deepdoctection 0.38__tar.gz → 0.39.1__tar.gz

{deepdoctection-0.38 → deepdoctection-0.39.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: deepdoctection
-Version: 0.38
+Version: 0.39.1
 Summary: Repository for Document AI
 Home-page: https://github.com/deepdoctection/deepdoctection
 Author: Dr. Janis Meyer
@@ -17,7 +17,7 @@ Requires-Python: >=3.9
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: catalogue==2.0.10
-Requires-Dist: huggingface_hub<0.26,>=0.12.0
+Requires-Dist: huggingface_hub>=0.26.0
 Requires-Dist: importlib-metadata>=5.0.0
 Requires-Dist: jsonlines==3.1.0
 Requires-Dist: lazy-imports==0.3.1
@@ -36,7 +36,7 @@ Requires-Dist: tabulate>=0.7.7
 Requires-Dist: tqdm==4.64.0
 Provides-Extra: tf
 Requires-Dist: catalogue==2.0.10; extra == "tf"
-Requires-Dist: huggingface_hub<0.26,>=0.12.0; extra == "tf"
+Requires-Dist: huggingface_hub>=0.26.0; extra == "tf"
 Requires-Dist: importlib-metadata>=5.0.0; extra == "tf"
 Requires-Dist: jsonlines==3.1.0; extra == "tf"
 Requires-Dist: lazy-imports==0.3.1; extra == "tf"
@@ -61,14 +61,14 @@ Requires-Dist: python-doctr==0.8.1; extra == "tf"
 Requires-Dist: pycocotools>=2.0.2; extra == "tf"
 Requires-Dist: boto3==1.34.102; extra == "tf"
 Requires-Dist: pdfplumber>=0.11.0; extra == "tf"
-Requires-Dist: fasttext==0.9.2; extra == "tf"
+Requires-Dist: fasttext-wheel; extra == "tf"
 Requires-Dist: jdeskew>=0.2.2; extra == "tf"
 Requires-Dist: apted==1.0.3; extra == "tf"
 Requires-Dist: distance==0.1.3; extra == "tf"
 Requires-Dist: lxml>=4.9.1; extra == "tf"
 Provides-Extra: pt
 Requires-Dist: catalogue==2.0.10; extra == "pt"
-Requires-Dist: huggingface_hub<0.26,>=0.12.0; extra == "pt"
+Requires-Dist: huggingface_hub>=0.26.0; extra == "pt"
 Requires-Dist: importlib-metadata>=5.0.0; extra == "pt"
 Requires-Dist: jsonlines==3.1.0; extra == "pt"
 Requires-Dist: lazy-imports==0.3.1; extra == "pt"
@@ -86,12 +86,12 @@ Requires-Dist: termcolor>=1.1; extra == "pt"
 Requires-Dist: tabulate>=0.7.7; extra == "pt"
 Requires-Dist: tqdm==4.64.0; extra == "pt"
 Requires-Dist: timm>=0.9.16; extra == "pt"
-Requires-Dist: transformers>=4.36.0; extra == "pt"
+Requires-Dist: transformers>=4.48.0; extra == "pt"
 Requires-Dist: accelerate>=0.29.1; extra == "pt"
 Requires-Dist: python-doctr==0.8.1; extra == "pt"
 Requires-Dist: boto3==1.34.102; extra == "pt"
 Requires-Dist: pdfplumber>=0.11.0; extra == "pt"
-Requires-Dist: fasttext==0.9.2; extra == "pt"
+Requires-Dist: fasttext-wheel; extra == "pt"
 Requires-Dist: jdeskew>=0.2.2; extra == "pt"
 Requires-Dist: apted==1.0.3; extra == "pt"
 Requires-Dist: distance==0.1.3; extra == "pt"
@@ -99,7 +99,7 @@ Requires-Dist: lxml>=4.9.1; extra == "pt"
 Provides-Extra: docs
 Requires-Dist: tensorpack==0.11; extra == "docs"
 Requires-Dist: boto3==1.34.102; extra == "docs"
-Requires-Dist: transformers>=4.36.0; extra == "docs"
+Requires-Dist: transformers>=4.48.0; extra == "docs"
 Requires-Dist: accelerate>=0.29.1; extra == "docs"
 Requires-Dist: pdfplumber>=0.11.0; extra == "docs"
 Requires-Dist: lxml>=4.9.1; extra == "docs"

