密集層的輸入 0 與該層不兼容：輸入形狀的預期軸 -1 具有值 8192，但接收到的輸入具有形狀（無，61608）

Question

我正在嘗試創建一個圖像處理 CNN。 我正在使用 VGG16 來加快一些學習過程。 下面創建我的 CNN 可以訓練和保存 model 和權重。 當我在加載 model 后嘗試運行預測 function 時會出現此問題。

image_gen = ImageDataGenerator()
train = image_gen.flow_from_directory('./data/train', class_mode='categorical', shuffle=False, batch_size=10, target_size=(151, 136))
val = image_gen.flow_from_directory('./data/validate', class_mode='categorical', shuffle=False, batch_size=10, target_size=(151, 136))

pretrained_model = VGG16(include_top=False, input_shape=(151, 136, 3), weights='imagenet')
pretrained_model.summary()

vgg_features_train = pretrained_model.predict(train)
vgg_features_val = pretrained_model.predict(val)

train_target = to_categorical(train.labels)
val_target = to_categorical(val.labels)

model = Sequential()
model.add(Flatten())
model.add(Dense(100, activation='relu'))
model.add(Dropout(0.5))
model.add(BatchNormalization())
model.add(Dense(2, activation='softmax'))

model.compile(optimizer='rmsprop', metrics=['accuracy'], loss='categorical_crossentropy')

target_dir = './models/weights-improvement'
if not os.path.exists(target_dir):
  os.mkdir(target_dir)

checkpoint = ModelCheckpoint(filepath=target_dir + 'weights-improvement-{epoch:02d}-{val_accuracy:.2f}.hdf5', monitor='val_accuracy', verbose=1, save_best_only=True, mode='max')
callbacks_list = [checkpoint]

model.fit(vgg_features_train, train_target, epochs=100, batch_size=8, validation_data=(vgg_features_val, val_target), callbacks=callbacks_list)

model.save('./models/model')
model.save_weights('./models/weights')

我有這個預測 function，我想加載到一個圖像中，然后返回 model 給出的這個圖像的分類。

from keras.preprocessing.image import load_img, img_to_array
def predict(file):
  x = load_img(file, target_size=(151,136,3))
  x = img_to_array(x)
  print(x.shape)
  print(x.shape)
  x = np.expand_dims(x, axis=0)
  array = model.predict(x)
  result = array[0]
  if result[0] > result[1]:
    if result[0] > 0.9:
      print("Predicted answer: Buy")
      answer = 'buy'
      print(result)
      print(array)
    else:
      print("Predicted answer: Not confident")
      answer = 'n/a'
      print(result)
  else:
    if result[1] > 0.9:
      print("Predicted answer: Sell")
      answer = 'sell'
      print(result)
    else:
      print("Predicted answer: Not confident")
      answer = 'n/a'
      print(result)

  return answer

我遇到的問題是當我運行這個預測 function 時，我收到以下錯誤。

  File "predict-binary.py", line 24, in predict
    array = model.predict(x)
  File ".venv\lib\site-packages\tensorflow\python\keras\engine\training.py", line 1629, in predict    
    tmp_batch_outputs = self.predict_function(iterator)
  File ".venv\lib\site-packages\tensorflow\python\eager\def_function.py", line 828, in __call__       
    result = self._call(*args, **kwds)
  File ".venv\lib\site-packages\tensorflow\python\eager\def_function.py", line 871, in _call
    self._initialize(args, kwds, add_initializers_to=initializers)
  File ".venv\lib\site-packages\tensorflow\python\eager\def_function.py", line 725, in _initialize    
    self._stateful_fn._get_concrete_function_internal_garbage_collected(  # pylint: disable=protected-access
  File ".venv\lib\site-packages\tensorflow\python\eager\function.py", line 2969, in _get_concrete_function_internal_garbage_collected
    graph_function, _ = self._maybe_define_function(args, kwargs)
  File ".venv\lib\site-packages\tensorflow\python\eager\function.py", line 3361, in _maybe_define_function
    graph_function = self._create_graph_function(args, kwargs)
  File ".venv\lib\site-packages\tensorflow\python\eager\function.py", line 3196, in _create_graph_function
    func_graph_module.func_graph_from_py_func(
  File ".venv\lib\site-packages\tensorflow\python\framework\func_graph.py", line 990, in func_graph_from_py_func
    func_outputs = python_func(*func_args, **func_kwargs)
  File ".venv\lib\site-packages\tensorflow\python\eager\def_function.py", line 634, in wrapped_fn     
    out = weak_wrapped_fn().__wrapped__(*args, **kwds)
  File ".venv\lib\site-packages\tensorflow\python\framework\func_graph.py", line 977, in wrapper      
    raise e.ag_error_metadata.to_exception(e)
ValueError: in user code:

