diff --git a/docling_eval/cli/main.py b/docling_eval/cli/main.py index d9a4ffac..e1b566f7 100644 --- a/docling_eval/cli/main.py +++ b/docling_eval/cli/main.py @@ -73,6 +73,13 @@ # Configure logging logging.getLogger("docling").setLevel(logging.WARNING) +logging.getLogger("PIL").setLevel(logging.WARNING) +logging.getLogger("transformers").setLevel(logging.WARNING) +logging.getLogger("datasets").setLevel(logging.WARNING) +logging.getLogger("filelock").setLevel(logging.WARNING) +logging.getLogger("urllib3").setLevel(logging.WARNING) +logging.getLogger("docling_ibm_models").setLevel(logging.WARNING) + _log = logging.getLogger(__name__) app = typer.Typer( @@ -188,14 +195,17 @@ def get_dataset_builder( name="CVAT", dataset_source=dataset_source, target=target, split=split ) elif benchmark == BenchMarkNames.PLAIN_FILES: - assert dataset_source is not None + if dataset_source is None: + raise ValueError("dataset_source is required for PLAIN_FILES") + return FileDatasetBuilder( name=dataset_source.name, dataset_source=dataset_source, target=target, split=split, + begin_index=begin_index, + end_index=end_index, ) - else: raise ValueError(f"Unsupported benchmark: {benchmark}") @@ -209,7 +219,11 @@ def get_prediction_provider( ): pipeline_options: PaginatedPipelineOptions """Get the appropriate prediction provider with default settings.""" - if provider_type == PredictionProviderType.DOCLING: + if ( + provider_type == PredictionProviderType.DOCLING + or provider_type == PredictionProviderType.OCR_DOCLING + or provider_type == PredictionProviderType.EasyOCR_DOCLING + ): ocr_factory = get_ocr_factory() ocr_options: OcrOptions = ocr_factory.create_options( # type: ignore @@ -238,6 +252,78 @@ def get_prediction_provider( ignore_missing_predictions=True, ) + elif provider_type == PredictionProviderType.MacOCR_DOCLING: + ocr_factory = get_ocr_factory() + + ocr_options: OcrOptions = ocr_factory.create_options( # type: ignore + kind="ocrmac", + ) + + pipeline_options = PdfPipelineOptions( + do_ocr=True, + ocr_options=ocr_options, + do_table_structure=True, + ) + + pipeline_options.images_scale = 2.0 + pipeline_options.generate_page_images = True + pipeline_options.generate_picture_images = True + + if artifacts_path is not None: + pipeline_options.artifacts_path = artifacts_path + + return DoclingPredictionProvider( + format_options={ + InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline_options), + InputFormat.IMAGE: PdfFormatOption(pipeline_options=pipeline_options), + }, + do_visualization=do_visualization, + ignore_missing_predictions=True, + ) + + elif provider_type == PredictionProviderType.PDF_DOCLING: + + ocr_factory = get_ocr_factory() + + ocr_options: OcrOptions = ocr_factory.create_options( # type: ignore + kind="easyocr", + ) + + pdf_pipeline_options = PdfPipelineOptions( + do_ocr=False, + ocr_options=ocr_options, # we need to provide OCR options in order to not break the parquet serialization + do_table_structure=True, + ) + + pdf_pipeline_options.images_scale = 2.0 + pdf_pipeline_options.generate_page_images = True + pdf_pipeline_options.generate_picture_images = True + + ocr_pipeline_options = PdfPipelineOptions( + do_ocr=True, + ocr_options=ocr_options, # we need to provide OCR options in order to not break the parquet serialization + do_table_structure=True, + ) + + ocr_pipeline_options.images_scale = 2.0 + ocr_pipeline_options.generate_page_images = True + ocr_pipeline_options.generate_picture_images = True + + if artifacts_path is not None: + pdf_pipeline_options.artifacts_path = artifacts_path + ocr_pipeline_options.artifacts_path = artifacts_path + + return DoclingPredictionProvider( + format_options={ + InputFormat.PDF: PdfFormatOption(pipeline_options=pdf_pipeline_options), + InputFormat.IMAGE: PdfFormatOption( + pipeline_options=ocr_pipeline_options + ), + }, + do_visualization=do_visualization, + ignore_missing_predictions=True, + ) + elif provider_type == PredictionProviderType.SMOLDOCLING: pipeline_options = VlmPipelineOptions() @@ -614,9 +700,14 @@ def create_cvat( output_dir: Annotated[Path, typer.Option(help="Output directory")], gt_dir: Annotated[Path, typer.Option(help="Dataset source path")], bucket_size: Annotated[int, typer.Option(help="Size of CVAT tasks")] = 20, + use_predictions: Annotated[bool, typer.Option(help="use predictions")] = False, ): + """Create dataset ready to upload to CVAT starting from (ground-truth) dataset.""" builder = CvatPreannotationBuilder( - dataset_source=gt_dir, target=output_dir, bucket_size=bucket_size + dataset_source=gt_dir, + target=output_dir, + bucket_size=bucket_size, + use_predictions=use_predictions, ) builder.prepare_for_annotation() diff --git a/docling_eval/datamodels/types.py b/docling_eval/datamodels/types.py index 93a5011f..120ee414 100644 --- a/docling_eval/datamodels/types.py +++ b/docling_eval/datamodels/types.py @@ -118,6 +118,11 @@ class PredictionProviderType(str, Enum): """Types of prediction providers available.""" DOCLING = "Docling" + PDF_DOCLING = "PDF_Docling" + OCR_DOCLING = "OCR_Docling" + MacOCR_DOCLING = "MacOCR_Docling" + EasyOCR_DOCLING = "EasyOCR_Docling" + TABLEFORMER = "TableFormer" FILE = "File" SMOLDOCLING = "SmolDocling" diff --git a/docling_eval/dataset_builders/dataset_builder.py b/docling_eval/dataset_builders/dataset_builder.py index a3b96a5e..012c95fd 100644 --- a/docling_eval/dataset_builders/dataset_builder.py +++ b/docling_eval/dataset_builders/dataset_builder.py @@ -7,6 +7,7 @@ import ibm_boto3 # type: ignore from docling.utils.utils import chunkify +from docling_core.types.doc.document import ImageRefMode from huggingface_hub import snapshot_download from pydantic import BaseModel @@ -15,7 +16,6 @@ TRUE_HTML_EXPORT_LABELS, ) from docling_eval.utils.utils import save_shard_to_disk, write_datasets_info -from docling_eval.visualisation.visualisations import save_inspection_html # Get logger _log = logging.getLogger(__name__) @@ -276,10 +276,11 @@ def save_to_disk( record_list.append(r.as_record_dict()) if do_visualization: viz_path = self.target / "visualizations" / f"{r.doc_id}.html" - save_inspection_html( + r.ground_truth_doc.save_as_html( filename=viz_path, - doc=r.ground_truth_doc, labels=TRUE_HTML_EXPORT_LABELS, + image_mode=ImageRefMode.EMBEDDED, + split_page_view=True, ) save_shard_to_disk( diff --git a/docling_eval/dataset_builders/file_dataset_builder.py b/docling_eval/dataset_builders/file_dataset_builder.py index d32a9b4e..111bd19f 100644 --- a/docling_eval/dataset_builders/file_dataset_builder.py +++ b/docling_eval/dataset_builders/file_dataset_builder.py @@ -100,7 +100,7 @@ def iterate(self) -> Iterable[DatasetRecord]: for filename in tqdm( selected_filenames, - desc="Processing files for DP-Bench", + desc=f"Processing files for {self.name}", ncols=128, ): mime_type, _ = mimetypes.guess_type(filename) @@ -108,6 +108,7 @@ def iterate(self) -> Iterable[DatasetRecord]: # Create the ground truth Document true_doc = DoclingDocument(name=f"{filename}") if mime_type == "application/pdf": + _log.info(f"add_pages_to_true_doc: {filename}") true_doc, _ = add_pages_to_true_doc( pdf_path=filename, true_doc=true_doc, image_scale=2.0 ) @@ -126,6 +127,7 @@ def iterate(self) -> Iterable[DatasetRecord]: image=image_ref, ) + _log.info(f"add_pages_to_true_doc: {filename}") true_doc.pages[1] = page_item else: raise ValueError( @@ -139,18 +141,20 @@ def iterate(self) -> Iterable[DatasetRecord]: page_images_column=BenchMarkColumns.GROUNDTRUTH_PAGE_IMAGES.value, ) - # Get PDF as binary data - pdf_bytes = get_binary(filename) - pdf_stream = DocumentStream(name=filename.name, stream=BytesIO(pdf_bytes)) + # Get source as binary data + source_bytes = get_binary(filename) + source_stream = DocumentStream( + name=filename.name, stream=BytesIO(source_bytes) + ) # Create dataset record record = DatasetRecord( doc_id=str(filename.name), - doc_hash=get_binhash(pdf_bytes), + doc_hash=get_binhash(source_bytes), ground_truth_doc=true_doc, ground_truth_pictures=true_pictures, ground_truth_page_images=true_page_images, - original=pdf_stream, + original=source_stream, mime_type=mime_type, ) diff --git a/docling_eval/evaluators/readingorder_evaluator.py b/docling_eval/evaluators/readingorder_evaluator.py index 3c3fa364..01c64fb5 100644 --- a/docling_eval/evaluators/readingorder_evaluator.py +++ b/docling_eval/evaluators/readingorder_evaluator.py @@ -292,7 +292,7 @@ def _show_items(self, true_doc: DoclingDocument): ) text = item.text if isinstance(item, TextItem) else None label = item.label # type: ignore - print(f"True {i}: {level} - {label}: {bbox} - {text}") + # print(f"True {i}: {level} - {label}: {bbox} - {text}") class ReadingOrderVisualizer: diff --git a/docling_eval/prediction_providers/base_prediction_provider.py b/docling_eval/prediction_providers/base_prediction_provider.py index 996cafdd..d280ed98 100644 --- a/docling_eval/prediction_providers/base_prediction_provider.py +++ b/docling_eval/prediction_providers/base_prediction_provider.py @@ -165,7 +165,6 @@ def