mirrors/docling
1
0
Fork 0
mirror of https://github.com/DS4SD/docling.git synced 2025-07-27 04:24:45 +00:00
Code Issues Packages Projects Releases Wiki Activity

Enabled figure support in vlm_pipeline

Browse Source 
Signed-off-by: Maksym Lysak <mly@zurich.ibm.com>
This commit is contained in:
Maksym Lysak 2025-01-10 13:56:46 +01:00
parent 1b968e4984
commit ef079e4e78
1 changed files with 22 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

40
docling/pipeline/vlm_pipeline.py
View File

@ -155,7 +155,7 @@ class VlmPipeline(PaginatedPipeline):
# def _turn_tags_into_doc(self, xml_content: str, image_bytes: bytes) -> (DoclingDocument, list): # def _turn_tags_into_doc(self, xml_content: str, image_bytes: bytes) -> (DoclingDocument, list):
def _turn_tags_into_doc( def _turn_tags_into_doc(
self, xml_content: str, input_image: Optional[Image] = None self, xml_content: str, pil_image: Optional[Image] = None
) -> DoclingDocument: ) -> DoclingDocument:
def extract_text(tag_content: str) -> str: def extract_text(tag_content: str) -> str:
return re.sub(r"<.*?>", "", tag_content).strip() return re.sub(r"<.*?>", "", tag_content).strip()
@ -457,23 +457,27 @@ class VlmPipeline(PaginatedPipeline):
bbox = extract_bounding_box(line) bbox = extract_bounding_box(line)
if bbox: if bbox:
bounding_boxes.append((bbox, "yellow")) bounding_boxes.append((bbox, "yellow"))
# Convert bounding box normalized to 0-100 into pixel coordinates for cropping if pil_image:
""" # Convert bounding box normalized to 0-100 into pixel coordinates for cropping
width, height = pil_image.size width, height = pil_image.size
crop_box = ( crop_box = (
int(bbox.l * width), int(bbox.l * width),
int(bbox.t * height), int(bbox.t * height),
int(bbox.r * width), int(bbox.r * width),
int(bbox.b * height), int(bbox.b * height),
) )
cropped_image = pil_image.crop(crop_box) cropped_image = pil_image.crop(crop_box)
doc.add_picture( doc.add_picture(
parent=current_group, parent=current_group,
image=ImageRef.from_pil(image=cropped_image, dpi=300), image=ImageRef.from_pil(image=cropped_image, dpi=300),
prov=ProvenanceItem(bbox=bbox, charspan=(0, 0), page_no=1), prov=ProvenanceItem(bbox=bbox, charspan=(0, 0), page_no=1),
) )
""" else:
doc.add_picture(
parent=current_group,
prov=ProvenanceItem(bbox=bbox, charspan=(0, 0), page_no=1),
)
elif line.startswith("<list>"): elif line.startswith("<list>"):
content = extract_text(line) content = extract_text(line)
prov_item_inst = None prov_item_inst = None