deepdoctection 0.32__py3-none-any.whl → 0.34__py3-none-any.whl

deepdoctection/__init__.py

@@ -15,7 +15,6 @@ if importlib.util.find_spec("dotenv") is not None:
 
 
 # pylint: disable=wrong-import-position
-import os
 import sys
 from typing import TYPE_CHECKING
 
@@ -25,11 +24,10 @@ from .utils.logger import LoggingRecord, logger
 
 # pylint: enable=wrong-import-position
 
-__version__ = 0.32
+__version__ = 0.34
 
 _IMPORT_STRUCTURE = {
     "analyzer": [
-        "maybe_copy_config_to_cache",
         "config_sanity_checks",
         "build_detector",
         "build_padder",
@@ -76,6 +74,7 @@ _IMPORT_STRUCTURE = {
     ],
     "datapoint": [
         "ann_from_dict",
+        "AnnotationMap",
         "Annotation",
         "CategoryAnnotation",
         "ImageAnnotation",
@@ -160,6 +159,8 @@ _IMPORT_STRUCTURE = {
         "EvalCallback",
     ],
     "extern": [
+        "ModelCategories",
+        "NerModelCategories",
         "PredictorBase",
         "DetectionResult",
         "ObjectDetector",
@@ -235,6 +236,7 @@ _IMPORT_STRUCTURE = {
         "LabelSummarizer",
         "curry",
         "match_anns_by_intersection",
+        "match_anns_by_distance",
         "to_image",
         "maybe_load_image",
         "maybe_remove_image",
@@ -263,6 +265,8 @@ _IMPORT_STRUCTURE = {
         "DetectResultGenerator",
         "SubImageLayoutService",
         "ImageCroppingService",
+        "IntersectionMatcher",
+        "NeighbourMatcher",
         "MatchingService",
         "PageParsingService",
         "AnnotationNmsService",
@@ -362,6 +366,7 @@ _IMPORT_STRUCTURE = {
         "get_configs_dir_path",
         "get_weights_dir_path",
         "get_dataset_dir_path",
+        "maybe_copy_config_to_cache",
         "is_uuid_like",
         "get_uuid_from_str",
         "get_uuid",
@@ -423,28 +428,6 @@ _IMPORT_STRUCTURE = {
 env_info = collect_env_info()
 logger.debug(LoggingRecord(msg=env_info))
 
-if os.environ.get("PYTORCH_AVAILABLE") and os.environ.get("DD_USE_TORCH") is None:
-    os.environ["DD_USE_TORCH"] = "1"
-    os.environ["USE_TORCH"] = "1"
-if os.environ.get("TENSORFLOW_AVAILABLE") and os.environ.get("DD_USE_TF") is None:
-    os.environ["DD_USE_TF"] = "1"
-    os.environ["USE_TF"] = "1"
-if os.environ.get("DD_USE_TORCH") and os.environ.get("DD_USE_TF"):
-    logger.warning(
-        "Both DD_USE_TORCH and DD_USE_TF are set. Defaulting to PyTorch. If you want a different "
-        "behaviour, set DD_USE_TORCH to None before importing deepdoctection."
-    )
-    os.environ.pop("DD_USE_TF")
-    os.environ.pop("USE_TF")
-
-if not os.environ.get("PYTORCH_AVAILABLE") and not os.environ.get("TENSORFLOW_AVAILABLE"):
-    logger.warning(
-        LoggingRecord(
-            msg="Neither Tensorflow or Pytorch are available. You will not be able to use any Deep Learning "
-            "model from the library."
-        )
-    )
-
 
 # Direct imports for type-checking
 if TYPE_CHECKING:

deepdoctection/analyzer/dd.py

@@ -23,10 +23,11 @@ Module for **deep**doctection analyzer.
 -user factory with a reduced config setting
 """
 
+from __future__ import annotations
+
 import os
 from os import environ
-from shutil import copyfile
-from typing import List, Optional, Union
+from typing import Optional, Union
 
