空维度形状“ ValueError：检查目标时出错：预期density_2具有2个维度，但数组的形状为（）”

Question

我在创建Keras CNN + LSTM模型时遇到问题：

  

ValueError：检查目标时出错：预期density_2具有2   尺寸，但数组的形状为（）

我已经删除了一些图层以测试问题。但是什么都没有改变。

背景： 我正在尝试分析4D数据：3D图像+ 1D时间序列。 我的工作方式是一次添加1张图片，然后由CNN + LSTM模型进行分析。我设法使尺寸正确并流经模型。但是后来我遇到了上面提到的错误。

# define CNN model
model = Sequential()

#Layer 1
model.add(TimeDistributed(Conv3D(32, kernel_size=(5, 5, 5), strides=(1, 1, 1),
                 activation='relu'), 
                input_shape=input_shape))

model.add(TimeDistributed(MaxPooling3D(pool_size=(2, 2, 2), strides=(2, 2, 2))))

#Layer 2
model.add(TimeDistributed(Conv3D(64, (5, 5, 5), activation='relu')))
model.add(TimeDistributed(MaxPooling3D(pool_size=(2, 2, 2))))

#Layer 3
model.add(TimeDistributed(Conv3D(128, (5, 5, 5), activation='relu')))
model.add(TimeDistributed(MaxPooling3D(pool_size=(2, 2, 2))))

#Flatten
model.add(TimeDistributed(Flatten()))

# LSTM
model.add(LSTM(512, return_sequences=True))
model.add(LSTM(512))
model.add(Dense(1201))

#Dense
model.add(Dense(num_classes, activation='sigmoid'))

#Compile
model.compile(loss='categorical_crossentropy',
              optimizer='adam',
              metrics=['accuracy'])

model.summary()

train_gen = generate_images(train_dataset)
model.fit_generator(train_gen, samples_per_epoch=5, nb_epoch=10)

###################################################################

#Data Generator
def generate_images(dataframe):

    while True:
        sub_dataframe = dataframe.sample(n=1)
        batch_input = []
#         batch_output = []

#         for index, row in sub_dataframe.iterrows(): # iterate through each row

        input_path = os.path.join(base_directory, sub_dataframe['Image'].values[0])
        img = get_fmri_sequence(input_path)
        img = np.expand_dims(img, axis=-1)
        batch_input.append(img)

        batch_input = np.array(batch_input)
        yield (batch_input, sub_dataframe["DX"].values[0])

###################################################################

这是模型摘要：

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
time_distributed_1 (TimeDist (None, None, 43, 54, 45,  4032      
_________________________________________________________________
time_distributed_2 (TimeDist (None, None, 21, 27, 22,  0         
_________________________________________________________________
time_distributed_3 (TimeDist (None, None, 17, 23, 18,  256064    
_________________________________________________________________
time_distributed_4 (TimeDist (None, None, 8, 11, 9, 64 0         
_________________________________________________________________
time_distributed_5 (TimeDist (None, None, 4, 7, 5, 128 1024128   
_________________________________________________________________
time_distributed_6 (TimeDist (None, None, 2, 3, 2, 128 0         
_________________________________________________________________
time_distributed_7 (TimeDist (None, None, 1536)        0         
_________________________________________________________________
lstm_1 (LSTM)                (None, None, 512)         4196352   
_________________________________________________________________
lstm_2 (LSTM)                (None, 512)               2099200   
_________________________________________________________________
dense_1 (Dense)              (None, 1201)              616113    
_________________________________________________________________
dense_2 (Dense)              (None, 2)                 2404      
=================================================================
Total params: 8,198,293
Trainable params: 8,198,293
Non-trainable params: 0
_________________________________________________________________