_________________________________________________________________
conv2d_transpose_4 (Conv2DTr (None, 24, 24, 64)        8256      
_________________________________________________________________
up_sampling2d_2 (UpSampling2 (None, 48, 48, 64)        0         
_________________________________________________________________
conv2d_transpose_5 (Conv2DTr (None, 105, 105, 1)       7745      
=================================================================
Total params: 32,129
Trainable params: 32,129
Non-trainable params: 0

【问题讨论】：

你能用模型架构更新问题吗？如果可能，您可以分享 Google Colab 的 Gist ==> 文件 -> 另存为 Github Gist。谢谢！

标签： python python-3.x tensorflow deep-learning

【解决方案1】：

我找到了答案，

更改 class_mode = 'input' 有效

还有，

如果你想使用class_mode = None，你可以通过添加这个函数来修复你的生成器

def fixed_generator(generator):
    for batch in generator:
        yield (batch, batch)

这是reference。

【讨论】：