 from lazy_imports import try_import
 
@@ -42,7 +43,7 @@ from ..extern.texocr import TextractOcrDetector
 from ..extern.tp.tfutils import disable_tp_layer_logging, get_tf_device
 from ..extern.tpdetect import TPFrcnnDetector
 from ..pipe.base import PipelineComponent
-from ..pipe.common import AnnotationNmsService, MatchingService, PageParsingService
+from ..pipe.common import AnnotationNmsService, IntersectionMatcher, MatchingService, PageParsingService
 from ..pipe.doctectionpipe import DoctectionPipe
 from ..pipe.layout import ImageLayoutService
 from ..pipe.order import TextOrderService
@@ -50,21 +51,21 @@ from ..pipe.refine import TableSegmentationRefinementService
 from ..pipe.segment import PubtablesSegmentationService, TableSegmentationService
 from ..pipe.sub_layout import DetectResultGenerator, SubImageLayoutService
 from ..pipe.text import TextExtractionService
-from ..utils.detection_types import Pathlike
+from ..utils.env_info import ENV_VARS_TRUE
 from ..utils.error import DependencyError
 from ..utils.file_utils import detectron2_available, tensorpack_available
-from ..utils.fs import get_configs_dir_path, get_package_path, mkdir_p
+from ..utils.fs import get_configs_dir_path, get_package_path, maybe_copy_config_to_cache
 from ..utils.logger import LoggingRecord, logger
 from ..utils.metacfg import AttrDict, set_config_by_yaml
-from ..utils.settings import CellType, LayoutType
+from ..utils.settings import CellType, LayoutType, Relationships
 from ..utils.transform import PadTransform
+from ..utils.types import PathLikeOrStr
 
 with try_import() as image_guard:
     from botocore.config import Config  # type: ignore
 
 
 __all__ = [
-    "maybe_copy_config_to_cache",
     "config_sanity_checks",
     "build_detector",
     "build_padder",
@@ -74,31 +75,37 @@ __all__ = [
     "build_doctr_word",
     "get_dd_analyzer",
     "build_analyzer",
+    "set_config_by_yaml",
 ]
 
 _DD_ONE = "deepdoctection/configs/conf_dd_one.yaml"
 _TESSERACT = "deepdoctection/configs/conf_tesseract.yaml"
-
-
-def maybe_copy_config_to_cache(
-    package_path: Pathlike, configs_dir_path: Pathlike, file_name: str, force_copy: bool = True
-) -> str:
-    """
-    Initial copying of various files
-    :param package_path: base path to directory of source file `file_name`
-    :param configs_dir_path: base path to target directory
-    :param file_name: file to copy
-    :param force_copy: If file is already in target directory, will re-copy the file
-
-    :return: path to the copied file_name
-    """
-
-    absolute_path_source = os.path.join(package_path, file_name)
-    absolute_path = os.path.join(configs_dir_path, os.path.join("dd", os.path.split(file_name)[1]))
-    mkdir_p(os.path.split(absolute_path)[0])
-    if not os.path.isfile(absolute_path) or force_copy:
-        copyfile(absolute_path_source, absolute_path)
-    return absolute_path
+_MODEL_CHOICES = {
+    "layout": [
+        "layout/d2_model_0829999_layout_inf_only.pt",
+        "xrf_layout/model_final_inf_only.pt",
+        "microsoft/table-transformer-detection/pytorch_model.bin",
+    ],
+    "segmentation": [
+        "item/model-1620000_inf_only.data-00000-of-00001",
+        "xrf_item/model_final_inf_only.pt",
+        "microsoft/table-transformer-structure-recognition/pytorch_model.bin",
+        "deepdoctection/tatr_tab_struct_v2/pytorch_model.bin",
+    ],
+    "ocr": ["Tesseract", "DocTr", "Textract"],
+    "doctr_word": ["doctr/db_resnet50/pt/db_resnet50-ac60cadc.pt"],
+    "doctr_recognition": [
+        "doctr/crnn_vgg16_bn/pt/crnn_vgg16_bn-9762b0b0.pt",
+        "doctr/crnn_vgg16_bn/pt/pytorch_model.bin",
+    ],
+    "llm": ["gpt-3.5-turbo", "gpt-4"],
+    "segmentation_choices": {
+        "item/model-1620000_inf_only.data-00000-of-00001": "cell/model-1800000_inf_only.data-00000-of-00001",
+        "xrf_item/model_final_inf_only.pt": "xrf_cell/model_final_inf_only.pt",
+        "microsoft/table-transformer-structure-recognition/pytorch_model.bin": None,
+        "deepdoctection/tatr_tab_struct_v2/pytorch_model.bin": None,
+    },
+}
 
 
 def config_sanity_checks(cfg: AttrDict) -> None:
@@ -115,7 +122,7 @@ def config_sanity_checks(cfg: AttrDict) -> None:
 
 def build_detector(
     cfg: AttrDict, mode: str
-) -> Union["D2FrcnnDetector", "TPFrcnnDetector", "HFDetrDerivedDetector", "D2FrcnnTracingDetector"]:
+) -> Union[D2FrcnnDetector, TPFrcnnDetector, HFDetrDerivedDetector, D2FrcnnTracingDetector]:
     """Building a D2-Detector, a TP-Detector as Detr-Detector or a D2-Torch Tracing Detector according to
     the config
 
