feat: introducing docling_backend (#26)

Uses our own docling_parse to reliably get PDF cells
To get page images, this backend uses pypdfium2

Signed-off-by: Maxim Lysak <[email protected]>
Co-authored-by: Maxim Lysak <[email protected]>

docling/backend/docling_parse_backend.py

@@ -0,0 +1,171 @@
+import random
+from io import BytesIO
+from pathlib import Path
+from typing import Iterable, List, Optional, Union
+
+import pypdfium2 as pdfium
+from docling_parse.docling_parse import pdf_parser
+from PIL import Image, ImageDraw
+from pypdfium2 import PdfPage
+
+from docling.backend.abstract_backend import PdfDocumentBackend, PdfPageBackend
+from docling.datamodel.base_models import BoundingBox, Cell, CoordOrigin, PageSize
+
+
+class DoclingParsePageBackend(PdfPageBackend):
+    def __init__(self, page_obj: PdfPage, docling_page_obj):
+        super().__init__(page_obj)
+        self._ppage = page_obj
+        self._dpage = docling_page_obj
+        self.text_page = None
+
+    def get_text_in_rect(self, bbox: BoundingBox) -> str:
+        # Find intersecting cells on the page
+        text_piece = ""
+        page_size = self.get_size()
+        parser_width = self._dpage["width"]
+        parser_height = self._dpage["height"]
+
+        scale = (
+            1  # FIX - Replace with param in get_text_in_rect across backends (optional)
+        )
+
+        for i in range(len(self._dpage["cells"])):
+            rect = self._dpage["cells"][i]["box"]["device"]
+            x0, y0, x1, y1 = rect
+            cell_bbox = BoundingBox(
+                l=x0 * scale * page_size.width / parser_width,
+                b=y0 * scale * page_size.height / parser_height,
+                r=x1 * scale * page_size.width / parser_width,
+                t=y1 * scale * page_size.height / parser_height,
+                coord_origin=CoordOrigin.BOTTOMLEFT,
+            ).to_top_left_origin(page_size.height * scale)
+
+            overlap_frac = cell_bbox.intersection_area_with(bbox) / cell_bbox.area()
+
+            if overlap_frac > 0.5:
+                if len(text_piece) > 0:
+                    text_piece += " "
+                text_piece += self._dpage["cells"][i]["content"]["rnormalized"]
+
+        return text_piece
+
+    def get_text_cells(self) -> Iterable[Cell]:
+        cells = []
+        cell_counter = 0
+
+        page_size = self.get_size()
+
+        parser_width = self._dpage["width"]
+        parser_height = self._dpage["height"]
+
+        for i in range(len(self._dpage["cells"])):
+            rect = self._dpage["cells"][i]["box"]["device"]
+            x0, y0, x1, y1 = rect
+            text_piece = self._dpage["cells"][i]["content"]["rnormalized"]
+            cells.append(
+                Cell(
+                    id=cell_counter,
+                    text=text_piece,
+                    bbox=BoundingBox(
+                        # l=x0, b=y0, r=x1, t=y1,
+                        l=x0 * page_size.width / parser_width,
+                        b=y0 * page_size.height / parser_height,
+                        r=x1 * page_size.width / parser_width,
+                        t=y1 * page_size.height / parser_height,
+                        coord_origin=CoordOrigin.BOTTOMLEFT,
+                    ).to_top_left_origin(page_size.height),
+                )
+            )
+            cell_counter += 1
+
+        def draw_clusters_and_cells():
+            image = self.get_page_image()
+            draw = ImageDraw.Draw(image)
+            for c in cells:
+                x0, y0, x1, y1 = c.bbox.as_tuple()
+                cell_color = (
+                    random.randint(30, 140),
+                    random.randint(30, 140),
+                    random.randint(30, 140),
+                )
+                draw.rectangle([(x0, y0), (x1, y1)], outline=cell_color)
+            image.show()
+
+        # before merge:
+        # draw_clusters_and_cells()
+
+        # cells = merge_horizontal_cells(cells)
+
+        # after merge:
+        # draw_clusters_and_cells()
+
+        return cells
+
+    def get_page_image(
+        self, scale: int = 1, cropbox: Optional[BoundingBox] = None
+    ) -> Image.Image:
+
+        page_size = self.get_size()
+
+        if not cropbox:
+            cropbox = BoundingBox(
+                l=0,
+                r=page_size.width,
+                t=0,
+                b=page_size.height,
+                coord_origin=CoordOrigin.TOPLEFT,
+            )
+            padbox = BoundingBox(
+                l=0, r=0, t=0, b=0, coord_origin=CoordOrigin.BOTTOMLEFT
+            )
+        else:
+            padbox = cropbox.to_bottom_left_origin(page_size.height)
+            padbox.r = page_size.width - padbox.r
+            padbox.t = page_size.height - padbox.t
+
+        image = (
+            self._ppage.render(
+                scale=scale * 1.5,
+                rotation=0,  # no additional rotation
+                crop=padbox.as_tuple(),
+            )
+            .to_pil()
+            .resize(size=(round(cropbox.width * scale), round(cropbox.height * scale)))
+        )  # We resize the image from 1.5x the given scale to make it sharper.
+
+        return image
+
+    def get_size(self) -> PageSize:
+        return PageSize(width=self._ppage.get_width(), height=self._ppage.get_height())
+
+    def unload(self):
+        self._ppage = None
+        self._dpage = None
+        self.text_page = None
+
+
+class DoclingParseDocumentBackend(PdfDocumentBackend):
+    def __init__(self, path_or_stream: Iterable[Union[BytesIO, Path]]):
+        super().__init__(path_or_stream)
+        self._pdoc = pdfium.PdfDocument(path_or_stream)
+        # Parsing cells with docling_parser call
+        print("PARSING WITH DOCLING PARSE")
+        parser = pdf_parser()
+        self._parser_doc = parser.find_cells(str(path_or_stream))
+
+    def page_count(self) -> int:
+        return len(self._parser_doc["pages"])
+
+    def load_page(self, page_no: int) -> PdfPage:
+        return DoclingParsePageBackend(
+            self._pdoc[page_no], self._parser_doc["pages"][page_no]
+        )
+
+    def is_valid(self) -> bool:
+        return self.page_count() > 0
+
+    def unload(self):
+        self._pdoc.close()
+        self._pdoc = None
+        self._parser_doc = None

