docling 1.19.1__py3-none-any.whl → 2.4.1__py3-none-any.whl

docling/models/page_preprocessing_model.py

@@ -0,0 +1,79 @@
+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):
+    images_scale: Optional[float]
+
+
+class PagePreprocessingModel(BasePageModel):
+    def __init__(self, options: PagePreprocessingOptions):
+        self.options = options
+
+    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:
+                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
+    def _populate_page_images(self, page: Page) -> Page:
+        # default scale
+        page.get_image(
+            scale=1.0
+        )  # puts the page image on the image cache at default scale
+
+        images_scale = self.options.images_scale
+        # user requested scales
+        if images_scale is not None:
+            page._default_image_scale = images_scale
+            page.get_image(
+                scale=images_scale
+            )  # this will trigger storing the image in the internal cache
+
+        return page
+
+    # Extract and populate the page cells and store it in the page object
+    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, 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")
+            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")
+
+        if settings.debug.visualize_cells:
+            draw_text_boxes(page.get_image(scale=1.0), page.cells)
+
+        return page

docling/models/table_structure_model.py

@@ -1,31 +1,30 @@
 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
 from docling_ibm_models.tableformer.data_management.tf_predictor import TFPredictor
 from PIL import ImageDraw
 
-from docling.datamodel.base_models import (
-    BoundingBox,
-    Page,
-    TableCell,
-    TableElement,
-    TableStructurePrediction,
-)
-from docling.datamodel.pipeline_options import TableFormerMode
+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:
-    def __init__(self, config):
-        self.config = config
-        self.do_cell_matching = config["do_cell_matching"]
-        self.mode = config["mode"]
+class TableStructureModel(BasePageModel):
+    def __init__(
+        self, enabled: bool, artifacts_path: Path, options: TableStructureOptions
+    ):
+        self.options = options
+        self.do_cell_matching = self.options.do_cell_matching
+        self.mode = self.options.mode
 
-        self.enabled = config["enabled"]
+        self.enabled = enabled
         if self.enabled:
-            artifacts_path: Path = config["artifacts_path"]
-
             if self.mode == TableFormerMode.ACCURATE:
                 artifacts_path = artifacts_path / "fat"
 
@@ -39,7 +38,15 @@ class TableStructureModel:
             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]):
+    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 = (
             page._backend.get_page_image()
         )  # make new image to avoid drawing on the saved ones
@@ -50,111 +57,150 @@ class TableStructureModel:
             draw.rectangle([(x0, y0), (x1, y1)], outline="red")
 
             for tc in table_element.table_cells:
-                x0, y0, x1, y1 = tc.bbox.as_tuple()
-                if tc.column_header:
-                    width = 3
-                else:
-                    width = 1
-                draw.rectangle([(x0, y0), (x1, y1)], outline="blue", width=width)
-                draw.text(
-                    (x0 + 3, y0 + 3),
-                    text=f"{tc.start_row_offset_idx}, {tc.start_col_offset_idx}",
-                    fill="black",
-                )
-
-        image.show()
-
-    def __call__(self, page_batch: Iterable[Page]) -> Iterable[Page]:
+                if tc.bbox is not None:
+                    x0, y0, x1, y1 = tc.bbox.as_tuple()
+                    if tc.column_header:
+                        width = 3
+                    else:
+                        width = 1
+                    draw.rectangle([(x0, y0), (x1, y1)], outline="blue", width=width)
+                    draw.text(
+                        (x0 + 3, y0 + 3),
+                        text=f"{tc.start_row_offset_idx}, {tc.start_col_offset_idx}",
+                        fill="black",
+                    )
+
+        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, conv_res: ConversionResult, page_batch: Iterable[Page]
+    ) -> Iterable[Page]:
 
         if not self.enabled:
             yield from page_batch
             return
 
         for page in page_batch:
-
-            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 == "Table"
-            ]
-            if not len(in_tables):
+            assert page._backend is not None
+            if not page._backend.is_valid():
                 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)
-
-            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:
-                            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 = TableElement(
-                        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="Table",
+            else:
+                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)
                     )
 
-                    page.predictions.tablestructure.table_map[table_cluster.id] = tbl
+                    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,
+                            )
+
+                            page.predictions.tablestructure.table_map[
+                                table_cluster.id
+                            ] = tbl
+
+                    # For debugging purposes:
+                    if settings.debug.visualize_tables:
+                        self.draw_table_and_cells(
+                            conv_res,
+                            page,
+                            page.predictions.tablestructure.table_map.values(),
+                        )
 
-                # For debugging purposes:
-                # self.draw_table_and_cells(page, page.predictions.tablestructure.table_map.values())
-
-            yield page
+                yield page

docling/models/tesseract_ocr_cli_model.py

@@ -2,13 +2,17 @@ import io
 import logging
 import tempfile
 from subprocess import DEVNULL, PIPE, Popen
-from typing import Iterable, Tuple
+from typing import Iterable, Optional, Tuple
 
 import pandas as pd
+from docling_core.types.doc import BoundingBox, CoordOrigin
 
-from docling.datamodel.base_models import BoundingBox, CoordOrigin, OcrCell, Page
+from docling.datamodel.base_models import OcrCell, Page
+from docling.datamodel.document import ConversionResult
 from docling.datamodel.pipeline_options import TesseractCliOcrOptions
+from docling.datamodel.settings import settings
 from docling.models.base_ocr_model import BaseOcrModel
+from docling.utils.profiling import TimeRecorder
 
 _log = logging.getLogger(__name__)
 
@@ -21,8 +25,8 @@ class TesseractOcrCliModel(BaseOcrModel):
 
         self.scale = 3  # multiplier for 72 dpi == 216 dpi.
 
