deepdoctection 0.42.1__py3-none-any.whl → 0.43.1__py3-none-any.whl

deepdoctection/datasets/info.py

@@ -16,7 +16,7 @@
 # limitations under the License.
 
 """
-Module for storing dataset info (e.g. general meta data or categories)
+General meta-data or categories
 """
 
 from copy import copy
@@ -53,10 +53,13 @@ def _get_dict(
     """
     Converts a list into a dict, where keys/values are the list indices.
 
-    :param l: A list of categories
-    :param name_as_key: Whether to return the dict with category names as key (True)
-    :param starts_with: index count start
-    :return: A dictionary of list indices/list elements.
+    Args:
+        l: A list of categories
+        name_as_key: Whether to return the dict with category names as key (`True`)
+        starts_with: index count start
+
+    Returns:
+        A dictionary of list indices/list elements.
     """
     if name_as_key:
         return {v: k for k, v in enumerate(l, starts_with)}
@@ -66,22 +69,17 @@ def _get_dict(
 @dataclass
 class DatasetInfo:
     """
-    DatasetInfo is a simple dataclass that stores some meta-data information about a dataset.
-
-    `name`: Name of the dataset. Using the name you can retrieve the dataset from the
-    `registry.DatasetRegistry`.
-
-    `description`: Short description of the dataset.
-
-    `license`: License to the dataset.
-
-    `url`: url, where the dataset can be downloaded from.
-
-    `splits`: A dict of splits. The value must store the relative path, where the split can be found.
-
-    `type`: The type describes whether this is a dataset for object detection (pass 'OBJECT_DETECTION'),
-    sequence classification (pass 'SEQUENCE_CLASSIFICATION') or token classification ('TOKEN_CLASSIFICATION').
-    Optionally, pass `None`.
+    `DatasetInfo` is a simple dataclass that stores some meta-data information about a dataset.
+
+     Attributes:
+        name: Name of the dataset. Using the name you can retrieve the dataset from the `registry.DatasetRegistry`.
+        description: Short description of the dataset.
+        license: License to the dataset.
+        url: url, where the dataset can be downloaded from.
+        splits: A `dict` of splits. The value must store the relative path, where the split can be found.
+        type: The type describes whether this is a dataset for object detection (pass 'OBJECT_DETECTION'),
+              sequence classification (pass 'SEQUENCE_CLASSIFICATION') or token classification ('TOKEN_CLASSIFICATION').
+              Optionally, pass `None`.
     """
 
     name: str
@@ -96,8 +94,11 @@ class DatasetInfo:
         """
         Get the split directory by its key (if it exists).
 
-        :param key: The key to a split (i.e. "train", "val", "test")
-        :return: The local directory path to the split. An empty string if the key doesn't exist.
+        Args:
+            key: The key to a split (i.e. `train`, `val`,`test`)
+
+        Returns:
+            The local directory path to the split. An empty string if the key doesn't exist.
         """
 
         return self.splits[key]
@@ -112,22 +113,26 @@ class DatasetCategories:
     for the index/category name relationship and guarantees that a sequence of natural numbers for the categories
     is always returned as the category-id even after replacing and/or filtering.
 
-    `init_categories`: A list of category names. The list must include all categories that can occur within the
-    annotations.
-
-    `init_sub_categories`: A dict of categories/sub-categories. Each sub-category that can appear in the
-    annotations in combination with a category must be listed.
+    Attributes:
+        init_categories: A list of `category_name`s. The list must include all categories that can occur within the
+                         annotations.
+        init_sub_categories: A dict of categories/sub-categories. Each sub-category that can appear in the
+                             annotations in combination with a category must be listed.
 
-    **Example:**
+    Example:
 
         An annotation file hast the category/sub-category combinations for three datapoints:
 
-            (cat1,s1),(cat1,s2), (cat2,s2).
+        ```python
+        (cat1,s1),(cat1,s2), (cat2,s2).
+        ```
 
         You must list `init_categories`, `init_sub_categories` as follows:
 
-            init_categories = [cat1,cat2]
-            init_sub_categories = {cat1: [s1,s2],cat2: [s2]}
+        ```python
+        init_categories = [cat1,cat2]
+        init_sub_categories = {cat1: [s1,s2],cat2: [s2]}
+        ```
 
     Use `filter_categories` or `set_cat_to_sub_cat` to filter or swap categories with sub-categories.
     """
@@ -173,14 +178,17 @@ class DatasetCategories:
         categories of replaced categories with sub categories. However, you must correctly pass arguments to return the
         state you want.
 