examples/convert.py

@@ -4,7 +4,8 @@
 from pathlib import Path
 from typing import Iterable
 
-from docling.backend.pypdfium2_backend import PyPdfiumDocumentBackend
+# from docling.backend.pypdfium2_backend import PyPdfiumDocumentBackend
+from docling.backend.docling_parse_backend import DoclingParseDocumentBackend
 from docling.datamodel.base_models import ConversionStatus, PipelineOptions
 from docling.datamodel.document import ConvertedDocument, DocumentConversionInput
 from docling.document_converter import DocumentConverter
@@ -54,11 +55,12 @@ def main():
     artifacts_path = DocumentConverter.download_models_hf()
 
     pipeline_options = PipelineOptions(do_table_structure=True)
-    # use text cells predicted from table structure model, instead of matching with pdf cells
-    pipeline_options.table_structure_options.do_cell_matching = False
+    pipeline_options.table_structure_options.do_cell_matching = True
 
     doc_converter = DocumentConverter(
-        artifacts_path=artifacts_path, pipeline_options=pipeline_options
+        artifacts_path=artifacts_path,
+        pipeline_options=pipeline_options,
+        pdf_backend=DoclingParseDocumentBackend,
     )
 
     input = DocumentConversionInput.from_paths(input_doc_paths)

pyproject.toml

@@ -32,6 +32,7 @@ pydantic-settings = "^2.3.0"
 huggingface_hub = ">=0.23,<1"
 requests = "^2.32.3"
 easyocr = { version = "^1.7", optional = true }
+docling-parse = "^0.0.1"
 
 [tool.poetry.group.dev.dependencies]
 black = {extras = ["jupyter"], version = "^24.4.2"}