docling 0.3.1__py3-none-any.whl → 0.4.0__py3-none-any.whl

docling/datamodel/base_models.py

@@ -1,3 +1,4 @@
+import copy
 from enum import Enum, auto
 from io import BytesIO
 from typing import Any, Dict, List, Optional, Tuple, Union
@@ -47,6 +48,15 @@ class BoundingBox(BaseModel):
     def height(self):
         return abs(self.t - self.b)
 
+    def scaled(self, scale: float) -> "BoundingBox":
+        out_bbox = copy.deepcopy(self)
+        out_bbox.l *= scale
+        out_bbox.r *= scale
+        out_bbox.t *= scale
+        out_bbox.b *= scale
+
+        return out_bbox
+
     def as_tuple(self):
         if self.coord_origin == CoordOrigin.TOPLEFT:
             return (self.l, self.t, self.r, self.b)
@@ -241,6 +251,17 @@ class DocumentStream(BaseModel):
     stream: BytesIO
 
 
+class TableStructureOptions(BaseModel):
+    do_cell_matching: bool = (
+        True
+        # True:  Matches predictions back to PDF cells. Can break table output if PDF cells
+        #        are merged across table columns.
+        # False: Let table structure model define the text cells, ignore PDF cells.
+    )
+
+
 class PipelineOptions(BaseModel):
-    do_table_structure: bool = True
-    do_ocr: bool = False
+    do_table_structure: bool = True  # True: perform table structure extraction
+    do_ocr: bool = False  # True: perform OCR, replace programmatic PDF text
+
+    table_structure_options: TableStructureOptions = TableStructureOptions()

docling/models/page_assemble_model.py

@@ -19,18 +19,6 @@ class PageAssembleModel:
     def __init__(self, config):
         self.config = config
 
-        # self.line_wrap_pattern = re.compile(r'(?<=[^\W_])- \n(?=\w)')
-
-    # def sanitize_text_poor(self, lines):
-    #     text = '\n'.join(lines)
-    #
-    #     # treat line wraps.
-    #     sanitized_text = self.line_wrap_pattern.sub('', text)
-    #
-    #     sanitized_text = sanitized_text.replace('\n', ' ')
-    #
-    #     return sanitized_text
-
     def sanitize_text(self, lines):
         if len(lines) <= 1:
             return " ".join(lines)

docling/models/table_structure_model.py

@@ -1,7 +1,10 @@
-from typing import Iterable
+import copy
+import random
+from typing import Iterable, List
 
 import numpy
 from docling_ibm_models.tableformer.data_management.tf_predictor import TFPredictor
+from PIL import ImageDraw
 
 from docling.datamodel.base_models import (
     BoundingBox,
@@ -28,6 +31,21 @@ class TableStructureModel:
             self.tm_model_type = self.tm_config["model"]["type"]
 
             self.tf_predictor = TFPredictor(self.tm_config)
+            self.scale = 2.0  # Scale up table input images to 144 dpi
+
+    def draw_table_and_cells(self, page: Page, tbl_list: List[TableElement]):
+        image = page._backend.get_page_image()
+        draw = ImageDraw.Draw(image)
+
+        for table_element in tbl_list:
+            x0, y0, x1, y1 = table_element.cluster.bbox.as_tuple()
+            draw.rectangle([(x0, y0), (x1, y1)], outline="red")
+
+            for tc in table_element.table_cells:
+                x0, y0, x1, y1 = tc.bbox.as_tuple()
+                draw.rectangle([(x0, y0), (x1, y1)], outline="blue")
+
+        image.show()
 
     def __call__(self, page_batch: Iterable[Page]) -> Iterable[Page]:
 