-        self._name = None
-        self._version = None
+        self._name: Optional[str] = None
+        self._version: Optional[str] = None
 
         if self.enabled:
             try:
@@ -39,7 +43,7 @@ class TesseractOcrCliModel(BaseOcrModel):
     def _get_name_and_version(self) -> Tuple[str, str]:
 
         if self._name != None and self._version != None:
-            return self._name, self._version
+            return self._name, self._version  # type: ignore
 
         cmd = [self.options.tesseract_cmd, "--version"]
 
@@ -101,67 +105,80 @@ class TesseractOcrCliModel(BaseOcrModel):
 
         return df_filtered
 
-    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
             return
 
         for page in page_batch:
-            ocr_rects = self.get_ocr_rects(page)
-
-            all_ocr_cells = []
-            for ocr_rect in ocr_rects:
-                # Skip zero area boxes
-                if ocr_rect.area() == 0:
-                    continue
-                high_res_image = page._backend.get_page_image(
-                    scale=self.scale, cropbox=ocr_rect
-                )
-
-                with tempfile.NamedTemporaryFile(suffix=".png", mode="w") as image_file:
-                    fname = image_file.name
-                    high_res_image.save(fname)
-
-                    df = self._run_tesseract(fname)
-
-                # _log.info(df)
-
-                # Print relevant columns (bounding box and text)
-                for ix, row in df.iterrows():
-                    text = row["text"]
-                    conf = row["conf"]
-
-                    l = float(row["left"])
-                    b = float(row["top"])
-                    w = float(row["width"])
-                    h = float(row["height"])
-
-                    t = b + h
-                    r = l + w
-
-                    cell = OcrCell(
-                        id=ix,
-                        text=text,
-                        confidence=conf / 100.0,
-                        bbox=BoundingBox.from_tuple(
-                            coord=(
-                                (l / self.scale) + ocr_rect.l,
-                                (b / self.scale) + ocr_rect.t,
-                                (r / self.scale) + ocr_rect.l,
-                                (t / self.scale) + ocr_rect.t,
-                            ),
-                            origin=CoordOrigin.TOPLEFT,
-                        ),
+            assert page._backend is not None
+            if not page._backend.is_valid():
+                yield page
+            else:
+                with TimeRecorder(conv_res, "ocr"):
+
+                    ocr_rects = self.get_ocr_rects(page)
+
+                    all_ocr_cells = []
+                    for ocr_rect in ocr_rects:
+                        # Skip zero area boxes
+                        if ocr_rect.area() == 0:
+                            continue
+                        high_res_image = page._backend.get_page_image(
+                            scale=self.scale, cropbox=ocr_rect
+                        )
+
+                        with tempfile.NamedTemporaryFile(
+                            suffix=".png", mode="w"
+                        ) as image_file:
+                            fname = image_file.name
+                            high_res_image.save(fname)
+
+                            df = self._run_tesseract(fname)
+
+                        # _log.info(df)
+
+                        # Print relevant columns (bounding box and text)
+                        for ix, row in df.iterrows():
+                            text = row["text"]
+                            conf = row["conf"]
+
+                            l = float(row["left"])
+                            b = float(row["top"])
+                            w = float(row["width"])
+                            h = float(row["height"])
+
+                            t = b + h
+                            r = l + w
+
+                            cell = OcrCell(
+                                id=ix,
+                                text=text,
+                                confidence=conf / 100.0,
+                                bbox=BoundingBox.from_tuple(
+                                    coord=(
+                                        (l / self.scale) + ocr_rect.l,
+                                        (b / self.scale) + ocr_rect.t,
+                                        (r / self.scale) + ocr_rect.l,
+                                        (t / self.scale) + ocr_rect.t,
+                                    ),
+                                    origin=CoordOrigin.TOPLEFT,
+                                ),
+                            )
+                            all_ocr_cells.append(cell)
+
+                    ## Remove OCR cells which overlap with programmatic cells.
+                    filtered_ocr_cells = self.filter_ocr_cells(
+                        all_ocr_cells, page.cells
                     )
-                    all_ocr_cells.append(cell)
-
-            ## Remove OCR cells which overlap with programmatic cells.
-            filtered_ocr_cells = self.filter_ocr_cells(all_ocr_cells, page.cells)
 
-            page.cells.extend(filtered_ocr_cells)
+                    page.cells.extend(filtered_ocr_cells)
 
-            # DEBUG code:
-            # self.draw_ocr_rects_and_cells(page, ocr_rects)
+                # DEBUG code:
+                if settings.debug.visualize_ocr:
+                    self.draw_ocr_rects_and_cells(conv_res, page, ocr_rects)
 
-            yield page
+                yield page