visualize_results( / f"{prediction_record.doc_id}.html", true_doc=gt_doc, pred_doc=pred_doc, - page_image=prediction_record.ground_truth_page_images[0], true_labels=self.true_labels, pred_labels=self.pred_labels, draw_reading_order=True, diff --git a/docling_eval/prediction_providers/docling_provider.py b/docling_eval/prediction_providers/docling_provider.py index 880e58b4..b86b619b 100644 --- a/docling_eval/prediction_providers/docling_provider.py +++ b/docling_eval/prediction_providers/docling_provider.py @@ -106,7 +106,7 @@ def predict(self, record: DatasetRecord) -> DatasetRecordWithPrediction: def info(self) -> Dict: """Get information about the prediction provider.""" - return { + result = { "asset": PredictionProviderType.DOCLING, "version": docling_version(), "package_versions": { @@ -128,10 +128,11 @@ def info(self) -> Dict: mode="json", exclude_defaults=True ) if v.pipeline_options is not None - else {} + else None # Parquet might not like empty dicts! ), } for k, v in self.doc_converter.format_to_options.items() if k in [InputFormat.PDF, InputFormat.IMAGE] }, } + return result diff --git a/docling_eval/visualisation/constants.py b/docling_eval/visualisation/constants.py index a0a04e3c..823c258a 100644 --- a/docling_eval/visualisation/constants.py +++ b/docling_eval/visualisation/constants.py @@ -309,7 +309,7 @@ display: flex; flex-direction: column; width: 25%; /* Adjust the width of each item */ - height: 100%; /* Adjust height to fill parent container */ + height: 50%; /* Adjust height to fill parent container */ border: 1px solid #ccc; /* Optional: Add borders */ box-shadow: 2px 2px 5px rgba(0, 0, 0, 0.1); /* Optional: Add shadow */ background-color: #fff; /* Optional: Add background */ diff --git a/docling_eval/visualisation/visualisations.py b/docling_eval/visualisation/visualisations.py index 330c634a..f782fca6 100644 --- a/docling_eval/visualisation/visualisations.py +++ b/docling_eval/visualisation/visualisations.py @@ -1,10 +1,16 @@ import copy import logging +import re from pathlib import Path from typing import Set from docling.datamodel.base_models import BoundingBox, Cluster from docling.utils.visualization import draw_clusters +from docling_core.experimental.serializer.html import ( + HTMLDocSerializer, + HTMLOutputStyle, + HTMLParams, +) from docling_core.types.doc.document import ( ContentLayer, DocItem, @@ -23,54 +29,6 @@ ) -def save_comparison_html( - filename: Path, - true_doc: DoclingDocument, - pred_doc: DoclingDocument, - page_image: Image.Image, - true_labels: Set[DocItemLabel], - pred_labels: Set[DocItemLabel], -): - - true_doc_html = true_doc.export_to_html( - image_mode=ImageRefMode.EMBEDDED, - html_head=HTML_DEFAULT_HEAD_FOR_COMP, - labels=true_labels, - ) - - pred_doc_html = pred_doc.export_to_html( - image_mode=ImageRefMode.EMBEDDED, - html_head=HTML_DEFAULT_HEAD_FOR_COMP, - labels=pred_labels, - ) - - # since the string in srcdoc are wrapped by ', we need to replace all ' by it HTML convention - true_doc_html = true_doc_html.replace("'", "'") - pred_doc_html = pred_doc_html.replace("'", "'") - - image_base64 = from_pil_to_base64(page_image) - - """ - # Convert the image to a bytes object - buffered = io.BytesIO() - page_image.save( - buffered, format="PNG" - ) # Specify the format (e.g., JPEG, PNG, etc.) - image_bytes = buffered.getvalue() - - # Encode the bytes to a Base64 string - image_base64 = base64.b64encode(image_bytes).decode("utf-8") - """ - - comparison_page = copy.deepcopy(HTML_COMPARISON_PAGE) - comparison_page = comparison_page.replace("BASE64PAGE", image_base64) - comparison_page = comparison_page.replace("TRUEDOC", true_doc_html) - comparison_page = comparison_page.replace("PREDDOC", pred_doc_html) - - with open(str(filename), "w") as fw: - fw.write(comparison_page) - - def draw_arrow( draw: ImageDraw.ImageDraw, arrow_coords: tuple[float, float, float, float], @@ -117,296 +75,151 @@ def draw_arrow( return draw -def draw_clusters_with_reading_order( - doc: DoclingDocument, - page_image: Image.Image, - labels: Set[DocItemLabel], - page_no: int = 1, - reading_order: bool = True, -): - - # img = copy.deepcopy(page_image) - img = page_image.copy() - draw = ImageDraw.Draw(img) - - # Load a font (adjust the font size and path as needed) - font = ImageFont.load_default() - try: - font = ImageFont.truetype("arial.ttf", size=15) - except IOError: - font = ImageFont.load_default() - - x0, y0 = None, None - - for item, level in doc.iterate_items( - included_content_layers={ContentLayer.BODY, ContentLayer.FURNITURE} - ): - if