docling 2.2.1__py3-none-any.whl → 2.3.1__py3-none-any.whl

docling/models/page_assemble_model.py

@@ -12,8 +12,10 @@ from docling.datamodel.base_models import (
     Table,
     TextElement,
 )
+from docling.datamodel.document import ConversionResult
 from docling.models.base_model import BasePageModel
 from docling.models.layout_model import LayoutModel
+from docling.utils.profiling import TimeRecorder
 
 _log = logging.getLogger(__name__)
 
@@ -51,122 +53,122 @@ class PageAssembleModel(BasePageModel):
 
         return sanitized_text.strip()  # Strip any leading or trailing whitespace
 
-    def __call__(self, page_batch: Iterable[Page]) -> Iterable[Page]:
+    def __call__(
+        self, conv_res: ConversionResult, page_batch: Iterable[Page]
+    ) -> Iterable[Page]:
         for page in page_batch:
             assert page._backend is not None
             if not page._backend.is_valid():
                 yield page
             else:
-                assert page.predictions.layout is not None
-
-                # assembles some JSON output page by page.
-
-                elements: List[PageElement] = []
-                headers: List[PageElement] = []
-                body: List[PageElement] = []
-
-                for cluster in page.predictions.layout.clusters:
-                    # _log.info("Cluster label seen:", cluster.label)
-                    if cluster.label in LayoutModel.TEXT_ELEM_LABELS:
-
-                        textlines = [
-                            cell.text.replace("\x02", "-").strip()
-                            for cell in cluster.cells
-                            if len(cell.text.strip()) > 0
-                        ]
-                        text = self.sanitize_text(textlines)
-                        text_el = TextElement(
-                            label=cluster.label,
-                            id=cluster.id,
-                            text=text,
-                            page_no=page.page_no,
-                            cluster=cluster,
-                        )
-                        elements.append(text_el)
-
-                        if cluster.label in LayoutModel.PAGE_HEADER_LABELS:
-                            headers.append(text_el)
-                        else:
-                            body.append(text_el)
-                    elif cluster.label == LayoutModel.TABLE_LABEL:
-                        tbl = None
-                        if page.predictions.tablestructure:
-                            tbl = page.predictions.tablestructure.table_map.get(
-                                cluster.id, None
-                            )
-                        if (
-                            not tbl
-                        ):  # fallback: add table without structure, if it isn't present
-                            tbl = Table(
+                with TimeRecorder(conv_res, "page_assemble"):
+
+                    assert page.predictions.layout is not None
+
+                    # assembles some JSON output page by page.
+
+                    elements: List[PageElement] = []
+                    headers: List[PageElement] = []
+                    body: List[PageElement] = []
+
+                    for cluster in page.predictions.layout.clusters:
+                        # _log.info("Cluster label seen:", cluster.label)
+                        if cluster.label in LayoutModel.TEXT_ELEM_LABELS:
+
+                            textlines = [
+                                cell.text.replace("\x02", "-").strip()
+                                for cell in cluster.cells
+                                if len(cell.text.strip()) > 0
+                            ]
+                            text = self.sanitize_text(textlines)
+                            text_el = TextElement(
                                 label=cluster.label,
                                 id=cluster.id,
-                                text="",
-                                otsl_seq=[],
-                                table_cells=[],
-                                cluster=cluster,
+                                text=text,
                                 page_no=page.page_no,
+                                cluster=cluster,
                             )
+                            elements.append(text_el)
+
+                            if cluster.label in LayoutModel.PAGE_HEADER_LABELS:
+                                headers.append(text_el)
+                            else:
+                                body.append(text_el)
+                        elif cluster.label == LayoutModel.TABLE_LABEL:
+                            tbl = None
+                            if page.predictions.tablestructure:
+                                tbl = page.predictions.tablestructure.table_map.get(
+                                    cluster.id, None
+                                )
+                            if (
+                                not tbl
+                            ):  # fallback: add table without structure, if it isn't present
+                                tbl = Table(
+                                    label=cluster.label,
+                                    id=cluster.id,
+                                    text="",
+                                    otsl_seq=[],
+                                    table_cells=[],
+                                    cluster=cluster,
+                                    page_no=page.page_no,
+                                )
 