-        :param as_dict: Will pass a dict if set to 'True' otherwise a list.
-        :param name_as_key: Categories are stored as key/value pair in a dict with integers as keys. name_as_key set to
-                            "False" will swap keys and values.
-        :param init: If set to "True" it will return the list/dict of categories as initially provided. Manipulations
-                     due to replacing/filtering will not be regarded.
-        :param filtered: If set to "True" will return an unfiltered list of all categories. If a replacing has been
-                         invoked selected sub categories will be returned.
-        :return: A dict of index/category names (or the other way around) or a list of category names.
+        Args:
+            as_dict: Will pass a dict if set to 'True' otherwise a list.
+            name_as_key: Categories are stored as key/value pair in a dict with integers as keys. `name_as_key` set to
+                         `False` will swap keys and values.
+            init: If set to `True` it will return the list/dict of categories as initially provided. Manipulations
+                  due to replacing/filtering will not be regarded.
+            filtered: If set to `True` will return an unfiltered list of all categories. If a replacing has been
+                      invoked selected sub categories will be returned.
+
+        Returns:
+            A dict of index/category names (or the other way around) or a list of category names.
         """
         if init:
             if as_dict:
@@ -209,14 +217,17 @@ class DatasetCategories:
         """
         Returns a dict of list with a category name and their sub categories.
 
-        :param categories: A single category or list of category names
-        :param sub_categories: A mapping of categories to sub category keys on which the result should be filtered. Only
+        Args:
+            categories: A single category or list of category names
+            sub_categories: A mapping of categories to sub category keys on which the result should be filtered. Only
                                relevant, if `keys=False`
-        :param keys: Will only pass keys if set to `True`.
-        :param values_as_dict: Will generate a dict with indices and sub category value names if set to `True`.
-        :param name_as_key: sub category values are stored as key/value pair in a dict with integers as keys.
+            keys: Will only pass keys if set to `True`.
+            values_as_dict: Will generate a dict with indices and sub category value names if set to `True`.
+            name_as_key: sub category values are stored as key/value pair in a dict with integers as keys.
                             name_as_key set to `False` will swap keys and values.
-        :return: Dict with all selected categories.
+
+        Returns:
+            Dict with all selected categories.
         """
         _categories: Sequence[ObjectTypes]
         if isinstance(categories, (ObjectTypes, str)):
@@ -293,14 +304,16 @@ class DatasetCategories:
         This method can only be called once per object. Re-setting or further replacing of categories would make the
         code messy and is therefore not allowed.
 
-        **Example:**
-
-                  cat_to_sub_cat={cat1: sub_cat1}
+        Example:
+            ```python
+            cat_to_sub_cat={cat1: sub_cat1}
+            ```
 
             will replace cat1 with sub_cat1 as category. This will also be respected when returning datapoints.
 
-        :param cat_to_sub_cat: A dict of pairs of category/sub-category. Note that the combination must be available
-                               according to the initial settings.
+        Args:
+            cat_to_sub_cat: A dict of pairs of category/sub-category. Note that the combination must be available
+                            according to the initial settings.
         """
 
         _cat_to_sub_cat = {get_type(key): get_type(value) for key, value in cat_to_sub_cat.items()}
@@ -327,7 +340,8 @@ class DatasetCategories:
         Filter categories of a dataset. This will keep all the categories chosen and remove all others.
         This method can only be called once per object.
 
-        :param categories: A single category name or a list of category names.
+        Args:
+            categories: A single `category_name` or a list of `category_name`s.
         """
 
         if not self._allow_update:
@@ -344,13 +358,14 @@ class DatasetCategories:
     @property
     def cat_to_sub_cat(self) -> Optional[Mapping[ObjectTypes, ObjectTypes]]:
         """
-        cat_to_sub_cat
+        `cat_to_sub_cat`
         """
         return self._cat_to_sub_cat
 
     def is_cat_to_sub_cat(self) -> bool:
         """
-        returns `True` if a category is replaced with sub categories
+        Returns:
+            `True` if a category is replaced with sub categories
         """
         if self._cat_to_sub_cat is not None:
             return True
@@ -358,7 +373,8 @@ class DatasetCategories:
 
     def is_filtered(self) -> bool:
         """
-        return `True` if categories are filtered
+        Returns:
+            `True` if categories are filtered
         """
         if hasattr(self, "_categories_filter_update"):
             return True
@@ -379,8 +395,11 @@ def get_merged_categories(*categories: DatasetCategories) -> DatasetCategories:
     as well but no sub category than the merged dataset will have no sub categories at all. Whereas in a similar setting
     dataset B has sub category `foo`:`bak`, then `bak` will be an optional sub category for the merged dataset as well.
 
