mirrors/docling
1
0
Fork 0
mirror of https://github.com/DS4SD/docling.git synced 2025-07-25 19:44:34 +00:00
Code Issues Packages Projects Releases Wiki Activity

Correctly compute PDF boxes from pymupdf

Browse Source 
Signed-off-by: Christoph Auer <cau@zurich.ibm.com>
This commit is contained in:
Christoph Auer 2025-06-11 14:19:55 +02:00
parent 06b408fa41
commit 9752e824fb
5 changed files with 101 additions and 82 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

25
docling/backend/docling_parse_backend.py
View File

@ -9,8 +9,6 @@ import pypdfium2 as pdfium
from docling_core.types.doc import BoundingBox, CoordOrigin, Size
from docling_core.types.doc.page import (
BoundingRectangle,
PdfPageBoundaryType,
PdfPageGeometry,
SegmentedPdfPage,
TextCell,
)
@ -19,6 +17,7 @@ from PIL import Image, ImageDraw
from pypdfium2 import PdfPage
from docling.backend.pdf_backend import PdfDocumentBackend, PdfPageBackend
from docling.backend.pypdfium2_backend import get_pdf_page_geometry
from docling.datamodel.document import InputDocument
_log = logging.getLogger(__name__)
@ -124,28 +123,10 @@ class DoclingParsePageBackend(PdfPageBackend):
if not self.valid:
return None
page_size = self.get_size()
text_cells = self._compute_text_cells()
# Create page geometry
crop_bbox = BoundingBox(
l=0,
r=page_size.width,
t=0,
b=page_size.height,
coord_origin=CoordOrigin.TOPLEFT,
).to_bottom_left_origin(page_size.height)
dimension = PdfPageGeometry(
angle=0.0,
rect=BoundingRectangle.from_bounding_box(crop_bbox),
boundary_type=PdfPageBoundaryType.CROP_BOX,
art_bbox=crop_bbox,
bleed_bbox=crop_bbox,
crop_bbox=crop_bbox,
media_bbox=crop_bbox,
trim_bbox=crop_bbox,
)
# Get the PDF page geometry from pypdfium2
dimension = get_pdf_page_geometry(self._ppage)
# Create SegmentedPdfPage
return SegmentedPdfPage(

24
docling/backend/docling_parse_v2_backend.py
View File

@ -19,6 +19,7 @@ from PIL import Image, ImageDraw
from pypdfium2 import PdfPage
from docling.backend.pdf_backend import PdfDocumentBackend, PdfPageBackend
from docling.backend.pypdfium2_backend import get_pdf_page_geometry
from docling.datamodel.base_models import Size
from docling.utils.locks import pypdfium2_lock
@ -139,28 +140,11 @@ class DoclingParseV2PageBackend(PdfPageBackend):
if not self.valid:
return None
page_size = self.get_size()
text_cells = self._compute_text_cells()
# Create page geometry
crop_bbox = BoundingBox(
l=0,
r=page_size.width,
t=0,
b=page_size.height,
coord_origin=CoordOrigin.TOPLEFT,
).to_bottom_left_origin(page_size.height)
dimension = PdfPageGeometry(
angle=0.0,
rect=BoundingRectangle.from_bounding_box(crop_bbox),
boundary_type=PdfPageBoundaryType.CROP_BOX,
art_bbox=crop_bbox,
bleed_bbox=crop_bbox,
crop_bbox=crop_bbox,
media_bbox=crop_bbox,
trim_bbox=crop_bbox,
)
# Get the PDF page geometry from pypdfium2
with pypdfium2_lock:
dimension = get_pdf_page_geometry(self._ppage)
# Create SegmentedPdfPage
return SegmentedPdfPage(

40
docling/backend/docling_parse_v4_backend.py
View File

@ -59,20 +59,6 @@ class DoclingParseV4PageBackend(PdfPageBackend):
return self._dpage
def get_text_cells(self) -> Iterable[TextCell]:
page_size = self.get_size()
[tc.to_top_left_origin(page_size.height) for tc in self._dpage.textline_cells]
# for cell in self._dpage.textline_cells:
# rect = cell.rect
#
# assert (
# rect.to_bounding_box().l <= rect.to_bounding_box().r
# ), f"left is > right on bounding box {rect.to_bounding_box()} of rect {rect}"
# assert (
# rect.to_bounding_box().t <= rect.to_bounding_box().b
# ), f"top is > bottom on bounding box {rect.to_bounding_box()} of rect {rect}"
return self._dpage.textline_cells
def get_bitmap_rects(self, scale: float = 1) -> Iterable[BoundingBox]:
@ -171,12 +157,28 @@ class DoclingParseV4DocumentBackend(PdfDocumentBackend):
self, page_no: int, create_words: bool = True, create_textlines: bool = True
) -> DoclingParseV4PageBackend:
with pypdfium2_lock:
seg_page = self.dp_doc.get_page(
page_no + 1,
create_words=create_words,
create_textlines=create_textlines,
)
# In Docling, all TextCell instances are expected with top-left origin.
[
tc.to_top_left_origin(seg_page.dimension.height)
for tc in seg_page.textline_cells
]
[
tc.to_top_left_origin(seg_page.dimension.height)
for tc in seg_page.char_cells
]
[
tc.to_top_left_origin(seg_page.dimension.height)
for tc in seg_page.word_cells
]
return DoclingParseV4PageBackend(
self.dp_doc.get_page(
page_no + 1,
create_words=create_words,
create_textlines=create_textlines,
),
seg_page,
self._pdoc[page_no],
)

