Python 必须使用dtype float和shape[？，？，？，，？，？]为占位符张量输入一个值

我正在使用tensorflow 2.3.0版和tensorflow gpu 1.14.0版的类似UNet的模型（实际上是注意力UNet）。我在GPU中运行我的模型。它接受一个张量，填充它，最终在扩展路径将其裁剪到原始大小后，这样填充就被移除了（来自Keras的简化摘要）：

。。。图层…（

我为训练集和验证集创建数据生成器，并编译模型。然后我将其与以下代码相匹配：

def run(self, load_weights: bool = False):
        config = tf.compat.v1.ConfigProto(device_count={'GPU': 1})
        config.gpu_options.allow_growth = True
        config.log_device_placement=True      
        
        sess =  tf.compat.v1.Session(config=config)
        
        with sess as sess:
                        
            self.get_model()
            print(self._model.summary())

            self.initialise_datagenerator()
            
            if load_weights:
                self.get_weights()
                
            callbacks = []
            if with_callbacks:
                earlystopper = EarlyStopping(patience = 5, verbose = 1, monitor = 'val_loss')
                checkpointer = ModelCheckpoint('../../save/model', verbose = 1, monitor = 'val_loss', save_best_only= True,save_weights_only= True)
                tb = TensorBoard(log_dir = '../../logs_tb/model', histogram_freq= 0, write_graph = True, write_grads = False, write_images = True)
                callbacks.extend([earlystopper, checkpointer, tb])

            print('Fitting started')
            self._model.fit(self.training_generator, epochs = self._epochs, validation_data = self.validation_generator, callbacks = callbacks)

它确实成功地运行了第一个时代。但是，当它必须输出验证丢失时，它会抛出以下错误，这似乎与裁剪层有关。我自己也无法理解，因为它在训练中运行良好

Traceback (most recent call last):
  File "/XXX/XXX/XXX/XXX/XXX/XXX/training/train_script.py", line 27, in <module>
    t.run()
  File "/XXX/XXX/XXX/XXX/XXX/XXX/model/train.py", line 243, in run
    self._model.fit(self.training_generator, epochs = self._epochs, validation_data = self.validation_generator, callbacks = callbacks)
  File "/home/XXX/miniconda3/envs/segment/lib/python3.7/site-packages/tensorflow/python/keras/engine/training_v1.py", line 809, in fit
    use_multiprocessing=use_multiprocessing)
  File "/home/XXX/miniconda3/envs/segment/lib/python3.7/site-packages/tensorflow/python/keras/engine/training_generator.py", line 590, in fit
    steps_name='steps_per_epoch')
  File "/home/kdqm927/miniconda3/envs/segment/lib/python3.7/site-packages/tensorflow/python/keras/engine/training_generator.py", line 310, in model_iteration
    steps_name='validation_steps')
  File "/home/XXX/miniconda3/envs/segment/lib/python3.7/site-packages/tensorflow/python/keras/engine/training_generator.py", line 256, in model_iteration
    batch_outs = batch_function(*batch_data)
  File "/home/XXX/miniconda3/envs/segment/lib/python3.7/site-packages/tensorflow/python/keras/engine/training_v1.py", line 1165, in test_on_batch
    outputs = self.test_function(inputs)  # pylint: disable=not-callable
  File "/home/XXX/miniconda3/envs/segment/lib/python3.7/site-packages/tensorflow/python/keras/backend.py", line 3825, in __call__
    run_metadata=self.run_metadata)
  File "/home/XXX/miniconda3/envs/segment/lib/python3.7/site-packages/tensorflow/python/client/session.py", line 1472, in __call__
    run_metadata_ptr)
tensorflow.python.framework.errors_impl.InvalidArgumentError: You must feed a value for placeholder tensor 'cropping3d_target' with dtype float and shape [?,?,?,?,?]
     [[{{node cropping3d_target}}]]

回溯（最近一次呼叫最后一次）：
文件“/XXX/XXX/XXX/XXX/XXX/training/train_script.py”，第27行，在
t、 运行（）
文件“/XXX/XXX/XXX/XXX/XXX/model/train.py”，第243行，运行中
self.\u model.fit（self.training\u generator，epochs=self.\u epochs，validation\u data=self.validation\u generator，callbacks=callbacks）
文件“/home/XXX/miniconda3/envs/segment/lib/python3.7/site packages/tensorflow/python/keras/engine/training_v1.py”，第809行，适合
使用多处理=使用多处理）
文件“/home/XXX/miniconda3/envs/segment/lib/python3.7/site packages/tensorflow/python/keras/engine/training_generator.py”，第590行
步骤（名称=“每个时代的步骤”）
文件“/home/kdqm927/miniconda3/envs/segment/lib/python3.7/site packages/tensorflow/python/keras/engine/training_generator.py”，第310行，在模型迭代中
步骤（name='validation'）
文件“/home/XXX/miniconda3/envs/segment/lib/python3.7/site packages/tensorflow/python/keras/engine/training\u generator.py”，第256行，在模型迭代中
批处理输出=批处理功能（*批处理数据）
文件“/home/XXX/miniconda3/envs/segment/lib/python3.7/site packages/tensorflow/python/keras/engine/training\u v1.py”，第1165行，批量测试
输出=自测试功能（输入）#pylint:disable=不可调用
文件“/home/XXX/miniconda3/envs/segment/lib/python3.7/site packages/tensorflow/python/keras/backend.py”，第3825行，在__
run\u元数据=self.run\u元数据）
文件“/home/XXX/miniconda3/envs/segment/lib/python3.7/site packages/tensorflow/python/client/session.py”，第1472行，在__
运行_元数据_ptr）
tensorflow.python.framework.errors\u impl.InvalidArgumentError：必须为带有数据类型float和形状[？，？，，，？，，？，？]的占位符tensor“cropping3d\u target”提供一个值
[{{node croping3d_target}}]]