{deepdoctection-0.38 → deepdoctection-0.39.1}/deepdoctection/__init__.py

@@ -6,6 +6,7 @@ Init file for deepdoctection package. This file is used to import all submodules
 """
 
 import importlib.util
+import os
 
 # Before doing anything else, check if the .env file exists and load it
 if importlib.util.find_spec("dotenv") is not None:
@@ -24,7 +25,7 @@ from .utils.logger import LoggingRecord, logger
 
 # pylint: enable=wrong-import-position
 
-__version__ = "0.38"
+__version__ = "0.39.1"
 
 _IMPORT_STRUCTURE = {
     "analyzer": ["config_sanity_checks", "get_dd_analyzer", "ServiceFactory"],
@@ -423,6 +424,9 @@ _IMPORT_STRUCTURE = {
 env_info = collect_env_info()
 logger.debug(LoggingRecord(msg=env_info))
 auto_select_pdf_render_framework()
+os.environ["DPI"] = "300"
+os.environ["IMAGE_WIDTH"] = ""
+os.environ["IMAGE_HEIGHT"] = ""
 
 # Direct imports for type-checking
 if TYPE_CHECKING:

{deepdoctection-0.38 → deepdoctection-0.39.1}/deepdoctection/analyzer/dd.py

@@ -32,7 +32,6 @@ from ..extern.pt.ptutils import get_torch_device
 from ..extern.tp.tfutils import disable_tp_layer_logging, get_tf_device
 from ..pipe.doctectionpipe import DoctectionPipe
 from ..utils.env_info import ENV_VARS_TRUE
-from ..utils.error import DependencyError
 from ..utils.file_utils import tensorpack_available
 from ..utils.fs import get_configs_dir_path, get_package_path, maybe_copy_config_to_cache
 from ..utils.logger import LoggingRecord, logger
@@ -118,13 +117,15 @@ def get_dd_analyzer(
     :return: A DoctectionPipe instance with given configs
     """
     config_overwrite = [] if config_overwrite is None else config_overwrite
-    lib = "TF" if os.environ.get("DD_USE_TF", "0") in ENV_VARS_TRUE else "PT"
-    if lib == "TF":
+    if os.environ.get("DD_USE_TF", "0") in ENV_VARS_TRUE:
+        lib = "TF"
         device = get_tf_device()
-    elif lib == "PT":
+    elif os.environ.get("DD_USE_TORCH", "0") in ENV_VARS_TRUE:
+        lib = "PT"
         device = get_torch_device()
     else:
-        raise DependencyError("At least one of the env variables DD_USE_TF or DD_USE_TORCH must be set.")
+        lib = None
+        device = None
     dd_one_config_path = maybe_copy_config_to_cache(
         get_package_path(), get_configs_dir_path() / "dd", _DD_ONE, reset_config_file
     )

{deepdoctection-0.38 → deepdoctection-0.39.1}/deepdoctection/analyzer/factory.py

@@ -48,6 +48,7 @@ from ..pipe.segment import PubtablesSegmentationService, TableSegmentationServic
 from ..pipe.sub_layout import DetectResultGenerator, SubImageLayoutService
 from ..pipe.text import TextExtractionService
 from ..pipe.transform import SimpleTransformService
+from ..utils.error import DependencyError
 from ..utils.file_utils import detectron2_available
 from ..utils.fs import get_configs_dir_path
 from ..utils.metacfg import AttrDict
@@ -62,8 +63,6 @@ __all__ = [
     "ServiceFactory",
 ]
 
