【问题标题】:<keras.preprocessing.image.DirectoryIterator> Object returns TypeError: unsupported operand type(s) for +: 'int' and 'str'<keras.preprocessing.image.DirectoryIterator> 对象返回类型错误:+ 的不支持的操作数类型:'int' 和 'str'
【发布时间】:2021-06-05 17:11:33
【问题描述】:

我使用以下代码 sn-p 为训练和验证生成器创建了 &lt;keras.preprocessing.image.DirectoryIterator&gt; 对象。

class DataLoader:

    @staticmethod
    def load_data(data_config, prefix = "blond"):
      
      train_datagen =  tf.keras.preprocessing.image.ImageDataGenerator(
            preprocessing_function=tf.keras.applications.inception_v3.preprocess_input,
            rotation_range=30,
            width_shift_range=0.2,
            height_shift_range=0.2,
            shear_range=0.2,
            zoom_range=0.2,
            horizontal_flip=True
        )

      valid_datagen = tf.keras.preprocessing.image.ImageDataGenerator(
            preprocessing_function=tf.keras.applications.inception_v3.preprocess_input,
            rotation_range=30,
            width_shift_range=0.2,
            height_shift_range=0.2,
            shear_range=0.2,
            zoom_range=0.2,
            horizontal_flip=True
        )
      train_generator = train_datagen.flow_from_directory(
            'data/celeba-dataset/{}-train'.format(prefix),
                target_size=(data_config.data.IMG_HEIGHT, data_config.data.IMG_WIDTH),
                batch_size=data_config.train.BATCH_SIZE)

然后我创建另一个 Model 类,其中包含以下 load_data 和 train 函数

class Model(BaseModel):

    def __init__(self, config):
        super().__init__(config)
        self.img_height = int(self.config.data.IMG_HEIGHT)
        self.img_width = int(self.config.data.IMG_WIDTH)
        self.base_model = tf.keras.applications.InceptionV3(weights='imagenet',
                        include_top=False,
                        input_shape=(self.img_height, self.img_width, 3))

        self.model = None
        self.training_samples = int(self.config.data.TRAINING_SAMPLES)
        self.batch_size = int(self.config.train.BATCH_SIZE)
        self.steps_per_epoch = int(self.training_samples) // int(self.batch_size)
        self.num_epochs = int(self.config.train.EPOCHS)

        self.train_generator = None
        self.validation_generator = None
    
    def load_data(self):

        self.train_generator, self.validation_generator = DataLoader().load_data(self.config)

    def train(self):
        self.model.compile(optimizer=tf.keras.optimizers.SGD(lr=0.0001, momentum=0.9), 
               loss='categorical_crossentropy', 
               metrics=[tf.keras.metrics.TopKCategoricalAccuracy(k = 1)])

        checkpointer = tf.keras.callbacks.ModelCheckpoint(filepath='weights.best.inc.blond.hdf5', 
                               verbose=1, save_best_only=True)

        model_history = self.model.fit(self.train_generator,
                                        validation_data = self.validation_generator,
                                        steps_per_epoch= self.steps_per_epoch,
                                        epochs = self.num_epochs,
                                        class_weight='auto',
                                        callbacks=[checkpointer])
        return model_history.history['loss'], model_history.history['val_loss']

在运行以下代码时,

model = Model(CFG)
model.load_data()
model.build()
model.train()

我得到以下 Traceback

Traceback (most recent call last):
  File "/Users/sauravmaheshkar/github/compression/train.py", line 14, in <module>
    model.train()
  File "/Users/sauravmaheshkar/github/compression/model/ForgetModel.py", line 70, in train
    callbacks=[checkpointer, ],
  File "/Users/sauravmaheshkar/opt/anaconda3/envs/compression/lib/python3.6/site-packages/tensorflow/python/keras/engine/training.py", line 1064, in fit
    steps_per_execution=self._steps_per_execution)
  File "/Users/sauravmaheshkar/opt/anaconda3/envs/compression/lib/python3.6/site-packages/tensorflow/python/keras/engine/data_adapter.py", line 1112, in __init__
    model=model)
  File "/Users/sauravmaheshkar/opt/anaconda3/envs/compression/lib/python3.6/site-packages/tensorflow/python/keras/engine/data_adapter.py", line 898, in __init__
    self._size = len(x)
  File "/Users/sauravmaheshkar/opt/anaconda3/envs/compression/lib/python3.6/site-packages/keras_preprocessing/image/iterator.py", line 68, in __len__
    return (self.n + self.batch_size - 1) // self.batch_size  # round up
TypeError: unsupported operand type(s) for +: 'int' and 'str'

Config.py 文件

"""Project Config in JSON"""

CFG = {
    "data": {
        "data_folder" : "data/CelebA/",
        "images_folder" : "data/CelebA/img_align_celeba/img_align_celeba/",
        "IMG_HEIGHT": "218",
        "IMG_WIDTH": "178",
        "TRAINING_SAMPLES": "10000",
        "VALIDATION_SAMPLES": "2000",
        "TEST_SAMPLES": "2000",
    },
    "train": {
        "BATCH_SIZE": "64",
        "EPOCHS": "10",
    }   
}

软件包版本

  • tensorflow==2.4.1
  • Keras-Preprocessing==1.1.2

预期输出

运行正常的张量流训练循环

【问题讨论】:

  • 堆栈跟踪中引用您的代码的第一行是:文件“/Users/sauravmaheshkar/github/compression/model/ForgetModel.py”,第 70 行,在 train callbacks=[checkpointer, ] , 你省略了checkpointer 的定义。请提供...代表的代码。最终,self.batch_size 是 str,但是如果没有丢失的代码,就很难确定哪里出了问题。
  • 我刚刚添加了整个代码@MichaelRuth

标签: python tensorflow keras tensorflow2.0 tf.keras


【解决方案1】:

我认为问题出在 DataLoader 类的 load_data 函数中。具体来说,在这一行:

batch_size=data_config.train.BATCH_SIZE.

那行,你的数据配置文件告诉我你只需要添加int,你的问题就解决了。因此,将该行替换为:

batch_size=int(data_config.train.BATCH_SIZE)

我建议你也检查一下其他参数。

或者,我认为只需从 Json 文件中的 int 值中删除引号也可以。

如您所见,here 数字/整数不需要引号。

【讨论】:

  • 谢谢,我在Model 类中明确地对batch_size 的每个引用进行了类型转换,但忘记了DataLoader 类。之后出现了一些后续错误,但已修复。
猜你喜欢
  • 2014-06-16
  • 1970-01-01
  • 1970-01-01
  • 1970-01-01
  • 2018-02-05
  • 1970-01-01
  • 2017-05-27
  • 2013-10-12
相关资源
最近更新 更多