docling 2.2.1__tar.gz → 2.3.1__tar.gz

{docling-2.2.1 → docling-2.3.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: docling
-Version: 2.2.1
+Version: 2.3.1
 Summary: Docling PDF conversion package
 Home-page: https://github.com/DS4SD/docling
 License: MIT
@@ -23,9 +23,9 @@ Provides-Extra: tesserocr
 Requires-Dist: beautifulsoup4 (>=4.12.3,<5.0.0)
 Requires-Dist: certifi (>=2024.7.4)
 Requires-Dist: deepsearch-glm (>=0.26.1,<0.27.0)
-Requires-Dist: docling-core (>=2.2.1,<3.0.0)
-Requires-Dist: docling-ibm-models (>=2.0.1,<3.0.0)
-Requires-Dist: docling-parse (>=2.0.0,<3.0.0)
+Requires-Dist: docling-core (>=2.3.0,<3.0.0)
+Requires-Dist: docling-ibm-models (>=2.0.3,<3.0.0)
+Requires-Dist: docling-parse (>=2.0.2,<3.0.0)
 Requires-Dist: easyocr (>=1.7,<2.0)
 Requires-Dist: filetype (>=1.2.0,<2.0.0)
 Requires-Dist: huggingface_hub (>=0.23,<1)
@@ -41,10 +41,6 @@ Requires-Dist: requests (>=2.32.3,<3.0.0)
 Requires-Dist: rtree (>=1.3.0,<2.0.0)
 Requires-Dist: scipy (>=1.14.1,<2.0.0)
 Requires-Dist: tesserocr (>=2.7.1,<3.0.0) ; extra == "tesserocr"
-Requires-Dist: torch (>=2.2.2,<2.3.0) ; sys_platform == "darwin" and platform_machine == "x86_64"
-Requires-Dist: torch (>=2.2.2,<3.0.0) ; sys_platform != "darwin" or platform_machine != "x86_64"
-Requires-Dist: torchvision (>=0,<1) ; sys_platform != "darwin" or platform_machine != "x86_64"
-Requires-Dist: torchvision (>=0.17.2,<0.18.0) ; sys_platform == "darwin" and platform_machine == "x86_64"
 Requires-Dist: typer (>=0.12.5,<0.13.0)
 Project-URL: Repository, https://github.com/DS4SD/docling
 Description-Content-Type: text/markdown
@@ -73,8 +69,9 @@ Docling parses documents and exports them to the desired format with ease and sp
 
 ## Features
 
-* 🗂️ Multi-format support for input (PDF, DOCX etc.) & output (Markdown, JSON etc.)
-* 📑 Advanced PDF document understanding incl. page layout, reading order & table structures
+* 🗂️ Reads popular document formats (PDF, DOCX, PPTX, Images, HTML, AsciiDoc, Markdown) and exports to Markdown and JSON
+* 📑 Advanced PDF document understanding including page layout, reading order & table structures
+* 🧩 Unified, expressive [DoclingDocument](https://ds4sd.github.io/docling/concepts/docling_document/) representation format
 * 📝 Metadata extraction, including title, authors, references & language
 * 🤖 Seamless LlamaIndex 🦙 & LangChain 🦜🔗 integration for powerful RAG / QA applications
 * 🔍 OCR support for scanned PDFs

{docling-2.2.1 → docling-2.3.1}/README.md

@@ -22,8 +22,9 @@ Docling parses documents and exports them to the desired format with ease and sp
 
 ## Features
 
-* 🗂️ Multi-format support for input (PDF, DOCX etc.) & output (Markdown, JSON etc.)
-* 📑 Advanced PDF document understanding incl. page layout, reading order & table structures
+* 🗂️ Reads popular document formats (PDF, DOCX, PPTX, Images, HTML, AsciiDoc, Markdown) and exports to Markdown and JSON
+* 📑 Advanced PDF document understanding including page layout, reading order & table structures
+* 🧩 Unified, expressive [DoclingDocument](https://ds4sd.github.io/docling/concepts/docling_document/) representation format
 * 📝 Metadata extraction, including title, authors, references & language
 * 🤖 Seamless LlamaIndex 🦙 & LangChain 🦜🔗 integration for powerful RAG / QA applications
 * 🔍 OCR support for scanned PDFs