isinstance(item, DocItem): # and item.label in labels: - for prov in item.prov: - - if page_no != prov.page_no: - continue - - bbox = prov.bbox.to_top_left_origin( - page_height=doc.pages[prov.page_no].size.height - ) - bbox = bbox.normalized(doc.pages[prov.page_no].size) - - bbox.l = round(bbox.l * img.width) - bbox.r = round(bbox.r * img.width) - bbox.t = round(bbox.t * img.height) - bbox.b = round(bbox.b * img.height) - - if bbox.b > bbox.t: - bbox.b, bbox.t = bbox.t, bbox.b - - if not reading_order: - x0, y0 = None, None - elif x0 is None and y0 is None: - x0 = (bbox.l + bbox.r) / 2.0 - y0 = (bbox.b + bbox.t) / 2.0 - else: - assert x0 is not None - assert y0 is not None - - x1 = (bbox.l + bbox.r) / 2.0 - y1 = (bbox.b + bbox.t) / 2.0 - - # Arrow parameters - start_point = (x0, y0) # Starting point of the arrow - end_point = (x1, y1) # Ending point of the arrow - arrowhead_length = 20 # Length of the arrowhead - arrowhead_width = 10 # Width of the arrowhead - - arrow_color = "red" - line_width = 2 - - # Draw the arrow shaft (line) - draw.line( - [start_point, end_point], fill=arrow_color, width=line_width - ) - - # Calculate the arrowhead points - dx = end_point[0] - start_point[0] - dy = end_point[1] - start_point[1] - angle = (dx**2 + dy**2) ** 0.5 + 0.01 # Length of the arrow shaft - - # Normalized direction vector for the arrow shaft - ux, uy = dx / angle, dy / angle - - # Base of the arrowhead - base_x = end_point[0] - ux * arrowhead_length - base_y = end_point[1] - uy * arrowhead_length - - # Left and right points of the arrowhead - left_x = base_x - uy * arrowhead_width - left_y = base_y + ux * arrowhead_width - right_x = base_x + uy * arrowhead_width - right_y = base_y - ux * arrowhead_width - - # Draw the arrowhead (triangle) - draw.polygon( - [end_point, (left_x, left_y), (right_x, right_y)], - fill=arrow_color, - ) - - x0, y0 = x1, y1 - - # Draw rectangle with only a border - rectangle_color = "blue" - border_width = 1 - draw.rectangle( - [bbox.l, bbox.b, bbox.r, bbox.t], - outline=rectangle_color, - width=border_width, - ) - - # Calculate label size using getbbox - text_bbox = font.getbbox(str(item.label)) - label_width = text_bbox[2] - text_bbox[0] - label_height = text_bbox[3] - text_bbox[1] - label_x = bbox.l - label_y = ( - bbox.b - label_height - ) # - 5 # Place the label above the rectangle - - # Draw label text - draw.text( - (label_x, label_y), - str(item.label), - fill=rectangle_color, - font=font, - ) - - return img - - def save_comparison_html_with_clusters( filename: Path, true_doc: DoclingDocument, pred_doc: DoclingDocument, - page_image: Image.Image, true_labels: Set[DocItemLabel], pred_labels: Set[DocItemLabel], draw_reading_order: bool = True, ): - if (1 not in true_doc.pages) or (1 not in pred_doc.pages): - logging.error(f"1 not in true_doc.pages -> skipping {filename} ") - return - - def draw_doc_layout(doc: DoclingDocument, image: Image.Image): - r""" - Draw the document clusters and optionaly the reading order - """ - clusters = [] - for idx, (elem, _) in enumerate( - doc.iterate_items( - included_content_layers={ContentLayer.BODY, ContentLayer.FURNITURE} - ) - ): - if not isinstance(elem, DocItem): - continue - if len(elem.prov) == 0: - continue # Skip elements without provenances - prov = elem.prov[0] - - if prov.page_no not in true_doc.pages or prov.page_no != 1: - logging.error(f"{prov.page_no} not in true_doc.pages -> skipping! ") - continue - - tlo_bbox = prov.bbox.to_top_left_origin( - page_height=doc.pages[prov.page_no].size.height - ) - cluster = Cluster( - id=idx, - label=elem.label, - bbox=BoundingBox.model_validate(tlo_bbox), - cells=[], - ) - clusters.append(cluster) - - scale_x = image.width / doc.pages[1].size.width - scale_y = image.height / doc.pages[1].size.height - draw_clusters(image, clusters, scale_x, scale_y) + """Save comparison html with clusters.""" - return image + def get_missing_pageimg(width=800, height=1100, text="MISSING PAGE"): + """Get missing page imgage.""" + import numpy as np + from PIL import Image, ImageDraw, ImageFont - def draw_doc_reading_order(doc: DoclingDocument, image: Image.Image): - r""" - Draw the reading order - """ + # Create a white background image + image = Image.new("RGB", (width, height), color="white") draw = ImageDraw.Draw(image) - x0, y0 = None, None - - for elem, _ in doc.iterate_items( - included_content_layers={ContentLayer.BODY, ContentLayer.FURNITURE} - ): - if not isinstance(elem, DocItem): - continue - if len(elem.prov) == 