-# from ._config import cfg
-
 
 class ServiceFactory:
     """
@@ -94,6 +93,8 @@ class ServiceFactory:
         :param config: configuration object
         :param mode: either `LAYOUT`,`CELL` or `ITEM`
         """
+        if config.LIB is None:
+            raise DependencyError("At least one of the env variables DD_USE_TF or DD_USE_TORCH must be set.")
         weights = (
             getattr(config.TF, mode).WEIGHTS
             if config.LIB == "TF"
@@ -310,6 +311,8 @@ class ServiceFactory:
                 config_overwrite=[f"LANGUAGES={config.LANGUAGE}"] if config.LANGUAGE is not None else None,
             )
         if config.OCR.USE_DOCTR:
+            if config.LIB is None:
+                raise DependencyError("At least one of the env variables DD_USE_TF or DD_USE_TORCH must be set.")
             weights = (
                 config.OCR.WEIGHTS.DOCTR_RECOGNITION.TF
                 if config.LIB == "TF"
@@ -353,6 +356,8 @@ class ServiceFactory:
         :param config: configuration object
         :return: DoctrTextlineDetector
         """
+        if config.LIB is None:
+            raise DependencyError("At least one of the env variables DD_USE_TF or DD_USE_TORCH must be set.")
         weights = config.OCR.WEIGHTS.DOCTR_WORD.TF if config.LIB == "TF" else config.OCR.WEIGHTS.DOCTR_WORD.PT
         weights_path = ModelDownloadManager.maybe_download_weights_and_configs(weights)
         profile = ModelCatalog.get_profile(weights)

{deepdoctection-0.38 → deepdoctection-0.39.1}/deepdoctection/datapoint/convert.py

@@ -154,7 +154,9 @@ def convert_pdf_bytes_to_np_array(pdf_bytes: bytes, dpi: Optional[int] = None) -
     return np_array.astype(uint8)
 
 
-def convert_pdf_bytes_to_np_array_v2(pdf_bytes: bytes, dpi: Optional[int] = 200) -> PixelValues:
+def convert_pdf_bytes_to_np_array_v2(
+    pdf_bytes: bytes, dpi: Optional[int] = None, width: Optional[int] = None, height: Optional[int] = None
+) -> PixelValues:
     """
     Converts a pdf passed as bytes into a numpy array. We use poppler or pdfmium to convert the pdf to an image.
     If both is available you can steer the selection of the render engine with environment variables:
@@ -165,17 +167,21 @@ def convert_pdf_bytes_to_np_array_v2(pdf_bytes: bytes, dpi: Optional[int] = 200)
     :param pdf_bytes: A pdf as bytes object. A byte representation can from a pdf file can be generated e.g. with
                       `utils.fs.load_bytes_from_pdf_file`
     :param dpi: The dpi value of the resulting output image. For high resolution set dpi=300.
+    :param width: The width of the resulting output image. This option does only work when using Poppler as
+    PDF renderer
+    :param height: The height of the resulting output image. This option does only work when using Poppler as
+    PDF renderer
     :return: Image as numpy array.
     """
 
-    with BytesIO(pdf_bytes) as pdf_file:
-        pdf = PdfReader(pdf_file).pages[0]
-    shape = pdf.mediabox  # pylint: disable=E1101
-    height = shape[3] - shape[1]
-    width = shape[2] - shape[0]
-
     if dpi is None:
-        return pdf_to_np_array(pdf_bytes, size=(int(width), int(height)))
+        if width is None or height is None:
+            with BytesIO(pdf_bytes) as pdf_file:
+                pdf = PdfReader(pdf_file).pages[0]
+            shape = pdf.mediabox  # pylint: disable=E1101
+            height = shape[3] - shape[1]
+            width = shape[2] - shape[0]
+        return pdf_to_np_array(pdf_bytes, size=(int(width), int(height)))  # type: ignore
     return pdf_to_np_array(pdf_bytes, dpi=dpi)
 
 

{deepdoctection-0.38 → deepdoctection-0.39.1}/deepdoctection/datapoint/image.py

@@ -153,7 +153,7 @@ class Image:
             self.set_width_height(self._image.shape[1], self._image.shape[0])
             self._self_embedding()
         elif isinstance(image, bytes):
-            self._image = convert_pdf_bytes_to_np_array_v2(image, dpi=environ.get("DPI", 300))  # type: ignore
+            self._image = convert_pdf_bytes_to_np_array_v2(image, dpi=int(environ["DPI"]))
             self.set_width_height(self._image.shape[1], self._image.shape[0])
             self._self_embedding()
         else:

{deepdoctection-0.38 → deepdoctection-0.39.1}/deepdoctection/datapoint/view.py

@@ -228,23 +228,33 @@ class Layout(ImageAnnotationBaseView):
 