@@ -133,8 +140,8 @@ def build_detector(
     config_path = ModelCatalog.get_full_path_configs(weights)
     weights_path = ModelDownloadManager.maybe_download_weights_and_configs(weights)
     profile = ModelCatalog.get_profile(weights)
-    categories = profile.categories
-    assert categories is not None
+    categories = profile.categories if profile.categories is not None else {}
+
     if profile.model_wrapper in ("TPFrcnnDetector",):
         return TPFrcnnDetector(config_path, weights_path, categories, filter_categories=filter_categories)
     if profile.model_wrapper in ("D2FrcnnDetector",):
@@ -202,11 +209,13 @@ def build_sub_image_service(detector: ObjectDetector, cfg: AttrDict, mode: str)
     padder = None
     if mode == "ITEM":
         if detector.__class__.__name__ in ("HFDetrDerivedDetector",):
-            exclude_category_ids.extend(["1", "3", "4", "5", "6"])
+            exclude_category_ids.extend([1, 3, 4, 5, 6])
             padder = build_padder(cfg, mode)
-    detect_result_generator = DetectResultGenerator(detector.categories, exclude_category_ids=exclude_category_ids)
+    detect_result_generator = DetectResultGenerator(
+        categories=detector.categories.categories, exclude_category_ids=exclude_category_ids
+    )
     return SubImageLayoutService(
-        detector, [LayoutType.table, LayoutType.table_rotated], None, detect_result_generator, padder
+        detector, [LayoutType.TABLE, LayoutType.TABLE_ROTATED], None, detect_result_generator, padder
     )
 
 
@@ -233,9 +242,9 @@ def build_ocr(cfg: AttrDict) -> Union[TesseractOcrDetector, DoctrTextRecognizer,
         )
     if cfg.OCR.USE_TEXTRACT:
         credentials_kwargs = {
-            "aws_access_key_id": environ.get("ACCESS_KEY"),
-            "aws_secret_access_key": environ.get("SECRET_KEY"),
-            "config": Config(region_name=environ.get("REGION")),
+            "aws_access_key_id": environ.get("ACCESS_KEY", None),
+            "aws_secret_access_key": environ.get("SECRET_KEY", None),
+            "config": Config(region_name=environ.get("REGION", None)),
         }
         return TextractOcrDetector(**credentials_kwargs)
     raise ValueError("You have set USE_OCR=True but any of USE_TESSERACT, USE_DOCTR, USE_TEXTRACT is set to False")
@@ -260,7 +269,7 @@ def build_analyzer(cfg: AttrDict) -> DoctectionPipe:
     :param cfg: A configuration
     :return: Analyzer pipeline
     """
-    pipe_component_list: List[PipelineComponent] = []
+    pipe_component_list: list[PipelineComponent] = []
 
     if cfg.USE_LAYOUT:
         d_layout = build_detector(cfg, "LAYOUT")
@@ -300,22 +309,22 @@ def build_analyzer(cfg: AttrDict) -> DoctectionPipe:
                 cfg.SEGMENTATION.REMOVE_IOU_THRESHOLD_ROWS,
                 cfg.SEGMENTATION.REMOVE_IOU_THRESHOLD_COLS,
                 cfg.SEGMENTATION.CELL_CATEGORY_ID,
-                LayoutType.table,
+                LayoutType.TABLE,
                 [
-                    CellType.spanning,
-                    CellType.row_header,
-                    CellType.column_header,
-                    CellType.projected_row_header,
-                    LayoutType.cell,
+                    CellType.SPANNING,
+                    CellType.ROW_HEADER,
+                    CellType.COLUMN_HEADER,
+                    CellType.PROJECTED_ROW_HEADER,
+                    LayoutType.CELL,
                 ],
                 [
-                    CellType.spanning,
-                    CellType.row_header,
-                    CellType.column_header,
-                    CellType.projected_row_header,
+                    CellType.SPANNING,
+                    CellType.ROW_HEADER,
+                    CellType.COLUMN_HEADER,
+                    CellType.PROJECTED_ROW_HEADER,
                 ],
-                [LayoutType.row, LayoutType.column],
-                [CellType.row_number, CellType.column_number],
+                [LayoutType.ROW, LayoutType.COLUMN],
+                [CellType.ROW_NUMBER, CellType.COLUMN_NUMBER],
                 stretch_rule=cfg.SEGMENTATION.STRETCH_RULE,
             )
             pipe_component_list.append(pubtables)
@@ -327,23 +336,23 @@ def build_analyzer(cfg: AttrDict) -> DoctectionPipe:
                 cfg.SEGMENTATION.FULL_TABLE_TILING,
                 cfg.SEGMENTATION.REMOVE_IOU_THRESHOLD_ROWS,
                 cfg.SEGMENTATION.REMOVE_IOU_THRESHOLD_COLS,
-                LayoutType.table,
-                [CellType.header, CellType.body, LayoutType.cell],
-                [LayoutType.row, LayoutType.column],
-                [CellType.row_number, CellType.column_number],
+                LayoutType.TABLE,
+                [CellType.HEADER, CellType.BODY, LayoutType.CELL],
+                [LayoutType.ROW, LayoutType.COLUMN],
+                [CellType.ROW_NUMBER, CellType.COLUMN_NUMBER],
                 cfg.SEGMENTATION.STRETCH_RULE,
             )
             pipe_component_list.append(table_segmentation)
 