-                        elements.append(tbl)
-                        body.append(tbl)
-                    elif cluster.label == LayoutModel.FIGURE_LABEL:
-                        fig = None
-                        if page.predictions.figures_classification:
-                            fig = (
-                                page.predictions.figures_classification.figure_map.get(
+                            elements.append(tbl)
+                            body.append(tbl)
+                        elif cluster.label == LayoutModel.FIGURE_LABEL:
+                            fig = None
+                            if page.predictions.figures_classification:
+                                fig = page.predictions.figures_classification.figure_map.get(
                                     cluster.id, None
                                 )
-                            )
-                        if (
-                            not fig
-                        ):  # fallback: add figure without classification, if it isn't present
-                            fig = FigureElement(
-                                label=cluster.label,
-                                id=cluster.id,
-                                text="",
-                                data=None,
-                                cluster=cluster,
-                                page_no=page.page_no,
-                            )
-                        elements.append(fig)
-                        body.append(fig)
-                    elif cluster.label == LayoutModel.FORMULA_LABEL:
-                        equation = None
-                        if page.predictions.equations_prediction:
-                            equation = (
-                                page.predictions.equations_prediction.equation_map.get(
+                            if (
+                                not fig
+                            ):  # fallback: add figure without classification, if it isn't present
+                                fig = FigureElement(
+                                    label=cluster.label,
+                                    id=cluster.id,
+                                    text="",
+                                    data=None,
+                                    cluster=cluster,
+                                    page_no=page.page_no,
+                                )
+                            elements.append(fig)
+                            body.append(fig)
+                        elif cluster.label == LayoutModel.FORMULA_LABEL:
+                            equation = None
+                            if page.predictions.equations_prediction:
+                                equation = page.predictions.equations_prediction.equation_map.get(
                                     cluster.id, None
                                 )
-                            )
-                        if (
-                            not equation
-                        ):  # fallback: add empty formula, if it isn't present
-                            text = self.sanitize_text(
-                                [
-                                    cell.text.replace("\x02", "-").strip()
-                                    for cell in cluster.cells
-                                    if len(cell.text.strip()) > 0
-                                ]
-                            )
-                            equation = TextElement(
-                                label=cluster.label,
-                                id=cluster.id,
-                                cluster=cluster,
-                                page_no=page.page_no,
-                                text=text,
-                            )
-                        elements.append(equation)
-                        body.append(equation)
+                            if (
+                                not equation
+                            ):  # fallback: add empty formula, if it isn't present
+                                text = self.sanitize_text(
+                                    [
+                                        cell.text.replace("\x02", "-").strip()
+                                        for cell in cluster.cells
+                                        if len(cell.text.strip()) > 0
+                                    ]
+                                )
+                                equation = TextElement(
+                                    label=cluster.label,
+                                    id=cluster.id,
+                                    cluster=cluster,
+                                    page_no=page.page_no,
+                                    text=text,
+                                )
+                            elements.append(equation)
+                            body.append(equation)
 
-                page.assembled = AssembledUnit(
-                    elements=elements, headers=headers, body=body
-                )
+                    page.assembled = AssembledUnit(
+                        elements=elements, headers=headers, body=body
+                    )
 
-                # Remove page images (can be disabled)
-                if not self.options.keep_images:
-                    page._image_cache = {}
+                    # Remove page images (can be disabled)
+                    if not self.options.keep_images:
+                        page._image_cache = {}
 
-                # Unload backend
-                page._backend.unload()
+                    # Unload backend
+                    page._backend.unload()
 
                 yield page

docling/models/page_preprocessing_model.py

@@ -1,10 +1,14 @@
+from pathlib import Path
 from typing import Iterable, Optional
 
 from PIL import ImageDraw
 from pydantic import BaseModel
 
 from docling.datamodel.base_models import Page
+from docling.datamodel.document import ConversionResult
+from docling.datamodel.settings import settings
 from docling.models.base_model import BasePageModel
+from docling.utils.profiling import TimeRecorder
 
 
 class PagePreprocessingOptions(BaseModel):
@@ -15,14 +19,17 @@ class PagePreprocessingModel(BasePageModel):
     def __init__(self, options: PagePreprocessingOptions):
         self.options = options
 
-    def __call__(self, page_batch: Iterable[Page]) -> Iterable[Page]:
+    def __call__(
+        self, conv_res: ConversionResult, page_batch: Iterable[Page]
+    ) -> Iterable[Page]:
         for page in page_batch:
             assert page._backend is not None
             if not page._backend.is_valid():
                 yield page
             else:
-                page = self._populate_page_images(page)
-                page = self._parse_page_cells(page)
+                with TimeRecorder(conv_res, "page_parse"):
+                    page = self._populate_page_images(page)
+                    page = self._parse_page_cells(conv_res, page)
                 yield page
 
     # Generate the page image and store it in the page object
@@ -43,19 +50,30 @@ class PagePreprocessingModel(BasePageModel):
         return page
 