         """
         words = self.get_ordered_words()
-        characters, ann_ids, token_classes, token_tags, token_classes_ids, token_tag_ids = zip(
-            *[
-                (
-                    word.characters,
-                    word.annotation_id,
-                    word.token_class,
-                    word.token_tag,
-                    word.get_sub_category(WordType.TOKEN_CLASS).category_id
-                    if WordType.TOKEN_CLASS in word.sub_categories
-                    else None,
-                    word.get_sub_category(WordType.TOKEN_TAG).category_id
-                    if WordType.TOKEN_TAG in word.sub_categories
-                    else None,
-                )
-                for word in words
-            ]
-        )
+        if words:
+            characters, ann_ids, token_classes, token_tags, token_classes_ids, token_tag_ids = zip(
+                *[
+                    (
+                        word.characters,
+                        word.annotation_id,
+                        word.token_class,
+                        word.token_tag,
+                        word.get_sub_category(WordType.TOKEN_CLASS).category_id
+                        if WordType.TOKEN_CLASS in word.sub_categories
+                        else None,
+                        word.get_sub_category(WordType.TOKEN_TAG).category_id
+                        if WordType.TOKEN_TAG in word.sub_categories
+                        else None,
+                    )
+                    for word in words
+                ]
+            )
+        else:
+            characters, ann_ids, token_classes, token_tags, token_classes_ids, token_tag_ids = (
+                [],  # type: ignore
+                [],  # type: ignore
+                [],  # type: ignore
+                [],  # type: ignore
+                [],  # type: ignore
+                [],  # type: ignore
+            )
         return {
             "text": " ".join(characters),
             "words": characters,
@@ -327,7 +337,7 @@ class Table(Layout):
         :return: A list of `Cell` objects that are row headers.
         """
         all_relation_ids = self.get_relationship(Relationships.CHILD)
-        all_cells: list[Cell] = self.base_page.get_annotation( # type: ignore
+        all_cells: list[Cell] = self.base_page.get_annotation(  # type: ignore
             category_names=[LayoutType.CELL, CellType.SPANNING], annotation_ids=all_relation_ids
         )
         row_header_cells = list(filter(lambda cell: CellType.ROW_HEADER in cell.sub_categories, all_cells))
@@ -363,18 +373,18 @@ class Table(Layout):
             category_names=[LayoutType.CELL, CellType.SPANNING], annotation_ids=all_relation_ids
         )
         row_cells = list(
-            filter(
-                lambda c: row_number in (c.row_number, c.row_number + c.row_span), all_cells  # type: ignore
-            )
+            filter(lambda c: row_number in (c.row_number, c.row_number + c.row_span), all_cells)  # type: ignore
         )
-        row_cells.sort(key=lambda c: c.column_number) # type: ignore
+        row_cells.sort(key=lambda c: c.column_number)  # type: ignore
         column_header_cells = self.column_header_cells
 
         kv_dict: Mapping[str, str] = {}
         for cell in row_cells:
             for header in column_header_cells:
-                if (cell.column_number == header.column_number and  # type: ignore
-                        cell.annotation_id != header.annotation_id):  # type: ignore
+                if (
+                    cell.column_number == header.column_number  # type: ignore
+                    and cell.annotation_id != header.annotation_id  # type: ignore
+                ):
                     kv_dict[(header.column_number, header.text)] = cell.text  # type: ignore
                     break
         return kv_dict

{deepdoctection-0.38 → deepdoctection-0.39.1}/deepdoctection/extern/model.py

@@ -24,7 +24,7 @@ from dataclasses import asdict, dataclass, field
 from typing import Any, Mapping, Optional, Union
 
 import jsonlines
-from huggingface_hub import cached_download, hf_hub_url  # type: ignore
+from huggingface_hub import hf_hub_download
 from tabulate import tabulate
 from termcolor import colored
 