@@ -36,16 +54,17 @@ class TableStructureModel:
             return
 
         for page in page_batch:
+
             page.predictions.tablestructure = TableStructurePrediction()  # dummy
 
             in_tables = [
                 (
                     cluster,
                     [
-                        round(cluster.bbox.l),
-                        round(cluster.bbox.t),
-                        round(cluster.bbox.r),
-                        round(cluster.bbox.b),
+                        round(cluster.bbox.l) * self.scale,
+                        round(cluster.bbox.t) * self.scale,
+                        round(cluster.bbox.r) * self.scale,
+                        round(cluster.bbox.b) * self.scale,
                     ],
                 )
                 for cluster in page.predictions.layout.clusters
@@ -65,20 +84,29 @@ class TableStructureModel:
                         ):
                             # Only allow non empty stings (spaces) into the cells of a table
                             if len(c.text.strip()) > 0:
-                                tokens.append(c.model_dump())
+                                new_cell = copy.deepcopy(c)
+                                new_cell.bbox = new_cell.bbox.scaled(scale=self.scale)
+
+                                tokens.append(new_cell.model_dump())
 
-            iocr_page = {
-                "image": numpy.asarray(page.image),
+            page_input = {
                 "tokens": tokens,
-                "width": page.size.width,
-                "height": page.size.height,
+                "width": page.size.width * self.scale,
+                "height": page.size.height * self.scale,
             }
+            # add image to page input.
+            if self.scale == 1.0:
+                page_input["image"] = numpy.asarray(page.image)
+            else:  # render new page image on the fly at desired scale
+                page_input["image"] = numpy.asarray(
+                    page._backend.get_page_image(scale=self.scale)
+                )
 
             table_clusters, table_bboxes = zip(*in_tables)
 
             if len(table_bboxes):
                 tf_output = self.tf_predictor.multi_table_predict(
-                    iocr_page, table_bboxes, do_matching=self.do_cell_matching
+                    page_input, table_bboxes, do_matching=self.do_cell_matching
                 )
 
                 for table_cluster, table_out in zip(table_clusters, tf_output):
@@ -91,6 +119,7 @@ class TableStructureModel:
                             element["bbox"]["token"] = text_piece
 
                         tc = TableCell.model_validate(element)
+                        tc.bbox = tc.bbox.scaled(1 / self.scale)
                         table_cells.append(tc)
 
                     # Retrieving cols/rows, after post processing:
@@ -111,4 +140,7 @@ class TableStructureModel:
 
                     page.predictions.tablestructure.table_map[table_cluster.id] = tbl
 
+                # For debugging purposes:
+                # self.draw_table_and_cells(page, page.predictions.tablestructure.table_map.values())
+
             yield page

docling/pipeline/standard_model_pipeline.py

@@ -34,7 +34,7 @@ class StandardModelPipeline(BaseModelPipeline):
                     "artifacts_path": artifacts_path
                     / StandardModelPipeline._table_model_path,
                     "enabled": pipeline_options.do_table_structure,
-                    "do_cell_matching": False,
+                    "do_cell_matching": pipeline_options.table_structure_options.do_cell_matching,
                 }
             ),
         ]