     # Extract and populate the page cells and store it in the page object
-    def _parse_page_cells(self, page: Page) -> Page:
+    def _parse_page_cells(self, conv_res: ConversionResult, page: Page) -> Page:
         assert page._backend is not None
 
         page.cells = list(page._backend.get_text_cells())
 
         # DEBUG code:
-        def draw_text_boxes(image, cells):
+        def draw_text_boxes(image, cells, show: bool = False):
             draw = ImageDraw.Draw(image)
             for c in cells:
                 x0, y0, x1, y1 = c.bbox.as_tuple()
                 draw.rectangle([(x0, y0), (x1, y1)], outline="red")
-            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"cells_page_{page.page_no:05}.png"
+                image.save(str(out_file), format="png")
 
-        # draw_text_boxes(page.get_image(scale=1.0), cells)
+        if settings.debug.visualize_cells:
+            draw_text_boxes(page.get_image(scale=1.0), page.cells)
 
         return page

docling/models/table_structure_model.py

@@ -1,6 +1,6 @@
 import copy
 from pathlib import Path
-from typing import Iterable, List
+from typing import Iterable
 
 import numpy
 from docling_core.types.doc import BoundingBox, DocItemLabel, TableCell
@@ -8,8 +8,11 @@ from docling_ibm_models.tableformer.data_management.tf_predictor import TFPredic
 from PIL import ImageDraw
 
 from docling.datamodel.base_models import Page, Table, TableStructurePrediction
+from docling.datamodel.document import ConversionResult
 from docling.datamodel.pipeline_options import TableFormerMode, TableStructureOptions
+from docling.datamodel.settings import settings
 from docling.models.base_model import BasePageModel
+from docling.utils.profiling import TimeRecorder
 
 
 class TableStructureModel(BasePageModel):
@@ -35,7 +38,13 @@ class TableStructureModel(BasePageModel):
             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[Table]):
+    def draw_table_and_cells(
+        self,
+        conv_res: ConversionResult,
+        page: Page,
+        tbl_list: Iterable[Table],
+        show: bool = False,
+    ):
         assert page._backend is not None
 
         image = (
@@ -61,9 +70,21 @@ class TableStructureModel(BasePageModel):
                         fill="black",
                     )
 
-        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"table_struct_page_{page.page_no:05}.png"
+            image.save(str(out_file), format="png")
 
-    def __call__(self, page_batch: Iterable[Page]) -> Iterable[Page]:
+    def __call__(
+        self, conv_res: ConversionResult, page_batch: Iterable[Page]
+    ) -> Iterable[Page]:
 
         if not self.enabled:
             yield from page_batch
@@ -74,98 +95,112 @@ class TableStructureModel(BasePageModel):
             if not page._backend.is_valid():
                 yield page
             else:
-
-                assert page.predictions.layout is not None
-                assert page.size is not None
-
-                page.predictions.tablestructure = TableStructurePrediction()  # dummy
-
-                in_tables = [
-                    (
-                        cluster,
-                        [
-                            round(cluster.bbox.l) * self.scale,
-                            round(cluster.bbox.t) * self.scale,
-                            round(cluster.bbox.r) * self.scale,
-                            round(cluster.bbox.b) * self.scale,
-                        ],
+                with TimeRecorder(conv_res, "table_structure"):
+
+                    assert page.predictions.layout is not None
+                    assert page.size is not None
+
+                    page.predictions.tablestructure = (
+                        TableStructurePrediction()
+                    )  # dummy
+
+                    in_tables = [
+                        (
+                            cluster,
+                            [
+                                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
+                        if cluster.label == DocItemLabel.TABLE
+                    ]
+                    if not len(in_tables):
+                        yield page
+                        continue
+
+                    tokens = []
+                    for c in page.cells:
+                        for cluster, _ in in_tables:
+                            if c.bbox.area() > 0:
+                                if (
+                                    c.bbox.intersection_area_with(cluster.bbox)
+                                    / c.bbox.area()
+                                    > 0.2
+                                ):
+                                    # Only allow non empty stings (spaces) into the cells of a table
+                                    if len(c.text.strip()) > 0:
+                                        new_cell = copy.deepcopy(c)
+                                        new_cell.bbox = new_cell.bbox.scaled(
+                                            scale=self.scale
+                                        )
+
+                                        tokens.append(new_cell.model_dump())
+
+                    page_input = {
+                        "tokens": tokens,
+                        "width": page.size.width * self.scale,
+                        "height": page.size.height * self.scale,
+                    }
+                    page_input["image"] = numpy.asarray(
+                        page.get_image(scale=self.scale)
                     )
-                    for cluster in page.predictions.layout.clusters
-                    if cluster.label == DocItemLabel.TABLE
-                ]
-                if not len(in_tables):
-                    yield page
-                    continue
-
-                tokens = []
-                for c in page.cells:
-                    for cluster, _ in in_tables:
-                        if c.bbox.area() > 0:
-                            if (
-                                c.bbox.intersection_area_with(cluster.bbox)
-                                / c.bbox.area()
-                                > 0.2
-                            ):
-                                # Only allow non empty stings (spaces) into the cells of a table
-                                if len(c.text.strip()) > 0:
-                                    new_cell = copy.deepcopy(c)
-                                    new_cell.bbox = new_cell.bbox.scaled(
-                                        scale=self.scale
-                                    )
-
-                                    tokens.append(new_cell.model_dump())
 
