如何在自定义 DataGenerator 中实现缩放？

Question

我使用 tf.keras.utils.Sequence:(https://www.tensorflow.org/api_docs/python/tf/keras/utils/Sequence) 创建了自定义 DataGenerator。这是自定义的 DataGenerator:

# Here, `x_set` is list of path to the images
# and `y_set` are the associated classes.

class DataGenerator(Sequence):

    def __init__(self, x_set, y_set, batch_size):
        self.x, self.y = x_set, y_set
        self.batch_size = batch_size

    def __len__(self):
        return math.ceil(len(self.x) / self.batch_size)

    def __getitem__(self, idx):
        batch_x = self.x[idx * self.batch_size:(idx + 1) *
        self.batch_size]
        batch_x = np.array(batch_x)
        batch_x = batch_x*1/255
        batch_y = self.y[idx * self.batch_size:(idx + 1) *
        self.batch_size]

        return np.array([
            resize(imread(file_name), (64, 128))
               for file_name in batch_x]), np.array(batch_y)

x_set 是我的图像的路径列表，y_set 是关联的 classes。我现在想添加一个函数来重新缩放图像的每个像素，方法是将其与 rescale = 1./255 相乘，如 ImageDataGenerator class: https://keras.io/api/preprocessing/image/#ImageDataGenerator%20class

当我将此代码与 model.fit_generator 一起应用时：

model.fit_generator(generator=training_generator,
                    validation_data=validation_generator,
                    steps_per_epoch = num_train_samples // 128,
                    validation_steps = num_val_samples // 128,
                    epochs = 10)

我收到这个错误：

---------------------------------------------------------------------------
UFuncTypeError                            Traceback (most recent call last)
<ipython-input-62-571a868b2d2a> in <module>()
      3                     steps_per_epoch = num_train_samples // 128,
      4                     validation_steps = num_val_samples // 128,
----> 5                     epochs = 10)

8 frames
<ipython-input-54-d98c3b0c7c56> in __getitem__(self, idx)
     15         self.batch_size]
     16         batch_x = np.array(batch_x)
---> 17         batch_x = batch_x*1/255
     18         batch_y = self.y[idx * self.batch_size:(idx + 1) *
     19         self.batch_size]

UFuncTypeError: ufunc 'multiply' did not contain a loop with signature matching types (dtype('<U77'), dtype('<U77')) -> dtype('<U77')

我必须如何修改我的代码？

Answer 1

这样试试

class DataGenerator(Sequence):

    def __init__(self, x_set, y_set, batch_size):
        self.x, self.y = x_set, y_set
        self.batch_size = batch_size

    def __len__(self):
        return math.ceil(len(self.x) / self.batch_size)

    def __getitem__(self, idx):
        batch_x = self.x[idx*self.batch_size : (idx + 1)*self.batch_size]
        batch_x = np.array([resize(imread(file_name), (64, 128)) for file_name in batch_x])
        batch_x = batch_x * 1./255
        batch_y = self.y[idx*self.batch_size : (idx + 1)*self.batch_size]
        batch_y = np.array(batch_y)

        return batch_x, batch_y

如何在自定义 DataGenerator 中实现缩放？

How to implement rescaling in custom DataGenerator?

python

data-generation

keras

tensorflow

image-preprocessing