    .venv\lib\site-packages\tensorflow\python\keras\engine\training.py:1478 predict_function  *       
        return step_function(self, iterator)
    .venv\lib\site-packages\tensorflow\python\keras\engine\training.py:1468 step_function  **
        outputs = model.distribute_strategy.run(run_step, args=(data,))
    .venv\lib\site-packages\tensorflow\python\distribute\distribute_lib.py:1259 run
        return self._extended.call_for_each_replica(fn, args=args, kwargs=kwargs)
    .venv\lib\site-packages\tensorflow\python\distribute\distribute_lib.py:2730 call_for_each_replica 
        return self._call_for_each_replica(fn, args, kwargs)
    .venv\lib\site-packages\tensorflow\python\distribute\distribute_lib.py:3417 _call_for_each_replica
        return fn(*args, **kwargs)
    .venv\lib\site-packages\tensorflow\python\keras\engine\training.py:1461 run_step  **
        outputs = model.predict_step(data)
    .venv\lib\site-packages\tensorflow\python\keras\engine\training.py:1434 predict_step
        return self(x, training=False)
    .venv\lib\site-packages\tensorflow\python\keras\engine\base_layer.py:1012 __call__
        outputs = call_fn(inputs, *args, **kwargs)
    .venv\lib\site-packages\tensorflow\python\keras\engine\sequential.py:375 call
        return super(Sequential, self).call(inputs, training=training, mask=mask)
    .venv\lib\site-packages\tensorflow\python\keras\engine\functional.py:424 call
        return self._run_internal_graph(
    .venv\lib\site-packages\tensorflow\python\keras\engine\functional.py:560 _run_internal_graph      
        outputs = node.layer(*args, **kwargs)
    .venv\lib\site-packages\tensorflow\python\keras\engine\base_layer.py:998 __call__
        input_spec.assert_input_compatibility(self.input_spec, inputs, self.name)
    .venv\lib\site-packages\tensorflow\python\keras\engine\input_spec.py:255 assert_input_compatibility
        raise ValueError(

    ValueError: Input 0 of layer dense is incompatible with the layer: expected axis -1 of input shape to have value 8192 but received input with shape (None, 61608)

我假設我需要在 model 的Flatten()和Dense()層之間進行一些更改，但我不確定是什么。 我試圖在這兩者之間添加model.add(Dense(61608, activation='relu))因為這似乎是我在另一篇文章中看到的建議（現在找不到鏈接），但它導致了同樣的錯誤。 （我也嘗試使用 8192 而不是 61608）。 任何幫助表示贊賞，謝謝。

編輯＃1：

更改 model 創建/培訓代碼，因為我認為Gerry P對此提出了建議

    img_shape = (151,136,3)
base_model=VGG19( include_top=False, input_shape=img_shape, pooling='max', weights='imagenet' )
x=base_model.output
x=Dense(100, activation='relu')(x)
x=Dropout(0.5)(x)
x=BatchNormalization(axis=-1, momentum=0.99, epsilon=0.001)(x)
output=Dense(2, activation='softmax')(x)
model=Model(inputs=base_model.input, outputs=output)

image_gen = ImageDataGenerator()
train = image_gen.flow_from_directory('./data/train', class_mode='categorical', shuffle=False, batch_size=10, target_size=(151, 136))
val = image_gen.flow_from_directory('./data/validate', class_mode='categorical', shuffle=False, batch_size=10, target_size=(151, 136))

vgg_features_train = base_model.predict(train)
vgg_features_val = base_model.predict(val)

train_target = to_categorical(train.labels)
val_target = to_categorical(val.labels)

model.compile(optimizer='rmsprop', metrics=['accuracy'], loss='categorical_crossentropy')

model.fit(vgg_features_train, train_target, epochs=100, batch_size=8, validation_data=(vgg_features_val, val_target), callbacks=callbacks_list)

這導致File "train-binary.py", line 37, in <module> model.fit(vgg_features_train, train_target, epochs=100, batch_size=8, validation_data=(vgg_features_val, val_target), callbacks=callbacks_list) ValueError: Input 0 is incompatible with layer model: expected shape=(None, 151, 136, 3), found shape=(None, 512)的不同輸入形狀錯誤File "train-binary.py", line 37, in <module> model.fit(vgg_features_train, train_target, epochs=100, batch_size=8, validation_data=(vgg_features_val, val_target), callbacks=callbacks_list) ValueError: Input 0 is incompatible with layer model: expected shape=(None, 151, 136, 3), found shape=(None, 512)

Answer 1

您的 model 期望看到 model.predict 的輸入，其尺寸與訓練時的尺寸相同。 在這種情況下，它是 vgg_features_train 的尺寸。您生成的 model.predict 的輸入是 VGG model 的輸入。 您實際上是在嘗試進行遷移學習，因此我建議您按照以下步驟進行

base_model=tf.keras.applications.VGG19( include_top=False, input_shape=img_shape, pooling='max', weights='imagenet' )
x=base_model.output
x=Dense(100, activation='relu'))(x)
x=Dropout(0.5)(x)
x=BatchNormalization(axis=-1, momentum=0.99, epsilon=0.001)(x)
output=Dense(2, activation='softmax')(x)
model=Model(inputs=base_model.input, outputs=output)
model.fit( train, epochs=100, batch_size=8, validation_data=val, callbacks=callbacks_list)

現在進行預測，您可以使用與訓練 model 相同的尺寸。