-    :param categories: A tuple/list of dataset categories
-    :return: An instance of `DatasetCategories` to be used as `DatasetCategories` for merged datasets
+    Args:
+        categories: A tuple/list of dataset categories
+
+    Returns:
+        An instance of `DatasetCategories` to be used as `DatasetCategories` for merged datasets
     """
 
     # working with lists is not possible as the order of categories is important here

deepdoctection/datasets/instances/__init__.py

@@ -16,7 +16,9 @@
 # limitations under the License.
 
 """
-Init file for instances package. Place all datasets in a directory using the structure
+Dataset samples for pre-training and fine-tuning models
+
+Place all datasets in a **deep**doctection's cache
 
     deepdoctection
     ├── datasets
@@ -24,9 +26,11 @@ Init file for instances package. Place all datasets in a directory using the str
     │ ├── dataset_2
     │ ├── dataset_3
 
-If not sure use
+If not sure:
 
+    ```python
     print(dataset_instance.dataflow.get_workdir())
+    ```
 """
 
 from .doclaynet import *

deepdoctection/datasets/instances/doclaynet.py

@@ -25,6 +25,7 @@ Module for DocLayNet dataset. Place the dataset as follows
     ├── PNG
     │ ├── 0a0d43e301facee9e99cc33b9b16e732dd207135f4027e75f6aea2bf117535a2.png
 """
+
 from __future__ import annotations
 
 import os
@@ -101,7 +102,7 @@ _SUB_CATEGORIES: Mapping[ObjectTypes, Mapping[ObjectTypes, Sequence[ObjectTypes]
 @dataset_registry.register("doclaynet")
 class DocLayNet(DatasetBase):
     """
-    DocLayNetSeq
+    `DocLayNet`
     """
 
     @classmethod
@@ -125,7 +126,7 @@ class DocLayNet(DatasetBase):
 
 class DocLayNetBuilder(DataFlowBaseBuilder):
     """
-    DocLayNetBuilder dataflow builder
+    `DocLayNetBuilder` dataflow builder
     """
 
     def build(self, **kwargs: Union[str, int]) -> DataFlow:
@@ -133,15 +134,14 @@ class DocLayNetBuilder(DataFlowBaseBuilder):
         Returns a dataflow from which you can stream datapoints of images. The following arguments affect the returns
         of the dataflow:
 
-        `split:` Split of the dataset. Can be `train`,`val` or `test`. Default: `val`
-
-        `max_datapoints:` Will stop iterating after max_datapoints. Default: None
-
-        `load_image:` Will load the image for each datapoint.  Default: False
-
-        `fake_score:` Will add a fake score so that annotations look like predictions
+        Args:
+            kwargs: (split) Split of the dataset. Can be `train`,`val` or `test`. Default: `val`
+                    (max_datapoints) Will stop iterating after `max_datapoints`. Default: `None`
+                    (load_image) Will load the image for each datapoint.  Default: `False`
+                    (fake_score) Will add a fake score so that annotations look like predictions
 
-        :return: dataflow
+        Returns:
+            Dataflow
         """
         split = str(kwargs.get("split", "val"))
         max_datapoints = kwargs.get("max_datapoints")
@@ -233,11 +233,14 @@ class DocLayNetSeqBuilder(DataFlowBaseBuilder):
         Returns a dataflow from which you can stream datapoints of images. The following arguments affect the returns
         of the dataflow:
 
-        `split:` Split of the dataset. Can be `train`, `val` or `test`. Default: `val`
-        `max_datapoints:` Will stop iterating after max_datapoints. Default: `None`
-        `load_image:` Will load the image for each datapoint.  Default: `False`
+        Args:
+            kwargs:
+                (split) Split of the dataset. Can be `train`, `val` or `test`. Default: `val`
+                (max_datapoints) Will stop iterating after `max_datapoints`. Default: `None`
+                (load_image) Will load the image for each datapoint. Default: `False`
 
