docling 2.1.0__py3-none-any.whl → 2.4.1__py3-none-any.whl

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

docling/models/tesseract_ocr_cli_model.py

@@ -8,8 +8,11 @@ import pandas as pd
 from docling_core.types.doc import BoundingBox, CoordOrigin
 
 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__)
 
@@ -102,7 +105,9 @@ 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
@@ -113,62 +118,67 @@ class TesseractOcrCliModel(BaseOcrModel):
             if not page._backend.is_valid():
                 yield page
             else:
-                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 TimeRecorder(conv_res, "ocr"):
 
-                    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,
-                            ),
+                    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
                         )
-                        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)
+                        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
+                    )
 
-                page.cells.extend(filtered_ocr_cells)
+                    page.cells.extend(filtered_ocr_cells)
 
                 # DEBUG code:
-                # self.draw_ocr_rects_and_cells(page, ocr_rects)
+                if settings.debug.visualize_ocr:
+                    self.draw_ocr_rects_and_cells(conv_res, page, ocr_rects)
 
                 yield page

docling/models/tesseract_ocr_model.py

@@ -4,8 +4,11 @@ from typing import Iterable
 from docling_core.types.doc import BoundingBox, CoordOrigin
 
 from docling.datamodel.base_models import OcrCell, Page
+from docling.datamodel.document import ConversionResult
 from docling.datamodel.pipeline_options import TesseractOcrOptions
+from docling.datamodel.settings import settings
 from docling.models.base_ocr_model import BaseOcrModel
+from docling.utils.profiling import TimeRecorder
 
 _log = logging.getLogger(__name__)
 
@@ -19,25 +22,37 @@ class TesseractOcrModel(BaseOcrModel):
         self.reader = None
 
         if self.enabled:
-            setup_errmsg = (
+            install_errmsg = (
                 "tesserocr is not correctly installed. "
                 "Please install it via `pip install tesserocr` to use this OCR engine. "
-                "Note that tesserocr might have to be manually compiled for working with"
+                "Note that tesserocr might have to be manually compiled for working with "
                 "your Tesseract installation. The Docling documentation provides examples for it. "
-                "Alternatively, Docling has support for other OCR engines. See the documentation."
+                "Alternatively, Docling has support for other OCR engines. See the documentation: "
+                "https://ds4sd.github.io/docling/installation/"
             )
+            missing_langs_errmsg = (
+                "tesserocr is not correctly configured. No language models have been detected. "
+                "Please ensure that the TESSDATA_PREFIX envvar points to tesseract languages dir. "
+                "You can find more information how to setup other OCR engines in Docling "
+                "documentation: "
+                "https://ds4sd.github.io/docling/installation/"
+            )
+
             try:
                 import tesserocr
             except ImportError:
-                raise ImportError(setup_errmsg)
-
+                raise ImportError(install_errmsg)
             try:
                 tesseract_version = tesserocr.tesseract_version()
-                _log.debug("Initializing TesserOCR: %s", tesseract_version)
             except:
-                raise ImportError(setup_errmsg)
+                raise ImportError(install_errmsg)
+
+            _, tesserocr_languages = tesserocr.get_languages()
+            if not tesserocr_languages:
+                raise ImportError(missing_langs_errmsg)
 
             # Initialize the tesseractAPI
+            _log.debug("Initializing TesserOCR: %s", tesseract_version)
             lang = "+".join(self.options.lang)
             if self.options.path is not None:
                 self.reader = tesserocr.PyTessBaseAPI(
@@ -61,7 +76,9 @@ class TesseractOcrModel(BaseOcrModel):
             # Finalize the tesseractAPI
             self.reader.End()
 
-    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
@@ -72,59 +89,66 @@ class TesseractOcrModel(BaseOcrModel):
             if not page._backend.is_valid():
                 yield page
             else:
-                assert self.reader is not None
+                with TimeRecorder(conv_res, "ocr"):
 
-                ocr_rects = self.get_ocr_rects(page)
+                    assert self.reader is not None
 
-                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
-                    )
+                    ocr_rects = self.get_ocr_rects(page)
 
-                    # Retrieve text snippets with their bounding boxes
-                    self.reader.SetImage(high_res_image)
-                    boxes = self.reader.GetComponentImages(
-                        self.reader_RIL.TEXTLINE, True
-                    )
+                    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
+                        )
 
-                    cells = []
-                    for ix, (im, box, _, _) in enumerate(boxes):
-                        # Set the area of interest. Tesseract uses Bottom-Left for the origin
-                        self.reader.SetRectangle(box["x"], box["y"], box["w"], box["h"])
-
-                        # Extract text within the bounding box
-                        text = self.reader.GetUTF8Text().strip()
-                        confidence = self.reader.MeanTextConf()
-                        left = box["x"] / self.scale
-                        bottom = box["y"] / self.scale
-                        right = (box["x"] + box["w"]) / self.scale
-                        top = (box["y"] + box["h"]) / self.scale
-
-                        cells.append(
-                            OcrCell(
-                                id=ix,
-                                text=text,
-                                confidence=confidence,
-                                bbox=BoundingBox.from_tuple(
-                                    coord=(left, top, right, bottom),
-                                    origin=CoordOrigin.TOPLEFT,
-                                ),
-                            )
+                        # Retrieve text snippets with their bounding boxes
+                        self.reader.SetImage(high_res_image)
+                        boxes = self.reader.GetComponentImages(
+                            self.reader_RIL.TEXTLINE, True
                         )
 
-                    # del high_res_image
-                    all_ocr_cells.extend(cells)
+                        cells = []
+                        for ix, (im, box, _, _) in enumerate(boxes):
+                            # Set the area of interest. Tesseract uses Bottom-Left for the origin
+                            self.reader.SetRectangle(
+                                box["x"], box["y"], box["w"], box["h"]
+                            )
+
+                            # Extract text within the bounding box
+                            text = self.reader.GetUTF8Text().strip()
+                            confidence = self.reader.MeanTextConf()
+                            left = box["x"] / self.scale
+                            bottom = box["y"] / self.scale
+                            right = (box["x"] + box["w"]) / self.scale
+                            top = (box["y"] + box["h"]) / self.scale
+
+                            cells.append(
+                                OcrCell(
+                                    id=ix,
+                                    text=text,
+                                    confidence=confidence,
+                                    bbox=BoundingBox.from_tuple(
+                                        coord=(left, top, right, bottom),
+                                        origin=CoordOrigin.TOPLEFT,
+                                    ),
+                                )
+                            )
+
+                        # del high_res_image
+                        all_ocr_cells.extend(cells)
 
-                ## Remove OCR cells which overlap with programmatic cells.
-                filtered_ocr_cells = self.filter_ocr_cells(all_ocr_cells, page.cells)
+                    ## 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)
+                if settings.debug.visualize_ocr:
+                    self.draw_ocr_rects_and_cells(conv_res, page, ocr_rects)
 
                 yield page