0: - continue # Skip elements without provenances - prov = elem.prov[0] - - if prov.page_no not in true_doc.pages or prov.page_no != 1: - logging.error(f"{prov.page_no} not in true_doc.pages -> skipping! ") - continue - - tlo_bbox = prov.bbox.to_top_left_origin( - page_height=doc.pages[prov.page_no].size.height - ) - ro_bbox = tlo_bbox.normalized(doc.pages[prov.page_no].size) - ro_bbox.l = round(ro_bbox.l * image.width) - ro_bbox.r = round(ro_bbox.r * image.width) - ro_bbox.t = round(ro_bbox.t * image.height) - ro_bbox.b = round(ro_bbox.b * image.height) - - if ro_bbox.b > ro_bbox.t: - ro_bbox.b, ro_bbox.t = ro_bbox.t, ro_bbox.b - - if x0 is None and y0 is None: - x0 = (ro_bbox.l + ro_bbox.r) / 2.0 - y0 = (ro_bbox.b + ro_bbox.t) / 2.0 - else: - assert x0 is not None - assert y0 is not None - - x1 = (ro_bbox.l + ro_bbox.r) / 2.0 - y1 = (ro_bbox.b + ro_bbox.t) / 2.0 - - draw = draw_arrow( - draw, - (x0, y0, x1, y1), - line_width=2, - color="red", - ) - x0, y0 = x1, y1 + + # Try to use a standard font or fall back to default + try: + # For larger installations, you might have Arial or other fonts + font = ImageFont.truetype("arial.ttf", size=60) + except IOError: + # Fall back to default font + font = ImageFont.load_default().font_variant(size=60) + + # Get text size to center it + text_width, text_height = ( + draw.textsize(text, font=font) + if hasattr(draw, "textsize") + else (draw.textlength(text, font=font), font.size) + ) + + # Position for the text (centered and angled) + position = ((width - text_width) // 2, (height - text_height) // 2) + + # Draw the watermark text (light gray and rotated) + draw.text(position, text, fill=(200, 200, 200), font=font) + + # Rotate the image 45 degrees to create diagonal watermark effect + image = image.rotate(45, expand=False, fillcolor="white") + return image - # HTML rendering - true_doc_html = true_doc.export_to_html( - image_mode=ImageRefMode.EMBEDDED, - html_head=HTML_DEFAULT_HEAD_FOR_COMP, - labels=true_labels, - ) + true_page_imgs = true_doc.get_visualization(show_label=False) + pred_page_imgs = pred_doc.get_visualization(show_label=False) - pred_doc_html = pred_doc.export_to_html( - image_mode=ImageRefMode.EMBEDDED, - html_head=HTML_DEFAULT_HEAD_FOR_COMP, - labels=pred_labels, + true_page_nos = true_page_imgs.keys() + pred_page_nos = pred_page_imgs.keys() + + if true_page_nos != pred_page_nos: + logging.error( + f"incompatible true_page_nos versus pred_page_nos: \ntrue_page_nos: {true_page_nos}\npred_page_nos: {pred_page_nos}" + ) + + page_nos = true_page_nos | pred_page_nos + + html_parts = [ + "", + "", + HTML_DEFAULT_HEAD_FOR_COMP, + "", + ] + + html_parts.append("") + html_parts.append("") + + # Compile a regular expression pattern to match content within tags + pattern = re.compile( + r"]*>\n
(.*?)
\n", + re.DOTALL | re.IGNORECASE, ) - # since the string in srcdoc are wrapped by ', we need to replace all ' by it HTML convention - true_doc_html = true_doc_html.replace("'", "'") - pred_doc_html = pred_doc_html.replace("'", "'") + for page_no in page_nos: - true_doc_img = draw_doc_layout(true_doc, copy.deepcopy(page_image)) - pred_doc_img = draw_doc_layout(pred_doc, copy.deepcopy(page_image)) + if page_no in true_page_imgs: + true_doc_img_b64 = from_pil_to_base64(true_page_imgs[page_no]) + else: + logging.error(f"{page_no} not in true_page_imgs, get default image.") + true_doc_img_b64 = from_pil_to_base64(get_missing_pageimg()) - if draw_reading_order: - true_doc_img = draw_doc_reading_order(true_doc, true_doc_img) - pred_doc_img = draw_doc_reading_order(pred_doc, pred_doc_img) + if page_no in pred_page_imgs: + pred_doc_img_b64 = from_pil_to_base64(pred_page_imgs[page_no]) + else: + logging.error(f"{page_no} not in pred_page_imgs, get default image.") + pred_doc_img_b64 = from_pil_to_base64(get_missing_pageimg()) - true_doc_img_b64 = from_pil_to_base64(true_doc_img) - pred_doc_img_b64 = from_pil_to_base64(pred_doc_img) + true_doc_page = true_doc.export_to_html( + image_mode=ImageRefMode.EMBEDDED, page_no=page_no + ) + pred_doc_page = pred_doc.export_to_html( + image_mode=ImageRefMode.EMBEDDED, page_no=page_no + ) - comparison_page = copy.deepcopy(HTML_COMPARISON_PAGE_WITH_CLUSTERS) - comparison_page = comparison_page.replace("BASE64TRUEPAGE", true_doc_img_b64) - comparison_page = comparison_page.replace("TRUEDOC", true_doc_html) - comparison_page = comparison_page.replace("BASE64PREDPAGE", pred_doc_img_b64) - comparison_page = comparison_page.replace("PREDDOC", pred_doc_html) + # Search for the pattern in the HTML string + mtch = pattern.search(true_doc_page) + if mtch: + true_doc_page_body = mtch.group(1).strip() + else: + logging.error(f"could not find body in true_doc_page") + true_doc_page_body = "