-        :return: dataflow
+        Returns:
+            Dataflow
         """
         split = str(kwargs.get("split", "val"))
         max_datapoints = kwargs.get("max_datapoints")

deepdoctection/datasets/instances/fintabnet.py

@@ -157,28 +157,22 @@ class FintabnetBuilder(DataFlowBaseBuilder):
         Returns a dataflow from which you can stream datapoints of images. The following arguments affect the returns
         of the dataflow:
 
-        `split:` Split of the dataset. Can be `train`,`val` or `test`. Default: `val`
-
-        `build_mode:` Returns the full image or crops a table according to the table bounding box. Pass `table`
-                           if you only want the cropped table. Default: ""
-
-        `max_datapoints:` Will stop iterating after max_datapoints. Default: `None`
-
-        `rows_and_cols:` Will add a 'item' image annotations that either represent a row or a column of a table.
-                              Note, that the type of the item (i.e. being a row or a column) can be inferred from the
-                              sub category added. Note further, that "item" are not originally part of the annotations
-                              and are inferred from cell positions and their associated table semantic. Default: `True`
-        `load_image:` Will load the image for each datapoint.  Default: `False`
-
-        `use_multi_proc:` As the original files are stored as pdf conversion into a numpy array is time-consuming.
-                        When setting use_multi_proc to True is will use several processes depending on the number
-                        of CPUs available.
-
-        `use_multi_proc_strict:` Will use strict mode in multiprocessing.
-
-        `fake_score:` Will add a fake score so that annotations look like predictions
-
-        :return: dataflow
+        Args:
+            kwargs:
+                (split) Split of the dataset. Can be `train`, `val` or `test`. Default: `val`
+                (build_mode) Returns the full image or crops a table according to the table bounding box. Pass `table`
+                             if you only want the cropped table. Default: `""`
+                (max_datapoints) Will stop iterating after `max_datapoints`. Default: `None`
+                (rows_and_cols) Will add 'item' image annotations that either represent a row or a column of a table.
+                                Default: `True`
+                (load_image) Will load the image for each datapoint. Default: `False`
+                (use_multi_proc) Uses multiple processes for PDF conversion. Default: `True`
+                (use_multi_proc_strict) Uses strict mode in multiprocessing. Default: `False`
+                (fake_score) Adds a fake score so that annotations look like predictions. Default: `False`
+                (pubtables_like) Treats the dataset as PubTables-like. Default: `False`
+
+        Returns:
+            Dataflow
         """
 
         split = str(kwargs.get("split", "val"))

deepdoctection/datasets/instances/funsd.py

@@ -54,8 +54,10 @@ def load_file(path_ann: PathLikeOrStr) -> FunsdDict:
     """
     Loading json file
 
-    :param path_ann: path
-    :return: dict
+    Args:
+        path_ann: path
+    Returns:
+        dict
     """
     anns = load_json(path_ann)
     path, file_name = os.path.split(path_ann)
@@ -144,11 +146,14 @@ class FunsdBuilder(DataFlowBaseBuilder):
         Returns a dataflow from which you can stream datapoints of images. The following arguments affect the returns
         of the dataflow:
 
-        `split:` Split of the dataset. "train" and "test" is available
-        `load_image:` Will load the image for each datapoint.  Default: `False`
-        `max_datapoints:` Will stop iterating after max_datapoints. Default: `None`
+        Args:
+            kwargs:
+                (split) Split of the dataset. Can be `train` or `test`. Default: `test`
+                (load_image) Will load the image for each datapoint. Default: `False`
+                (max_datapoints) Will stop iterating after `max_datapoints`. Default: `None`
 
-        :return: Dataflow
+        Returns:
+            Dataflow
         """
 
         split = str(kwargs.get("split", "test"))

deepdoctection/datasets/instances/iiitar13k.py

@@ -124,15 +124,15 @@ class IIITar13KBuilder(DataFlowBaseBuilder):
         Returns a dataflow from which you can stream datapoints of images. The following arguments affect the return
         values of the dataflow:
 