             if cfg.USE_TABLE_REFINEMENT:
                 table_segmentation_refinement = TableSegmentationRefinementService(
-                    [LayoutType.table, LayoutType.table_rotated],
+                    [LayoutType.TABLE, LayoutType.TABLE_ROTATED],
                     [
-                        LayoutType.cell,
-                        CellType.column_header,
-                        CellType.projected_row_header,
-                        CellType.spanning,
-                        CellType.row_header,
+                        LayoutType.CELL,
+                        CellType.COLUMN_HEADER,
+                        CellType.PROJECTED_ROW_HEADER,
+                        CellType.SPANNING,
+                        CellType.ROW_HEADER,
                     ],
                 )
                 pipe_component_list.append(table_segmentation_refinement)
@@ -363,24 +372,28 @@ def build_analyzer(cfg: AttrDict) -> DoctectionPipe:
 
         ocr = build_ocr(cfg)
         skip_if_text_extracted = cfg.USE_PDF_MINER
-        extract_from_roi = LayoutType.word if cfg.OCR.USE_DOCTR else None
+        extract_from_roi = LayoutType.WORD if cfg.OCR.USE_DOCTR else None
         text = TextExtractionService(
             ocr, skip_if_text_extracted=skip_if_text_extracted, extract_from_roi=extract_from_roi
         )
         pipe_component_list.append(text)
 
     if cfg.USE_PDF_MINER or cfg.USE_OCR:
-        match = MatchingService(
-            parent_categories=cfg.WORD_MATCHING.PARENTAL_CATEGORIES,
-            child_categories=LayoutType.word,
+        matcher = IntersectionMatcher(
             matching_rule=cfg.WORD_MATCHING.RULE,
             threshold=cfg.WORD_MATCHING.THRESHOLD,
             max_parent_only=cfg.WORD_MATCHING.MAX_PARENT_ONLY,
         )
+        match = MatchingService(
+            parent_categories=cfg.WORD_MATCHING.PARENTAL_CATEGORIES,
+            child_categories=LayoutType.WORD,
+            matcher=matcher,
+            relationship_key=Relationships.CHILD,
+        )
         pipe_component_list.append(match)
 
         order = TextOrderService(
-            text_container=LayoutType.word,
+            text_container=LayoutType.WORD,
             text_block_categories=cfg.TEXT_ORDERING.TEXT_BLOCK_CATEGORIES,
             floating_text_block_categories=cfg.TEXT_ORDERING.FLOATING_TEXT_BLOCK_CATEGORIES,
             include_residual_text_container=cfg.TEXT_ORDERING.INCLUDE_RESIDUAL_TEXT_CONTAINER,
@@ -392,7 +405,7 @@ def build_analyzer(cfg: AttrDict) -> DoctectionPipe:
         pipe_component_list.append(order)
 
     page_parsing_service = PageParsingService(
-        text_container=LayoutType.word,
+        text_container=LayoutType.WORD,
         floating_text_block_categories=cfg.TEXT_ORDERING.FLOATING_TEXT_BLOCK_CATEGORIES,
         include_residual_text_container=cfg.TEXT_ORDERING.INCLUDE_RESIDUAL_TEXT_CONTAINER,
     )
@@ -403,8 +416,8 @@ def build_analyzer(cfg: AttrDict) -> DoctectionPipe:
 