@@ -136,51 +136,6 @@ class ModelCatalog:
             dl_library="TF",
             model_wrapper="TPFrcnnDetector",
         ),
-        "item/model-1620000.data-00000-of-00001": ModelProfile(
-            name="item/model-1620000.data-00000-of-00001",
-            description="Tensorpack row/column detection model trained on Pubtabnet",
-            config="dd/tp/conf_frcnn_rows.yaml",
-            size=[823546048, 25787],
-            tp_model=True,
-            hf_repo_id="deepdoctection/tp_casc_rcnn_X_32xd4_50_FPN_GN_2FC_pubtabnet_rc",
-            hf_model_name="model-1620000",
-            hf_config_file=["conf_frcnn_rows.yaml"],
-            categories={1: LayoutType.ROW, 2: LayoutType.COLUMN},
-            dl_library="TF",
-            model_wrapper="TPFrcnnDetector",
-        ),
-        "layout/model-800000.data-00000-of-00001": ModelProfile(
-            name="layout/model-800000.data-00000-of-00001",
-            description="Tensorpack layout detection model trained on Publaynet",
-            config="dd/tp/conf_frcnn_layout.yaml",
-            size=[823656748, 25796],
-            tp_model=True,
-            hf_repo_id="deepdoctection/tp_casc_rcnn_X_32xd4_50_FPN_GN_2FC_publaynet",
-            hf_model_name="model-800000",
-            hf_config_file=["conf_frcnn_layout.yaml"],
-            dl_library="TF",
-            categories={
-                1: LayoutType.TEXT,
-                2: LayoutType.TITLE,
-                3: LayoutType.LIST,
-                4: LayoutType.TABLE,
-                5: LayoutType.FIGURE,
-            },
-            model_wrapper="TPFrcnnDetector",
-        ),
-        "cell/model-1800000.data-00000-of-00001": ModelProfile(
-            name="cell/model-1800000.data-00000-of-00001",
-            description="Tensorpack cell detection model trained on Pubtabnet",
-            config="dd/tp/conf_frcnn_cell.yaml",
-            size=[823509160, 25905],
-            tp_model=True,
-            hf_repo_id="deepdoctection/tp_casc_rcnn_X_32xd4_50_FPN_GN_2FC_pubtabnet_c",
-            hf_model_name="model-1800000",
-            hf_config_file=["conf_frcnn_cell.yaml"],
-            categories={1: LayoutType.CELL},
-            dl_library="TF",
-            model_wrapper="TPFrcnnDetector",
-        ),
         "layout/d2_model_0829999_layout_inf_only.pt": ModelProfile(
             name="layout/d2_model_0829999_layout_inf_only.pt",
             description="Detectron2 layout detection model trained on Publaynet",
@@ -200,25 +155,6 @@ class ModelCatalog:
             dl_library="PT",
             model_wrapper="D2FrcnnDetector",
         ),
-        "layout/d2_model_0829999_layout.pth": ModelProfile(
-            name="layout/d2_model_0829999_layout.pth",
-            description="Detectron2 layout detection model trained on Publaynet. Checkpoint for resuming training",
-            config="dd/d2/layout/CASCADE_RCNN_R_50_FPN_GN.yaml",
-            size=[548377327],
-            tp_model=False,
-            hf_repo_id="deepdoctection/d2_casc_rcnn_X_32xd4_50_FPN_GN_2FC_publaynet_inference_only",
-            hf_model_name="d2_model_0829999_layout.pth",
-            hf_config_file=["Base-RCNN-FPN.yaml", "CASCADE_RCNN_R_50_FPN_GN.yaml"],
-            categories={
-                1: LayoutType.TEXT,
-                2: LayoutType.TITLE,
-                3: LayoutType.LIST,
-                4: LayoutType.TABLE,
-                5: LayoutType.FIGURE,
-            },
-            dl_library="PT",
-            model_wrapper="D2FrcnnDetector",
-        ),
         "layout/d2_model_0829999_layout_inf_only.ts": ModelProfile(
             name="layout/d2_model_0829999_layout_inf_only.ts",
             description="Detectron2 layout detection model trained on Publaynet. Torchscript export",
@@ -264,32 +200,6 @@ class ModelCatalog:
             dl_library="PT",
             model_wrapper="D2FrcnnTracingDetector",
         ),
-        "cell/d2_model_1849999_cell.pth": ModelProfile(
-            name="cell/d2_model_1849999_cell.pth",
-            description="Detectron2 cell detection inference only model trained on Pubtabnet",
-            config="dd/d2/cell/CASCADE_RCNN_R_50_FPN_GN.yaml",
-            size=[548279023],
-            tp_model=False,
-            hf_repo_id="deepdoctection/d2_casc_rcnn_X_32xd4_50_FPN_GN_2FC_pubtabnet_c_inference_only",
-            hf_model_name="cell/d2_model_1849999_cell.pth",
-            hf_config_file=["Base-RCNN-FPN.yaml", "CASCADE_RCNN_R_50_FPN_GN.yaml"],
-            categories={1: LayoutType.CELL},
-            dl_library="PT",
-            model_wrapper="D2FrcnnDetector",
-        ),
-        "item/d2_model_1639999_item.pth": ModelProfile(
-            name="item/d2_model_1639999_item.pth",
-            description="Detectron2 item detection model trained on Pubtabnet",
-            config="dd/d2/item/CASCADE_RCNN_R_50_FPN_GN.yaml",
-            size=[548303599],
-            tp_model=False,
-            hf_repo_id="deepdoctection/d2_casc_rcnn_X_32xd4_50_FPN_GN_2FC_pubtabnet_rc_inference_only",
-            hf_model_name="d2_model_1639999_item.pth",
-            hf_config_file=["Base-RCNN-FPN.yaml", "CASCADE_RCNN_R_50_FPN_GN.yaml"],
-            categories={1: LayoutType.ROW, 2: LayoutType.COLUMN},
-            dl_library="PT",
-            model_wrapper="D2FrcnnDetector",
-        ),
         "item/d2_model_1639999_item_inf_only.pt": ModelProfile(
             name="item/d2_model_1639999_item_inf_only.pt",
             description="Detectron2 item detection model inference only trained on Pubtabnet",
@@ -1232,20 +1142,19 @@ class ModelDownloadManager:
     def _load_from_hf_hub(
         repo_id: str, file_name: str, cache_directory: PathLikeOrStr, force_download: bool = False
     ) -> int:
-        url = hf_hub_url(repo_id=repo_id, filename=file_name)
         token = os.environ.get("HF_CREDENTIALS", None)
-        f_path = cached_download(
-            url,
-            cache_dir=cache_directory,
+        f_path = hf_hub_download(
+            repo_id,
+            file_name,
+            local_dir=cache_directory,  # type: ignore
             force_filename=file_name,
             force_download=force_download,
             token=token,
-            legacy_cache_layout=True,
         )
         if f_path:
             stat_info = os.stat(f_path)
             size = stat_info.st_size
 