-        `split:` Split of the dataset. Can be `train`,`val` or `test`. Default: `val`
-
-        `max_datapoints:` Will stop iterating after max_datapoints. Default: `None`
-
-        `load_image:` Will load the image for each datapoint.  Default: `False`
-
-        `fake_score:` Will add a fake score so that annotations look like predictions
-
-        :return: dataflow
+        Args:
+            kwargs:
+                (split) Split of the dataset. Can be `train`, `val` or `test`. Default: `val`
+                (max_datapoints) Will stop iterating after `max_datapoints`. Default: `None`
+                (load_image) Will load the image for each datapoint. Default: `False`
+                (fake_score) Will add a fake score so that annotations look like predictions. Default: `False`
+
+        Returns:
+            Dataflow
         """
 
         if not lxml_available():

deepdoctection/datasets/instances/layouttest.py

@@ -66,7 +66,7 @@ _INIT_CATEGORIES = [LayoutType.TEXT, LayoutType.TITLE, LayoutType.LIST, LayoutTy
 @dataset_registry.register("testlayout")
 class LayoutTest(_BuiltInDataset):
     """
-    LayoutTest
+    `LayoutTest`
     """
 
     _name = _NAME
@@ -99,15 +99,15 @@ class LayoutTestBuilder(DataFlowBaseBuilder):
         Returns a dataflow from which you can stream datapoints of images. The following arguments affect the returns
         of the dataflow:
 
-        `split:` Split of the dataset. Only "test" is for this small sample available
+        Args:
+            kwargs:
+                (split) Split of the dataset. Only `test` is available for this small sample. Default: `test`
+                (max_datapoints) Will stop iterating after `max_datapoints`. Default: `None`
+                (load_image) Will load the image for each datapoint. Default: `False`
+                (fake_score) Will add a fake score so that annotations look like predictions. Default: `False`
 
-        `max_datapoints:` Will stop iterating after max_datapoints. Default: `None`
-
-        `load_image:` Will load the image for each datapoint.  Default: `False`
-
-        `fake_score:` Will add a fake score so that annotations look like predictions
-
-        :return: Dataflow
+        Returns:
+            Dataflow
         """
         split = str(kwargs.get("split", "test"))
         max_datapoints = kwargs.get("max_datapoints")

deepdoctection/datasets/instances/publaynet.py

@@ -73,7 +73,7 @@ _INIT_CATEGORIES = [LayoutType.TEXT, LayoutType.TITLE, LayoutType.LIST, LayoutTy
 @dataset_registry.register("publaynet")
 class Publaynet(_BuiltInDataset):
     """
-    Publaynet
+    `Publaynet`
     """
 
     _name = _NAME
@@ -107,15 +107,15 @@ class PublaynetBuilder(DataFlowBaseBuilder):
         Returns a dataflow from which you can stream datapoints of images. The following arguments affect the returns
         of the dataflow:
 
-        `split:` Split of the dataset. Can be `train`,`val` or `test`. Default: `val`
+        Args:
+            kwargs:
+                (split) Split of the dataset. Can be `train`, `val` or `test`. Default: `val`
+                (max_datapoints) Will stop iterating after `max_datapoints`. Default: `None`
+                (load_image) Will load the image for each datapoint. Default: `False`
+                (fake_score) Will add a fake score so that annotations look like predictions. Default: `False`
 
-        `max_datapoints:` Will stop iterating after max_datapoints. Default: `None`
-
-        `load_image:` Will load the image for each datapoint.  Default: `False`
-
-        `fake_score:` Will add a fake score so that annotations look like predictions
-
-        :return: dataflow
+        Returns:
+            Dataflow
         """
         split = str(kwargs.get("split", "val"))
         max_datapoints = kwargs.get("max_datapoints")

deepdoctection/datasets/instances/pubtables1m.py

@@ -119,23 +119,23 @@ class Pubtables1MDet(_BuiltInDataset):
 
 class Pubtables1MBuilder(DataFlowBaseBuilder):
     """
-    Pubtables1M dataflow builder
+    `Pubtables1M` dataflow builder
     """
 
     def build(self, **kwargs: Union[str, int]) -> DataFlow:
         """
-        Returns a dataflow from which you can stream datapoints of images. The following arguments affect the return
-        values of the dataflow:
-
-        `split:` Split of the dataset. Can be `train`, `val` or `test`. Default: `val`
-
-        `max_datapoints:` Will stop iterating after max_datapoints. Default: `None`
-
-        `load_image:` Will load the image for each datapoint.  Default: `False`
-
-        `fake_score:` Will add a fake score so that annotations look like predictions
-
-        :return: dataflow
+        Returns a dataflow from which you can stream datapoints of images. The following arguments affect the returns
+        of the dataflow:
+
+        Args:
+            kwargs:
+                (split) Split of the dataset. Can be `train`, `val` or `test`. Default: `val`
+                (max_datapoints) Will stop iterating after `max_datapoints`. Default: `None`
+                (load_image) Will load the image for each datapoint. Default: `False`
+                (fake_score) Will add a fake score so that annotations look like predictions. Default: `False`
+
+        Returns:
+            Dataflow
         """
 
         if not lxml_available():

deepdoctection/datasets/instances/pubtabnet.py

@@ -109,7 +109,7 @@ _SUB_CATEGORIES = {
 @dataset_registry.register("pubtabnet")
 class Pubtabnet(_BuiltInDataset):
     """
-    Pubtabnet
+    `Pubtabnet`
     """
 