 def get_dd_analyzer(
     reset_config_file: bool = True,
-    config_overwrite: Optional[List[str]] = None,
-    path_config_file: Optional[Pathlike] = None,
+    config_overwrite: Optional[list[str]] = None,
+    path_config_file: Optional[PathLikeOrStr] = None,
 ) -> DoctectionPipe:
     """
     Factory function for creating the built-in **deep**doctection analyzer.
@@ -431,7 +444,7 @@ 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") else "PT"
+    lib = "TF" if os.environ.get("DD_USE_TF", "0") in ENV_VARS_TRUE else "PT"
     if lib == "TF":
         device = get_tf_device()
     elif lib == "PT":
@@ -439,9 +452,9 @@ def get_dd_analyzer(
     else:
         raise DependencyError("At least one of the env variables DD_USE_TF or DD_USE_TORCH must be set.")
     dd_one_config_path = maybe_copy_config_to_cache(
-        get_package_path(), get_configs_dir_path(), _DD_ONE, reset_config_file
+        get_package_path(), get_configs_dir_path() / "dd", _DD_ONE, reset_config_file
     )
-    maybe_copy_config_to_cache(get_package_path(), get_configs_dir_path(), _TESSERACT)
+    maybe_copy_config_to_cache(get_package_path(), get_configs_dir_path() / "dd", _TESSERACT)
 
     # Set up of the configuration and logging
     cfg = set_config_by_yaml(dd_one_config_path if not path_config_file else path_config_file)

deepdoctection/dataflow/common.py

@@ -12,7 +12,7 @@ Some DataFlow classes for transforming and processing datapoints. Many classes h
 """
 import itertools
 from copy import copy
-from typing import Any, Callable, Iterator, List, Union
+from typing import Any, Callable, Iterator, Union
 
 import tqdm
 
@@ -164,6 +164,10 @@ class RepeatedData(ProxyDataFlow):
                 Set to -1 to repeat ``ds`` infinite times.
         """
         self.num = num
+        if self.num != -1:
+            self.dfs = itertools.tee(df, self.num)
+        else:
+            self.dfs = ()
         super().__init__(df)
 
     def __len__(self) -> int:
@@ -180,8 +184,8 @@ class RepeatedData(ProxyDataFlow):
             while True:
                 yield from self.df
         else:
-            for _ in range(self.num):
-                yield from self.df
+            for df in self.dfs:
+                yield from df
 
 
 class ConcatData(DataFlow):
@@ -197,7 +201,7 @@ class ConcatData(DataFlow):
            df = ConcatData([df_1,df_2])
     """
 
-    def __init__(self, df_lists: List[DataFlow]) -> None:
+    def __init__(self, df_lists: list[DataFlow]) -> None:
         """
         :param df_lists: a list of DataFlow.
         """
@@ -233,7 +237,7 @@ class JoinData(DataFlow):
     `JoinData` will stop once the first Dataflow throws a StopIteration
     """
 
-    def __init__(self, df_lists: List[DataFlow]) -> None:
+    def __init__(self, df_lists: list[DataFlow]) -> None:
         """
         :param df_lists: a list of DataFlow. When these dataflows have different sizes, JoinData will stop when any
                         of them is exhausted.

deepdoctection/dataflow/custom.py

@@ -21,7 +21,7 @@ from
 
 <https://github.com/tensorpack/dataflow/blob/master/dataflow/dataflow/common.py>
 """
-from typing import Any, Callable, Iterable, Iterator, List, Optional
+from typing import Any, Callable, Iterable, Iterator, Optional
 
 import numpy as np
 
@@ -54,7 +54,7 @@ class CacheData(ProxyDataFlow):
         :param shuffle: whether to shuffle the cache before yielding from it.
         """
         self.shuffle = shuffle
-        self.buffer: List[Any] = []
+        self.buffer: list[Any] = []
         self._guard: Optional[DataFlowReentrantGuard] = None
         self.rng = get_rng(self)
         super().__init__(df)
@@ -78,7 +78,7 @@ class CacheData(ProxyDataFlow):
                     yield dp
                     self.buffer.append(dp)
 
-    def get_cache(self) -> List[Any]:
+    def get_cache(self) -> list[Any]:
         """
         get the cache of the whole dataflow as a list
 
@@ -115,10 +115,10 @@ class CustomDataFromList(DataFromList):
 
     def __init__(
         self,
-        lst: List[Any],
+        lst: list[Any],
         shuffle: bool = False,
         max_datapoints: Optional[int] = None,
-        rebalance_func: Optional[Callable[[List[Any]], List[Any]]] = None,
+        rebalance_func: Optional[Callable[[list[Any]], list[Any]]] = None,
     ):
         """
         :param lst: the input list. Each element represents a datapoint.