-            assert size > 0, f"Downloaded an empty file from {url}!"
+            assert size > 0, f"Downloaded an empty file from {f_path}!"
             return size
         raise TypeError("Returned value from cached_download cannot be Null")

{deepdoctection-0.38 → deepdoctection-0.39.1}/deepdoctection/mapper/cats.py

@@ -73,18 +73,21 @@ def re_assign_cat_ids(
     Annotations that are not in the dictionary provided will be removed.
 
     :param dp: Image
-    :param categories_dict_name_as_key: e.g. `{LayoutType.word: '1'}`
+    :param categories_dict_name_as_key: e.g. `{LayoutType.word: 1}`
     :param cat_to_sub_cat_mapping: e.g. `{<LayoutType.word>:
         {<WordType.token_class>:
-            {<FundsFirstPage.report_date>: '1',
-            <FundsFirstPage.report_type>: '2',
-            <FundsFirstPage.umbrella>: '3',
-            <FundsFirstPage.fund_name>: '4',
-            <TokenClasses.other>: '5'},
-            <WordType.tag>:
-            {<BioTag.inside>: '1',
-            <BioTag.outside>: '2',
-            <BioTag.begin>: '3'}}}`
+            {<FundsFirstPage.REPORT_DATE>: 1,
+            <FundsFirstPage.REPORT_TYPE>: 2,
+            <FundsFirstPage.UMBRELLA>: 3,
+            <FundsFirstPage.FUND_NAME>: 4,
+            <TokenClasses.OTHER>: 5},
+            <WordType.TAG>:
+            {<BioTag.INSIDE>: 1,
+            <BioTag.OUTSIDE>: 2,
+            <BioTag.BEGIN>: 3}}}`
+            To re-assign the category ids of an image summary, use the key 'default_type' for the default category, e.g.
+            `{DefaultType.DEFAULT_TYPE: {<PageType.DOCUMENT_TYPE>: {<DocumentType.INVOICE>:1,
+            <DocumentType.BANK_STATEMENT>:2}}}`
     :return: Image
     """
 
@@ -104,6 +107,14 @@ def re_assign_cat_ids(
                     sub_category = ann.get_sub_category(key)
                     sub_category.category_id = sub_cat_values_dict.get(sub_category.category_name, DEFAULT_CATEGORY_ID)
 