Nothing Found

" - with open(str(filename), "w") as fw: - fw.write(comparison_page) + # Search for the pattern in the HTML string + mtch = pattern.search(pred_doc_page) + if mtch: + pred_doc_page_body = mtch.group(1).strip() + else: + logging.error(f"could not find body in pred_doc_page") + pred_doc_page_body = "

Nothing Found

" + if len(true_doc_page_body) == 0: + true_doc_page_body = "

Nothing Found

" -def save_inspection_html( - filename: Path, doc: DoclingDocument, labels: Set[DocItemLabel] -): + if len(pred_doc_page_body) == 0: + pred_doc_page_body = "

Nothing Found

" + + html_parts.append("") + + html_parts.append("") + + html_parts.append("") - html_doc = doc.export_to_html(image_mode=ImageRefMode.EMBEDDED, labels=labels) - html_doc = html_doc.replace("'", "'") + html_parts.append("") - page_images = [] - page_template = '
Example Image
' - for page_no, page in doc.pages.items(): - # page_img = page.image.pil_image + html_parts.append("") - if page.image is not None and page.image.pil_image is not None: + html_parts.append("") - page_img = draw_clusters_with_reading_order( - doc=doc, - page_image=page.image.pil_image, - labels=labels, - page_no=page_no, - reading_order=True, - ) + html_parts.append("") + html_parts.append("
") + html_parts.append(f'') + html_parts.append("") + html_parts.append(f"
\n{true_doc_page_body}\n
") + html_parts.append("		") + html_parts.append(f'') + html_parts.append("") + html_parts.append(f"
\n{pred_doc_page_body}\n
") + html_parts.append("
") - page_base64 = from_pil_to_base64(page_img) - page_images.append(page_template.replace("BASE64PAGE", page_base64)) + # Close HTML structure + html_parts.extend(["", ""]) - html_viz = copy.deepcopy(HTML_INSPECTION) - html_viz = html_viz.replace("PREDDOC", html_doc) - html_viz = html_viz.replace("PAGE_IMAGES", "\n".join(page_images)) + # Join with newlines + html_content = "\n".join(html_parts) with open(str(filename), "w") as fw: - fw.write(html_viz) + fw.write(html_content) diff --git a/docs/CVAT_create_groundtruth.md b/docs/CVAT_create_groundtruth.md index e0e9925e..20d3d074 100644 --- a/docs/CVAT_create_groundtruth.md +++ b/docs/CVAT_create_groundtruth.md @@ -17,6 +17,12 @@ Alternatively, you can first create a plain dataset from a folder of PDF or imag docling_eval create-gt --benchmark PlainFiles --dataset-source ./tests/data/files --output-dir ./benchmarks/my_dataset/ ``` +If you want to pre-annotate the dataset (eg with Docling), then use + +```shell +docling_eval create --benchmark PlainFiles --dataset-source ./tests/data/files --output-dir ./benchmarks/my_dataset/ --prediction-provider Docling +``` + ## Pre-annotation with Docling Now you can create the files you need to upload to [CVAT](https://www.cvat.ai/). These files will be created using the `create-cvat` function of the `docling_eval` CLI. @@ -26,19 +32,21 @@ Now you can create the files you need to upload to [CVAT](https://www.cvat.ai/). Usage: docling_eval create-cvat [OPTIONS] -╭─ Options ────────────────────────────────────────────────────────────────────────╮ -│ * --output-dir PATH Output directory [default: None] [required] │ -│ * --gt-dir PATH Dataset source path [default: None] [required] │ -│ --bucket-size INTEGER Size of CVAT tasks [default: 20] │ -│ --help Show this message and exit. │ -╰──────────────────────────────────────────────────────────────────────────────────╯ +╭─ Options ────────────────────────────────────────────────────────────────────────────────────────╮ +│ * --output-dir PATH Output directory [default: None] [required] │ +│ * --gt-dir PATH Dataset source path [default: None] [required] │ +│ --bucket-size INTEGER Size of CVAT tasks [default: 20] │ +│ --use-predictions --no-use-predictions use predictions [default: no-use-predictions] │ +│ --help Show this message and exit. │ +╰──────────────────────────────────────────────────────────────────────────────────────────────────╯ ``` For example, try: ```sh docling_eval create-cvat \ --gt-dir ./benchmarks/DPBench-GT/gt_dataset/test/ \ - --output-dir ./benchmarks/DPBench-CVAT/ + --output-dir ./benchmarks/DPBench-CVAT/ \ + --no-use-predictions ``` In essence, this will read the parquet files and set up a new directory structure (designated output) that has the following layout, diff --git