{docling-2.2.1 → docling-2.3.1}/docling/backend/asciidoc_backend.py

@@ -1,24 +1,20 @@
 import logging
-import os
 import re
 from io import BytesIO
 from pathlib import Path
 from typing import Set, Union
 
 from docling_core.types.doc import (
-    DocItem,
     DocItemLabel,
     DoclingDocument,
     DocumentOrigin,
     GroupItem,
     GroupLabel,
     ImageRef,
-    NodeItem,
     Size,
     TableCell,
     TableData,
 )
-from pydantic import AnyUrl
 
 from docling.backend.abstract_backend import DeclarativeDocumentBackend
 from docling.datamodel.base_models import InputFormat

{docling-2.2.1 → docling-2.3.1}/docling/backend/html_backend.py

@@ -179,31 +179,31 @@ class HTMLDocumentBackend(DeclarativeDocumentBackend):
             self.parents[self.level] = doc.add_text(
                 parent=self.parents[0], label=DocItemLabel.TITLE, text=text
             )
-
-        elif hlevel > self.level:
-
-            # add invisible group
-            for i in range(self.level + 1, hlevel):
-                self.parents[i] = doc.add_group(
-                    name=f"header-{i}",
-                    label=GroupLabel.SECTION,
-                    parent=self.parents[i - 1],
-                )
-            self.level = hlevel
-
-        elif hlevel < self.level:
-
-            # remove the tail
-            for key, val in self.parents.items():
-                if key > hlevel:
-                    self.parents[key] = None
-            self.level = hlevel
-
-        self.parents[hlevel] = doc.add_heading(
-            parent=self.parents[hlevel - 1],
-            text=text,
-            level=hlevel,
-        )
+        else:
+            if hlevel > self.level:
+
+                # add invisible group
+                for i in range(self.level + 1, hlevel):
+                    self.parents[i] = doc.add_group(
+                        name=f"header-{i}",
+                        label=GroupLabel.SECTION,
+                        parent=self.parents[i - 1],
+                    )
+                self.level = hlevel
+
+            elif hlevel < self.level:
+
+                # remove the tail
+                for key, val in self.parents.items():
+                    if key > hlevel:
+                        self.parents[key] = None
+                self.level = hlevel
+
+            self.parents[hlevel] = doc.add_heading(
+                parent=self.parents[hlevel - 1],
+                text=text,
+                level=hlevel,
+            )
 
     def handle_paragraph(self, element, idx, doc):
         """Handles paragraph tags (p)."""

{docling-2.2.1 → docling-2.3.1}/docling/datamodel/base_models.py

@@ -1,6 +1,6 @@
 from enum import Enum, auto
 from io import BytesIO
-from typing import TYPE_CHECKING, Dict, List, Optional, Set, Union
+from typing import TYPE_CHECKING, Dict, List, Optional, Union
 
 from docling_core.types.doc import (
     BoundingBox,

{docling-2.2.1 → docling-2.3.1}/docling/datamodel/document.py

@@ -3,7 +3,7 @@ import re
 from enum import Enum
 from io import BytesIO
 from pathlib import Path, PurePath
-from typing import TYPE_CHECKING, Dict, Iterable, List, Optional, Tuple, Type, Union
+from typing import TYPE_CHECKING, Dict, Iterable, List, Optional, Type, Union
 
 import filetype
 from docling_core.types.doc import (
@@ -52,6 +52,7 @@ from docling.datamodel.base_models import (
     Page,
 )
 from docling.datamodel.settings import DocumentLimits
+from docling.utils.profiling import ProfilingItem
 from docling.utils.utils import create_file_hash, create_hash
 
 if TYPE_CHECKING:
@@ -187,6 +188,7 @@ class ConversionResult(BaseModel):
 
     pages: List[Page] = []
     assembled: AssembledUnit = AssembledUnit()
+    timings: Dict[str, ProfilingItem] = {}
 
     document: DoclingDocument = _EMPTY_DOCLING_DOC
 

{docling-2.2.1 → docling-2.3.1}/docling/datamodel/settings.py

@@ -1,4 +1,5 @@
 import sys
+from pathlib import Path
 
 from pydantic import BaseModel
 from pydantic_settings import BaseSettings
@@ -26,8 +27,21 @@ class BatchConcurrencySettings(BaseModel):
     # To force models into single core: export OMP_NUM_THREADS=1
 
 
+class DebugSettings(BaseModel):
+    visualize_cells: bool = False
+    visualize_ocr: bool = False
+    visualize_layout: bool = False
+    visualize_tables: bool = False
+
+    profile_pipeline_timings: bool = False
+
+    # Path used to output debug information.
+    debug_output_path: str = str(Path.cwd() / "debug")
+
+
 class AppSettings(BaseSettings):
     perf: BatchConcurrencySettings
+    debug: DebugSettings
 
 
-settings = AppSettings(perf=BatchConcurrencySettings())
+settings = AppSettings(perf=BatchConcurrencySettings(), debug=DebugSettings())

{docling-2.2.1 → docling-2.3.1}/docling/document_converter.py

@@ -139,6 +139,10 @@ class DocumentConverter:
 
         self.initialized_pipelines: Dict[Type[BasePipeline], BasePipeline] = {}
 
+    def initialize_pipeline(self, format: InputFormat):
+        """Initialize the conversion pipeline for the selected format."""
+        self._get_pipeline(doc_format=format)
+
     @validate_call(config=ConfigDict(strict=True))
     def convert(
         self,
@@ -189,32 +193,43 @@ class DocumentConverter:
     ) -> Iterator[ConversionResult]:
         assert self.format_to_options is not None
 
+        start_time = time.monotonic()
+
         for input_batch in chunkify(
             conv_input.docs(self.format_to_options),
             settings.perf.doc_batch_size,  # pass format_options
         ):
             _log.info(f"Going to convert document batch...")
+
             # parallel processing only within input_batch
             # with ThreadPoolExecutor(
             #    max_workers=settings.perf.doc_batch_concurrency
             # ) as pool:
             #   yield from pool.map(self.process_document, input_batch)
-
             # Note: PDF backends are not thread-safe, thread pool usage was disabled.
+
             for item in map(
                 partial(self._process_document, raises_on_error=raises_on_error),
                 input_batch,
             ):
+                elapsed = time.monotonic() - start_time
+                start_time = time.monotonic()
+
                 if item is not None:
+                    _log.info(
+                        f"Finished converting document {item.input.file.name} in {elapsed:.2f} sec."
+                    )
                     yield item
+                else:
+                    _log.info(f"Skipped a document. We lost {elapsed:.2f} sec.")
 
-    def _get_pipeline(self, doc: InputDocument) -> Optional[BasePipeline]:
+    def _get_pipeline(self, doc_format: InputFormat) -> Optional[BasePipeline]:
         assert self.format_to_options is not None
 
-        fopt = self.format_to_options.get(doc.format)
+        fopt = self.format_to_options.get(doc_format)
 
         if fopt is None:
-            raise RuntimeError(f"Could not get pipeline for document {doc.file}")
+            raise RuntimeError(f"Could not get pipeline for {doc_format}")
         else:
             pipeline_class = fopt.pipeline_cls
             pipeline_options = fopt.pipeline_options
@@ -237,22 +252,15 @@ class DocumentConverter:
         assert self.allowed_formats is not None
         assert in_doc.format in self.allowed_formats
 
-        start_doc_time = time.time()
-
         conv_res = self._execute_pipeline(in_doc, raises_on_error=raises_on_error)
 
-        end_doc_time = time.time() - start_doc_time
-        _log.info(
-            f"Finished converting document {in_doc.file.name} in {end_doc_time:.2f} seconds."
-        )
-
         return conv_res
 
     def _execute_pipeline(
         self, in_doc: InputDocument, raises_on_error: bool
     ) -> ConversionResult:
         if in_doc.valid:
-            pipeline = self._get_pipeline(in_doc)
+            pipeline = self._get_pipeline(in_doc.format)
             if pipeline is None:  # Can't find a default pipeline. Should this raise?
                 if raises_on_error:
                     raise RuntimeError(

{docling-2.2.1 → docling-2.3.1}/docling/models/base_model.py

@@ -4,11 +4,14 @@ from typing import Any, Iterable
 from docling_core.types.doc import DoclingDocument, NodeItem
 
 from docling.datamodel.base_models import Page
+from docling.datamodel.document import ConversionResult
 
 
 class BasePageModel(ABC):
     @abstractmethod
-    def __call__(self, page_batch: Iterable[Page]) -> Iterable[Page]:
+    def __call__(
+        self, conv_res: ConversionResult, page_batch: Iterable[Page]
+    ) -> Iterable[Page]:
         pass
 
 

{docling-2.2.1 → docling-2.3.1}/docling/models/base_ocr_model.py

@@ -1,6 +1,7 @@
 import copy
 import logging
 from abc import abstractmethod
+from pathlib import Path
 from typing import Iterable, List
 
 import numpy as np
@@ -10,12 +11,15 @@ from rtree import index
 from scipy.ndimage import find_objects, label
 
 from docling.datamodel.base_models import OcrCell, Page
+from docling.datamodel.document import ConversionResult
 from docling.datamodel.pipeline_options import OcrOptions
+from docling.datamodel.settings import settings
+from docling.models.base_model import BasePageModel
 
 _log = logging.getLogger(__name__)
 
 
-class BaseOcrModel:
+class BaseOcrModel(BasePageModel):
     def __init__(self, enabled: bool, options: OcrOptions):
         self.enabled = enabled
         self.options = options
@@ -113,7 +117,7 @@ class BaseOcrModel:
         ]
         return filtered_ocr_cells
 
-    def draw_ocr_rects_and_cells(self, page, ocr_rects):
+    def draw_ocr_rects_and_cells(self, conv_res, page, ocr_rects, show: bool = False):
         image = copy.deepcopy(page.image)
         draw = ImageDraw.Draw(image, "RGBA")
 
@@ -130,8 +134,21 @@ class BaseOcrModel:
             if isinstance(tc, OcrCell):
                 color = "magenta"
             draw.rectangle([(x0, y0), (x1, y1)], outline=color)
-        image.show()
+
+        if show:
+            image.show()
+        else:
+            out_path: Path = (
+                Path(settings.debug.debug_output_path)
+                / f"debug_{conv_res.input.file.stem}"
+            )
+            out_path.mkdir(parents=True, exist_ok=True)
+
+            out_file = out_path / f"ocr_page_{page.page_no:05}.png"
+            image.save(str(out_file), format="png")
 
     @abstractmethod
-    def __call__(self, page_batch: Iterable[Page]) -> Iterable[Page]:
+    def __call__(
+        self, conv_res: ConversionResult, page_batch: Iterable[Page]
+    ) -> Iterable[Page]:
         pass

{docling-2.2.1 → docling-2.3.1}/docling/models/ds_glm_model.py

@@ -1,5 +1,6 @@
 import copy
 import random
+from pathlib import Path
 from typing import List, Union
 
 from deepsearch_glm.nlp_utils import init_nlp_model
@@ -27,6 +28,8 @@ from pydantic import BaseModel, ConfigDict
 
 from docling.datamodel.base_models import Cluster, FigureElement, Table, TextElement
 from docling.datamodel.document import ConversionResult, layout_label_to_ds_type
+from docling.datamodel.settings import settings
+from docling.utils.profiling import ProfilingScope, TimeRecorder
 from docling.utils.utils import create_hash
 
 
@@ -226,23 +229,24 @@ class GlmModel:
         return ds_doc
 
     def __call__(self, conv_res: ConversionResult) -> DoclingDocument:
-        ds_doc = self._to_legacy_document(conv_res)
-        ds_doc_dict = ds_doc.model_dump(by_alias=True)
+        with TimeRecorder(conv_res, "glm", scope=ProfilingScope.DOCUMENT):
+            ds_doc = self._to_legacy_document(conv_res)
+            ds_doc_dict = ds_doc.model_dump(by_alias=True)
 
-        glm_doc = self.model.apply_on_doc(ds_doc_dict)
+            glm_doc = self.model.apply_on_doc(ds_doc_dict)
 
-        docling_doc: DoclingDocument = to_docling_document(glm_doc)  # Experimental
+            docling_doc: DoclingDocument = to_docling_document(glm_doc)  # Experimental
 
         # DEBUG code:
-        def draw_clusters_and_cells(ds_document, page_no):
+        def draw_clusters_and_cells(ds_document, page_no, show: bool = False):
             clusters_to_draw = []
             image = copy.deepcopy(conv_res.pages[page_no].image)
             for ix, elem in enumerate(ds_document.main_text):
                 if isinstance(elem, BaseText):
-                    prov = elem.prov[0]
+                    prov = elem.prov[0]  # type: ignore
                 elif isinstance(elem, Ref):
                     _, arr, index = elem.ref.split("/")
-                    index = int(index)
+                    index = int(index)  # type: ignore
                     if arr == "tables":
                         prov = ds_document.tables[index].prov[0]
                     elif arr == "figures":
@@ -256,7 +260,7 @@ class GlmModel:
                             id=ix,
                             label=elem.name,
                             bbox=BoundingBox.from_tuple(
-                                coord=prov.bbox,
+                                coord=prov.bbox,  # type: ignore
                                 origin=CoordOrigin.BOTTOMLEFT,
                             ).to_top_left_origin(conv_res.pages[page_no].size.height),
                         )
@@ -276,9 +280,21 @@ class GlmModel:
                 for tc in c.cells:  # [:1]:
                     x0, y0, x1, y1 = tc.bbox.as_tuple()
                     draw.rectangle([(x0, y0), (x1, y1)], outline=cell_color)
-            image.show()
 