+    if cat_to_sub_cat_mapping:
+        if "default_type" in cat_to_sub_cat_mapping:
+            sub_cat_keys_to_sub_cat_values = cat_to_sub_cat_mapping[get_type("default_type")]
+            for key in sub_cat_keys_to_sub_cat_values:
+                sub_cat_values_dict = sub_cat_keys_to_sub_cat_values[key]
+                sub_category = dp.summary.get_sub_category(key)
+                sub_category.category_id = sub_cat_values_dict.get(sub_category.category_name, DEFAULT_CATEGORY_ID)
+
     dp.remove(annotation_ids=ann_ids_to_remove)
 
     return dp

{deepdoctection-0.38 → deepdoctection-0.39.1}/deepdoctection/mapper/match.py

@@ -101,17 +101,6 @@ def match_anns_by_intersection(
         ]
     )
 
-    # second try, if ann has empty image
-    n_dim = child_ann_boxes.ndim
-    if n_dim != 2:
-        child_ann_boxes = np.array(
-            [
-                ann.bounding_box.transform(dp.width, dp.height, absolute_coords=True).to_list(mode="xyxy")
-                for ann in child_anns
-                if ann.bounding_box is not None
-            ]
-        )
-
     parent_anns = dp.get_annotation(annotation_ids=parent_ann_ids, category_names=parent_ann_category_names)
     parent_ann_boxes = np.array(
         [
@@ -120,17 +109,6 @@ def match_anns_by_intersection(
         ]
     )
 
-    # same for parent
-    n_dim = parent_ann_boxes.ndim
-    if n_dim != 2:
-        parent_ann_boxes = np.array(
-            [
-                ann.bounding_box.transform(dp.width, dp.height, absolute_coords=True).to_list(mode="xyxy")
-                for ann in parent_anns
-                if ann.bounding_box is not None
-            ]
-        )
-
     if matching_rule in ["iou"] and parent_anns and child_anns:
         iou_matrix = iou(child_ann_boxes, parent_ann_boxes)
         output = iou_matrix > threshold

{deepdoctection-0.38 → deepdoctection-0.39.1}/deepdoctection/mapper/misc.py

@@ -38,12 +38,20 @@ with try_import() as import_guard:
     from lxml import etree  # pylint: disable=W0611
 
 
-def to_image(dp: Union[str, Mapping[str, Union[str, bytes]]], dpi: Optional[int] = None) -> Optional[Image]:
+def to_image(
+    dp: Union[str, Mapping[str, Union[str, bytes]]],
+    dpi: Optional[int] = None,
+    width: Optional[int] = None,
+    height: Optional[int] = None,
+) -> Optional[Image]:
     """
     Mapping an input from `dataflow.SerializerFiles` or similar to an Image
 
     :param dp: Image
     :param dpi: dot per inch definition for pdf resolution when converting to numpy array
+    :param width: target width of the image. This option does only work when using Poppler as PDF renderer
+    :param height: target width of the image. This option does only work when using Poppler as PDF renderer
+    :param height: target height of the image
     :return: Image
     """
 
@@ -77,7 +85,9 @@ def to_image(dp: Union[str, Mapping[str, Union[str, bytes]]], dpi: Optional[int]
                 dp_image.pdf_bytes = dp.get("pdf_bytes")
                 if dp_image.pdf_bytes is not None:
                     if isinstance(dp_image.pdf_bytes, bytes):
-                        dp_image.image = convert_pdf_bytes_to_np_array_v2(dp_image.pdf_bytes, dpi=dpi)
+                        dp_image.image = convert_pdf_bytes_to_np_array_v2(
+                            dp_image.pdf_bytes, dpi=dpi, width=width, height=height
+                        )
             elif image_bytes is not None:
                 dp_image.image = convert_bytes_to_np_array(image_bytes)
             else:

{deepdoctection-0.38 → deepdoctection-0.39.1}/deepdoctection/mapper/pubstruct.py

@@ -393,7 +393,7 @@ def pub_to_image_uncur(  # pylint: disable=R0914
             np_image = load_image_from_file(dp["filename"])
         if is_file_extension(dp["filename"], ".pdf"):
             pdf_bytes = load_bytes_from_pdf_file(dp["filename"])
-            np_image = convert_pdf_bytes_to_np_array_v2(pdf_bytes)
+            np_image = convert_pdf_bytes_to_np_array_v2(pdf_bytes, dpi=200)
             dp = _convert_boxes(dp, np_image.shape[0])
 
         if load_image and np_image is not None: