How to convert ImageDataGenerator(DirectoryIterator) to dataset?

Question

I want to load a large dataset of images which does not fit into the memory for semantic segmentation. I'm trying to use data generator and I created two generators for image and mask. However I can't find how do I make these generators into the dataset. What I want to do is to load large numbers of images and masks from the directory, and apply preprocessing to the mask(converting them from RGB to categorical) then train U-Net for binary semantic segmentation. Here is the code I tried so far.

# we create two instances with the same arguments
data_gen_args = dict(featurewise_center=True,
                     featurewise_std_normalization=True,
                     rotation_range=90,
                     width_shift_range=0.1,
                     height_shift_range=0.1,
                     zoom_range=0.2)
image_datagen = ImageDataGenerator(**data_gen_args)
mask_datagen = ImageDataGenerator(**data_gen_args)
seed=123
image_generator = image_datagen.flow_from_directory(
    "image_directory/",
    classes=['images'],
    class_mode=None,
    seed=seed)
mask_generator = mask_datagen.flow_from_directory(
    "label_directory",
    classes=['labels'],
    class_mode=None,
    seed=seed)```

> Found 12996 images belonging to 1 classes.
> Found 12996 images belonging to 1 classes.
> 
> Blockquote

Edit:
After I run 
ds = tf.data.Dataset.from_generator(image_generator,output_types=(tf.float32))
I still get this error.

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Input In [20], in <cell line: 1>()
----> 1 ds = tf.data.Dataset.from_generator(image_generator,output_types=(tf.float32))

File /opt/conda/lib/python3.9/site-packages/tensorflow/python/util/deprecation.py:548, in deprecated_args.<locals>.deprecated_wrapper.<locals>.new_func(*args, **kwargs)
    540         _PRINTED_WARNING[(func, arg_name)] = True
    541       logging.warning(
    542           'From %s: calling %s (from %s) with %s is deprecated and will '
    543           'be removed %s.\nInstructions for updating:\n%s',
   (...)
    546           'in a future version' if date is None else ('after %s' % date),
    547           instructions)
--> 548 return func(*args, **kwargs)

File /opt/conda/lib/python3.9/site-packages/tensorflow/python/data/ops/dataset_ops.py:919, in DatasetV2.from_generator(generator, output_types, output_shapes, args, output_signature, name)
    840 """Creates a `Dataset` whose elements are generated by `generator`.
    841 
    842 The `generator` argument must be a callable object that returns
   (...)
    916   Dataset: A `Dataset`.
    917 """
    918 if not callable(generator):
--> 919   raise TypeError("`generator` must be a Python callable.")
    921 if output_signature is not None:
    922   if output_types is not None:

TypeError: `generator` must be a Python callable.