-        # draw_clusters_and_cells(ds_doc, 0)
-        # draw_clusters_and_cells(exported_doc, 0)
+            if show:
+                image.show()
+            else:
+                out_path: Path = (
+                    Path(settings.debug.debug_output_path)
+                    / f"debug_{conv_res.input.file.stem}"
+                )
+                out_path.mkdir(parents=True, exist_ok=True)
+
+                out_file = out_path / f"doc_page_{page_no:05}.png"
+                image.save(str(out_file), format="png")
+
+        # for item in ds_doc.page_dimensions:
+        #    page_no = item.page
+        #    draw_clusters_and_cells(ds_doc, page_no)
 
         return docling_doc

docling-2.3.1/docling/models/easyocr_model.py

@@ -0,0 +1,100 @@
+import logging
+from typing import Iterable
+
+import numpy
+from docling_core.types.doc import BoundingBox, CoordOrigin
+
+from docling.datamodel.base_models import OcrCell, Page
+from docling.datamodel.document import ConversionResult
+from docling.datamodel.pipeline_options import EasyOcrOptions
+from docling.datamodel.settings import settings
+from docling.models.base_ocr_model import BaseOcrModel
+from docling.utils.profiling import TimeRecorder
+
+_log = logging.getLogger(__name__)
+
+
+class EasyOcrModel(BaseOcrModel):
+    def __init__(self, enabled: bool, options: EasyOcrOptions):
+        super().__init__(enabled=enabled, options=options)
+        self.options: EasyOcrOptions
+
+        self.scale = 3  # multiplier for 72 dpi == 216 dpi.
+
+        if self.enabled:
+            try:
+                import easyocr
+            except ImportError:
+                raise ImportError(
+                    "EasyOCR is not installed. Please install it via `pip install easyocr` to use this OCR engine. "
+                    "Alternatively, Docling has support for other OCR engines. See the documentation."
+                )
+
+            self.reader = easyocr.Reader(
+                lang_list=self.options.lang,
+                model_storage_directory=self.options.model_storage_directory,
+                download_enabled=self.options.download_enabled,
+            )
+
+    def __call__(
+        self, conv_res: ConversionResult, page_batch: Iterable[Page]
+    ) -> Iterable[Page]:
+
+        if not self.enabled:
+            yield from page_batch
+            return
+
+        for page in page_batch:
+
+            assert page._backend is not None
+            if not page._backend.is_valid():
+                yield page
+            else:
+                with TimeRecorder(conv_res, "ocr"):
+                    ocr_rects = self.get_ocr_rects(page)
+
+                    all_ocr_cells = []
+                    for ocr_rect in ocr_rects:
+                        # Skip zero area boxes
+                        if ocr_rect.area() == 0:
+                            continue
+                        high_res_image = page._backend.get_page_image(
+                            scale=self.scale, cropbox=ocr_rect
+                        )
+                        im = numpy.array(high_res_image)
+                        result = self.reader.readtext(im)
+
+                        del high_res_image
+                        del im
+
+                        cells = [
+                            OcrCell(
+                                id=ix,
+                                text=line[1],
+                                confidence=line[2],
+                                bbox=BoundingBox.from_tuple(
+                                    coord=(
+                                        (line[0][0][0] / self.scale) + ocr_rect.l,
+                                        (line[0][0][1] / self.scale) + ocr_rect.t,
+                                        (line[0][2][0] / self.scale) + ocr_rect.l,
+                                        (line[0][2][1] / self.scale) + ocr_rect.t,
+                                    ),
+                                    origin=CoordOrigin.TOPLEFT,
+                                ),
+                            )
+                            for ix, line in enumerate(result)
+                        ]
+                        all_ocr_cells.extend(cells)
+
+                    ## Remove OCR cells which overlap with programmatic cells.
+                    filtered_ocr_cells = self.filter_ocr_cells(
+                        all_ocr_cells, page.cells
+                    )
+
+                    page.cells.extend(filtered_ocr_cells)
+
+                # DEBUG code:
+                if settings.debug.visualize_ocr:
+                    self.draw_ocr_rects_and_cells(conv_res, page, ocr_rects)
+
+                yield page