-                page_input = {
-                    "tokens": tokens,
-                    "width": page.size.width * self.scale,
-                    "height": page.size.height * self.scale,
-                }
-                page_input["image"] = numpy.asarray(page.get_image(scale=self.scale))
+                    table_clusters, table_bboxes = zip(*in_tables)
 
-                table_clusters, table_bboxes = zip(*in_tables)
-
-                if len(table_bboxes):
-                    tf_output = self.tf_predictor.multi_table_predict(
-                        page_input, table_bboxes, do_matching=self.do_cell_matching
-                    )
-
-                    for table_cluster, table_out in zip(table_clusters, tf_output):
-                        table_cells = []
-                        for element in table_out["tf_responses"]:
-
-                            if not self.do_cell_matching:
-                                the_bbox = BoundingBox.model_validate(
-                                    element["bbox"]
-                                ).scaled(1 / self.scale)
-                                text_piece = page._backend.get_text_in_rect(the_bbox)
-                                element["bbox"]["token"] = text_piece
-
-                            tc = TableCell.model_validate(element)
-                            if self.do_cell_matching and tc.bbox is not None:
-                                tc.bbox = tc.bbox.scaled(1 / self.scale)
-                            table_cells.append(tc)
-
-                        # Retrieving cols/rows, after post processing:
-                        num_rows = table_out["predict_details"]["num_rows"]
-                        num_cols = table_out["predict_details"]["num_cols"]
-                        otsl_seq = table_out["predict_details"]["prediction"]["rs_seq"]
-
-                        tbl = Table(
-                            otsl_seq=otsl_seq,
-                            table_cells=table_cells,
-                            num_rows=num_rows,
-                            num_cols=num_cols,
-                            id=table_cluster.id,
-                            page_no=page.page_no,
-                            cluster=table_cluster,
-                            label=DocItemLabel.TABLE,
+                    if len(table_bboxes):
+                        tf_output = self.tf_predictor.multi_table_predict(
+                            page_input, table_bboxes, do_matching=self.do_cell_matching
                         )
 
-                        page.predictions.tablestructure.table_map[table_cluster.id] = (
-                            tbl
-                        )
+                        for table_cluster, table_out in zip(table_clusters, tf_output):
+                            table_cells = []
+                            for element in table_out["tf_responses"]:
+
+                                if not self.do_cell_matching:
+                                    the_bbox = BoundingBox.model_validate(
+                                        element["bbox"]
+                                    ).scaled(1 / self.scale)
+                                    text_piece = page._backend.get_text_in_rect(
+                                        the_bbox
+                                    )
+                                    element["bbox"]["token"] = text_piece
+
+                                tc = TableCell.model_validate(element)
+                                if self.do_cell_matching and tc.bbox is not None:
+                                    tc.bbox = tc.bbox.scaled(1 / self.scale)
+                                table_cells.append(tc)
+
+                            # Retrieving cols/rows, after post processing:
+                            num_rows = table_out["predict_details"]["num_rows"]
+                            num_cols = table_out["predict_details"]["num_cols"]
+                            otsl_seq = table_out["predict_details"]["prediction"][
+                                "rs_seq"
+                            ]
+
+                            tbl = Table(
+                                otsl_seq=otsl_seq,
+                                table_cells=table_cells,
+                                num_rows=num_rows,
+                                num_cols=num_cols,
+                                id=table_cluster.id,
+                                page_no=page.page_no,
+                                cluster=table_cluster,
+                                label=DocItemLabel.TABLE,
+                            )
+
+                            page.predictions.tablestructure.table_map[
+                                table_cluster.id
+                            ] = tbl
 
                     # For debugging purposes:
-                    # self.draw_table_and_cells(page, page.predictions.tablestructure.table_map.values())
+                    if settings.debug.visualize_tables:
+                        self.draw_table_and_cells(
+                            conv_res,
+                            page,
+                            page.predictions.tablestructure.table_map.values(),
+                        )
 
                 yield page