     _name = _NAME
@@ -143,20 +143,18 @@ class PubtabnetBuilder(DataFlowBaseBuilder):
         Returns a dataflow from which you can stream datapoints of images. The following arguments affect the returns
         of the dataflow:
 
-        `split:` Split of the dataset. Can be `train`, `val` or `test`. Default: `val`
-
-        `max_datapoints:` Will stop iterating after max_datapoints. Default: `None`
-
-        `load_image:` Will load the image for each datapoint.  Default: `False`
-
-        `rows_and_cols:` Will add a 'item' image annotations that either represent a row or a column of a table.
-                         Note, that the type of the item (i.e. being a row or a column) can be inferred from the
-                         sub category added. Note further, that 'ITEM' are not originally part of the annotations
-                         and are inferred from cell positions and their associated table semantic. Default: `True`
-
-        `fake_score:` Will add a fake score so that annotations look like predictions
-
-        :return: dataflow
+        Args:
+            kwargs:
+                (split) Split of the dataset. Can be `train`, `val` or `test`. Default: `val`
+                (max_datapoints) Will stop iterating after `max_datapoints`. Default: `None`
+                (load_image) Will load the image for each datapoint. Default: `False`
+                (rows_and_cols) Will add 'item' image annotations that represent rows or columns of a
+                                table. Default: `True`
+                (fake_score) Will add a fake score so that annotations look like predictions. Default: `False`
+                (dd_pipe_like) If `True`, sets `load_image` to `True`. Default: `False`
+
+        Returns:
+            Dataflow
         """
         split = str(kwargs.get("split", "val"))
         if split == "val":

deepdoctection/datasets/instances/rvlcdip.py

@@ -123,16 +123,16 @@ class RvlcdipBuilder(DataFlowBaseBuilder):
 
     def build(self, **kwargs: Union[str, int]) -> DataFlow:
         """
-        Returns a dataflow from which you can stream datapoints of images. The following arguments affect the returns
-        of the dataflow:
+        Returns a dataflow from which you can stream datapoints of images.
 
-        `split:` Split of the dataset. Can be `train`, `val` or `test`. Default: `val`
+        Args:
+            kwargs:
+                split (str): Split of the dataset. Can be `train`, `val` or `test`. Default: `val`
+                max_datapoints (int): Will stop iterating after max_datapoints. Default: `None`
+                load_image (bool): Will load the image for each datapoint. Default: `False`
 
-        max_datapoints: Will stop iterating after max_datapoints. Default: `None`
-
-        load_image: Will load the image for each datapoint.  Default: `False`
-
-        :return: dataflow
+        Returns:
+            Dataflow
         """
 
         split = str(kwargs.get("split", "val"))

deepdoctection/datasets/instances/xfund.py

@@ -131,15 +131,17 @@ class XfundBuilder(DataFlowBaseBuilder):
         Returns a dataflow from which you can stream datapoints of images. The following arguments affect the returns
         of the dataflow:
 
-        `split:` Split of the dataset. `train` and `val` is available
-
-        `load_image:` Will load the image for each datapoint.  Default: `False`
-
-        `max_datapoints:` Will stop iterating after max_datapoints. Default: `None`
-
-        `languages:` Will select only samples of selected languages. Available languages: `de`, `es`, `fr`, `it`, `ja` ,
-                     `pt`, `zh`. If default will take any language.
-        :return: Dataflow
+        Args:
+            kwargs:
+                (split) Split of the dataset. `train` and `val` are available. Default: `val`
+                (load_image) Will load the image for each datapoint. Default: `False`
+                (max_datapoints) Will stop iterating after `max_datapoints`. Default: `None`
+                (languages) Will select only samples of selected languages. Available languages:
+                            `de`, `es`, `fr`, `it`, `ja`, `pt`, `zh`. If `None`, all
+                            languages are taken. Default: `None`
+
+        Returns:
+            Dataflow
         """
 
         split = str(kwargs.get("split", "val"))