{docling-0.3.1.dist-info → docling-0.4.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: docling
-Version: 0.3.1
+Version: 0.4.0
 Summary: Docling PDF conversion package
 Home-page: https://github.com/DS4SD/docling
 License: MIT
@@ -79,7 +79,9 @@ python examples/convert.py
 ```
 The output of the above command will be written to `./scratch`.
 
-### Enable or disable pipeline features
+### Adjust pipeline features
+
+**Control pipeline options**
 
 You can control if table structure recognition or OCR should be performed by arguments passed to `DocumentConverter`:
 ```python
@@ -92,6 +94,23 @@ doc_converter = DocumentConverter(
 )
 ```
 
+**Control table extraction options**
+
+You can control if table structure recognition should map the recognized structure back to PDF cells (default) or use text cells from the structure prediction itself.
+This can improve output quality if you find that multiple columns in extracted tables are erroneously merged into one.
+
+
+```python
+
+pipeline_options = PipelineOptions(do_table_structure=True)
+pipeline_options.table_structure_options.do_cell_matching = False # Uses text cells predicted from table structure model
+
+doc_converter = DocumentConverter(
+    artifacts_path=artifacts_path,
+    pipeline_options=pipeline_options,
+)
+```
+
 ### Impose limits on the document size
 
 You can limit the file size and number of pages which should be allowed to process per document:

{docling-0.3.1.dist-info → docling-0.4.0.dist-info}/RECORD

@@ -3,7 +3,7 @@ docling/backend/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 docling/backend/abstract_backend.py,sha256=dINr8oTax9Fq31Y1AR0CGWNZtAHN5aqB_M7TAPkJNVQ,1122
 docling/backend/pypdfium2_backend.py,sha256=sJMoActFyc3qdKB6RFly3auHXuXM4noQAG0ypUlj26o,7647
 docling/datamodel/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-docling/datamodel/base_models.py,sha256=GKeRryRuCS6mPWJf0IPJ5manXwiuS0v8wFOnVXF38b0,6128
+docling/datamodel/base_models.py,sha256=k7gLFPnq3ArEMAFz6qUcp5qemlYzVhOmR9qtBTkAiX4,6862
 docling/datamodel/document.py,sha256=S4USz13mqLS9WUwTgEkoocykcmY6B3cC3f4JlfTSYcM,12635
 docling/datamodel/settings.py,sha256=t5g6wrEJnPa9gBzMMl8ppgBRUYz-8xgopEtfMS0ZH28,733
 docling/document_converter.py,sha256=MZw23oPlRmRi1ggzoD1PukUnqo-6boO3RZB06dZ5Xt0,7305
@@ -11,15 +11,15 @@ docling/models/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 docling/models/ds_glm_model.py,sha256=wmb--2JKFQby-kvidw6PyM8wURPXYPQ_Z_eKKCBAdYQ,3192
 docling/models/easyocr_model.py,sha256=NaHVs8IN0eW9KB076E2Kae1s-bq74_4IMWueze9QqtE,2290
 docling/models/layout_model.py,sha256=4AfPFiu6pXc8wIQ1sQlEZnHRt7SnBmfzDdctiRveOWw,10944
-docling/models/page_assemble_model.py,sha256=jhjQt0NOkVi-dWBaovJ2KsBim5FF6e47y21uZ8EWfBg,5906
-docling/models/table_structure_model.py,sha256=uvkK2NPvltk9-zScbORUA05JbvymkGX6Dfsal4wLwsI,4103
+docling/models/page_assemble_model.py,sha256=8eoG2WiFxPxq9TPvM-wkngb2gkr0tdtCRVXg1JcTETo,5550
+docling/models/table_structure_model.py,sha256=ryZrmkNkCbw5SCpgdQabkmcRAEi_4VqOMv2VGdpvGZo,5499
 docling/pipeline/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 docling/pipeline/base_model_pipeline.py,sha256=ozHdJak0yQAxQf7pQN_C480vI35A2e5KL5Qq1xSkq5c,560
-docling/pipeline/standard_model_pipeline.py,sha256=pDbgVO0oOJry7Q-3KYdMuaypXCQOdoVikR80veizo9o,1489
+docling/pipeline/standard_model_pipeline.py,sha256=UTwodKUKrisLoVcntbNUBDhjzRyFvpdUvyVw-gNmBlM,1541
 docling/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 docling/utils/layout_utils.py,sha256=FOFbL0hKzUoWXdZaeUvEtFqKv0IkPifIr4sdGW4suKs,31804
 docling/utils/utils.py,sha256=llhXSbIDNZ1MHOwBEfLHBAoJIAYI7QlPIonlI1jLUJ0,1208
-docling-0.3.1.dist-info/LICENSE,sha256=ACwmltkrXIz5VsEQcrqljq-fat6ZXAMepjXGoe40KtE,1069
-docling-0.3.1.dist-info/METADATA,sha256=5OpesJEMNC_jdf88GO7drN0XHZkmpmk0J13mM1E50rk,5390
-docling-0.3.1.dist-info/WHEEL,sha256=sP946D7jFCHeNz5Iq4fL4Lu-PrWrFsgfLXbbkciIZwg,88
-docling-0.3.1.dist-info/RECORD,,
+docling-0.4.0.dist-info/LICENSE,sha256=ACwmltkrXIz5VsEQcrqljq-fat6ZXAMepjXGoe40KtE,1069
+docling-0.4.0.dist-info/METADATA,sha256=aWx7RrxtFIXHkPqEBrzO_gJfPjgWcpgENqx02cdBQys,6044
+docling-0.4.0.dist-info/WHEEL,sha256=sP946D7jFCHeNz5Iq4fL4Lu-PrWrFsgfLXbbkciIZwg,88
+docling-0.4.0.dist-info/RECORD,,