92
docling/backend/pypdfium2_backend.py
View File

@ -22,6 +22,75 @@ from pypdfium2._helpers.misc import PdfiumError
from docling.backend.pdf_backend import PdfDocumentBackend, PdfPageBackend
from docling.utils.locks import pypdfium2_lock
def get_pdf_page_geometry(
ppage: pdfium.PdfPage,
angle: float = 0.0,
boundary_type: PdfPageBoundaryType = PdfPageBoundaryType.CROP_BOX,
) -> PdfPageGeometry:
"""
Create PdfPageGeometry from a pypdfium2 PdfPage object.
Args:
ppage: pypdfium2 PdfPage object
angle: Page rotation angle in degrees (default: 0.0)
boundary_type: The boundary type for the page (default: CROP_BOX)
Returns:
PdfPageGeometry with all the different bounding boxes properly set
"""
# Get the main bounding box (intersection of crop_box and media_box)
bbox_tuple = ppage.get_bbox()
bbox = BoundingBox.from_tuple(bbox_tuple, CoordOrigin.BOTTOMLEFT)
# Get all the different page boxes from pypdfium2
media_box_tuple = ppage.get_mediabox()
crop_box_tuple = ppage.get_cropbox()
art_box_tuple = ppage.get_artbox()
bleed_box_tuple = ppage.get_bleedbox()
trim_box_tuple = ppage.get_trimbox()
# Convert to BoundingBox objects using existing from_tuple method
# pypdfium2 returns (x0, y0, x1, y1) in PDF coordinate system (bottom-left origin)
# Use bbox as fallback when specific box types are not defined
media_bbox = (
BoundingBox.from_tuple(media_box_tuple, CoordOrigin.BOTTOMLEFT)
if media_box_tuple
else bbox
)
crop_bbox = (
BoundingBox.from_tuple(crop_box_tuple, CoordOrigin.BOTTOMLEFT)
if crop_box_tuple
else bbox
)
art_bbox = (
BoundingBox.from_tuple(art_box_tuple, CoordOrigin.BOTTOMLEFT)
if art_box_tuple
else bbox
)
bleed_bbox = (
BoundingBox.from_tuple(bleed_box_tuple, CoordOrigin.BOTTOMLEFT)
if bleed_box_tuple
else bbox
)
trim_bbox = (
BoundingBox.from_tuple(trim_box_tuple, CoordOrigin.BOTTOMLEFT)
if trim_box_tuple
else bbox
)
return PdfPageGeometry(
angle=angle,
rect=BoundingRectangle.from_bounding_box(bbox),
boundary_type=boundary_type,
art_bbox=art_bbox,
bleed_bbox=bleed_bbox,
crop_bbox=crop_bbox,
media_bbox=media_bbox,
trim_bbox=trim_bbox,
)
if TYPE_CHECKING:
from docling.datamodel.document import InputDocument
@ -213,28 +282,11 @@ class PyPdfiumPageBackend(PdfPageBackend):
if not self.valid:
return None
page_size = self.get_size()
text_cells = self._compute_text_cells()
# Create page geometry
crop_bbox = BoundingBox(
l=0,
r=page_size.width,
t=0,
b=page_size.height,
coord_origin=CoordOrigin.TOPLEFT,
).to_bottom_left_origin(page_size.height)
dimension = PdfPageGeometry(
angle=0.0,
rect=BoundingRectangle.from_bounding_box(crop_bbox),
boundary_type=PdfPageBoundaryType.CROP_BOX,
art_bbox=crop_bbox,
bleed_bbox=crop_bbox,
crop_bbox=crop_bbox,
media_bbox=crop_bbox,
trim_bbox=crop_bbox,
)
# Get the PDF page geometry from pypdfium2
with pypdfium2_lock:
dimension = get_pdf_page_geometry(self._ppage)
# Create SegmentedPdfPage
return SegmentedPdfPage(

2
docling/models/base_ocr_model.py
View File

@ -145,7 +145,7 @@ class BaseOcrModel(BasePageModel, BaseModelWithOptions):
# Update parsed_page.textline_cells directly
page.parsed_page.textline_cells = final_cells
page.parsed_page.has_lines = bool(final_cells)
page.parsed_page.has_lines = len(final_cells) > 0
def _combine_cells(self, existing_cells, ocr_cells):
"""Combine existing and OCR cells with filtering and re-indexing."""