a/docs/examples/package_pdfs.py b/docs/examples/package_pdfs.py index 0db3041b..2be54849 100644 --- a/docs/examples/package_pdfs.py +++ b/docs/examples/package_pdfs.py @@ -17,7 +17,6 @@ save_shard_to_disk, ) from docling_eval.visualisation.constants import HTML_INSPECTION -from docling_eval.visualisation.visualisations import draw_clusters_with_reading_order # Configure logging logging.basicConfig( @@ -164,43 +163,27 @@ def main(): for key, item in conv_results.timings.items(): timings[key] = json.loads(item.model_dump_json()) - html_doc = pred_doc.export_to_html( + filename = viz_dir / f"{os.path.basename(pdf_file)}.html" + pred_doc.save_as_html( + filename=filename, + labels=PRED_HTML_EXPORT_LABELS, image_mode=ImageRefMode.EMBEDDED, - # html_head=HTML_DEFAULT_HEAD_FOR_COMP, - # labels=pred_labels, + split_page_view=True, ) - html_doc = html_doc.replace("'", "'") - - page_images = [] - page_template = '
Example Image
' - for page_no, page in pred_doc.pages.items(): - page_img = page.image.pil_image - - page_img = draw_clusters_with_reading_order( - doc=pred_doc, - page_image=page_img, - labels=PRED_HTML_EXPORT_LABELS, - page_no=page_no, - reading_order=True, - ) - - page_base64 = from_pil_to_base64(page_img) - page_images.append(page_template.replace("BASE64PAGE", page_base64)) - - page = copy.deepcopy(HTML_INSPECTION) - page = page.replace("PREDDOC", html_doc) - page = page.replace("PAGE_IMAGES", "\n".join(page_images)) - - filename = viz_dir / f"{os.path.basename(pdf_file)}.html" - with open(str(filename), "w") as fw: - fw.write(page) + pred_doc, pred_pictures, pred_page_images = extract_images( + document=pred_doc, + pictures_column=BenchMarkColumns.PREDICTION_PICTURES.value, # pictures_column, + page_images_column=BenchMarkColumns.PREDICTION_PAGE_IMAGES.value, # page_images_column, + ) record = { BenchMarkColumns.CONVERTER_VERSION: docling_version(), BenchMarkColumns.STATUS: str(conv_results.status.value), BenchMarkColumns.DOC_ID: str(os.path.basename(pdf_file)), BenchMarkColumns.PREDICTION: json.dumps(pred_doc.export_to_dict()), + BenchMarkColumns.PREDICTION_PAGE_IMAGES: pred_page_images, + BenchMarkColumns.PREDICTION_PICTURES: pred_pictures, BenchMarkColumns.ORIGINAL: get_binary(pdf_file), BenchMarkColumns.MIMETYPE: "application/pdf", BenchMarkColumns.TIMINGS: json.dumps(timings), diff --git a/docs/examples/package_pngs.py b/docs/examples/package_pngs.py index b640dbd0..5ab47c22 100644 --- a/docs/examples/package_pngs.py +++ b/docs/examples/package_pngs.py @@ -19,7 +19,8 @@ save_shard_to_disk, ) from docling_eval.visualisation.constants import HTML_INSPECTION -from docling_eval.visualisation.visualisations import draw_clusters_with_reading_order + +# from docling_eval.visualisation.visualisations import draw_clusters_with_reading_order # Configure logging logging.basicConfig( @@ -173,44 +174,14 @@ def main(): for key, item in conv_results.timings.items(): timings[key] = json.loads(item.model_dump_json()) - html_doc = pred_doc.export_to_html( + filename = viz_dir / f"{os.path.basename(img_file)}.html" + pred_doc.save_as_html( + filename=filename, + labels=PRED_HTML_EXPORT_LABELS, image_mode=ImageRefMode.EMBEDDED, - # html_head=HTML_DEFAULT_HEAD_FOR_COMP, - # labels=pred_labels, + split_page_view=True, ) - html_doc = html_doc.replace("'", "'") - - page_images = [] - page_template = '
Example Image
' - for page_no, page in pred_doc.pages.items(): - - page_img = page.image.pil_image - - # page_img = PILImage.open(png_file) - # assert page.size.width==page_img.width, f"page.size.width==page_img.width {page.size.width}=={page_img.width}" - # assert page.size.height==page_img.height, f"page.size.height==page_img.height {page.size.height}=={page_img.height}" - # page_img.show() - - page_img = draw_clusters_with_reading_order( - doc=pred_doc, - page_image=page_img, - labels=PRED_HTML_EXPORT_LABELS, - page_no=page_no, - reading_order=True, - ) - - page_base64 = from_pil_to_base64(page_img) - page_images.append(page_template.replace("BASE64PAGE", page_base64)) - - page = copy.deepcopy(HTML_INSPECTION) - page = page.replace("PREDDOC", html_doc) - page = page.replace("PAGE_IMAGES", "\n".join(page_images)) - - filename = viz_dir / f"{os.path.basename(img_file)}.html" - with open(str(filename), "w") as fw: - fw.write(page) - pred_doc, pred_pictures, pred_page_images = extract_images( document=pred_doc, pictures_column=BenchMarkColumns.PREDICTION_PICTURES.value, # pictures_column, diff --git a/poetry.lock b/poetry.lock index ba2a192d..5be3a7a7 100644 --- a/poetry.lock +++ b/poetry.lock @@ -1532,14 +1532,14 @@ vlm = ["accelerate (>=1.2.1,<2.0.0) ; sys_platform != \"darwin\" or platform_mac [[package]] name = "docling-core" -version = "2.27.0" +version = "2.28.0" description = "A python library to define and validate data types in Docling." optional = false python-versions = "<4.0,>=3.9" groups = ["main"] files = [ - {file = "docling_core-2.27.0-py3-none-any.whl", hash = "sha256:3bd5ac5e2673b3688c45c9c5beaef9922006c380895933c90393008ae15ed693"}, - {file = "docling_core-2.27.0.tar.gz", hash = "sha256:b0bee5501a6f7b0a8b888f47ba6e51f82cd559b6afd56bf0c9bce7f71168f87d"}, + {file = "docling_core-2.28.0-py3-none-any.whl", hash = "sha256:f1a01446996b90c4c151ec0ad247283888e6372f9dac0e356d06f8b9838ca4ca"}, + {file = "docling_core-2.28.0.tar.gz", hash = "sha256:16a762c251063839d7b20624cd6f89c2488872377f8546b037e604606014fb66"}, ] [package.dependencies] @@ -1550,7 +1550,7 @@ pandas = ">=2.1.4,<3.0.0" pillow = ">=10.0.0,<12.0.0" pydantic = ">=2.6.0,<2.10.0 || >2.10.0,<2.10.1 || >2.10.1,<2.10.2 || >2.10.2,<3.0.0" pyyaml = ">=5.1,<7.0.0" -semchunk = {version = ">=2.2.0,<3.0.0", optional = true, markers = "extra == \"chunking\""} +semchunk = {version = ">=2.2.0,<3.0.0", optional = true, markers = "extra == \"chunking\" or extra == \"chunking-openai\""} tabulate = ">=0.9.0,<0.10.0" transformers = {version = ">=4.34.0,<5.0.0", optional = true, markers = "extra == \"chunking\""} typer = ">=0.12.5,<0.16.0" @@ -1558,6 +1558,7 @@ typing-extensions = ">=4.12.2,<5.0.0" [package.extras] chunking = ["semchunk (>=2.2.0,<3.0.0)", "transformers (>=4.34.0,<5.0.0)"] +chunking-openai = ["semchunk (>=2.2.0,<3.0.0)", "tiktoken (>=0.9.0,<0.10.0)"] [[package]] name = "docling-ibm-models" @@ -4294,8 +4295,8 @@ files = [ numpy = [ {version = ">=1.21.4", markers = "python_version >= \"3.10\" and platform_system == \"Darwin\""}, {version = ">=1.21.2", markers = "platform_system != \"Darwin\" and python_version >= \"3.10\""}, - {version = ">=1.23.5", markers = "python_version >= \"3.11\""}, {version = ">=1.26.0", markers = "python_version >= \"3.12\""}, + {version = ">=1.23.5", markers = "python_version >= \"3.11\""}, ] [[package]] @@ -4335,43 +4336,31 @@ groups = ["main"] files = [ {file = "pandas-2.2.3-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:1948ddde24197a0f7add2bdc4ca83bf2b1ef84a1bc8ccffd95eda17fd836ecb5"}, {file = "pandas-2.2.3-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:381175499d3802cde0eabbaf6324cce0c4f5d52ca6f8c377c29ad442f50f6348"}, - {file = "pandas-2.2.3-cp310-cp310-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:d9c45366def9a3dd85a6454c0e7908f2b3b8e9c138f5dc38fed7ce720d8453ed"}, {file = "pandas-2.2.3-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:86976a1c5b25ae3f8ccae3a5306e443569ee3c3faf444dfd0f41cda24667ad57"}, - {file = "pandas-2.2.3-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:b8661b0238a69d7aafe156b7fa86c44b881387509653fdf857bebc5e4008ad42"}, {file = "pandas-2.2.3-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:37e0aced3e8f539eccf2e099f65cdb9c8aa85109b0be6e93e2baff94264bdc6f"}, {file = "pandas-2.2.3-cp310-cp310-win_amd64.whl", hash = "sha256:56534ce0746a58afaf7942ba4863e0ef81c9c50d3f0ae93e9497d6a41a057645"}, {file = "pandas-2.2.3-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:66108071e1b935240e74525006034333f98bcdb87ea116de573a6a0dccb6c039"}, {file = "pandas-2.2.3-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:7c2875855b0ff77b2a64a0365e24455d9990730d6431b9e0ee18ad8acee13dbd"}, - {file = "pandas-2.2.3-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:cd8d0c3be0515c12fed0bdbae072551c8b54b7192c7b1fda0ba56059a0179698"}, {file = "pandas-2.2.3-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:c124333816c3a9b03fbeef3a9f230ba9a737e9e5bb4060aa2107a86cc0a497fc"}, - {file = "pandas-2.2.3-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:63cc132e40a2e084cf01adf0775b15ac515ba905d7dcca47e9a251819c575ef3"}, {file = "pandas-2.2.3-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:29401dbfa9ad77319367d36940cd8a0b3a11aba16063e39632d98b0e931ddf32"}, {file = "pandas-2.2.3-cp311-cp311-win_amd64.whl", hash = "sha256:3fc6873a41186404dad67245896a6e440baacc92f5b716ccd1bc9ed2995ab2c5"}, {file = "pandas-2.2.3-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:b1d432e8d08679a40e2a6d8b2f9770a5c21793a6f9f47fdd52c5ce1948a5a8a9"}, {file = "pandas-2.2.3-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:a5a1595fe639f5988ba6a8e5bc9649af3baf26df3998a0abe56c02609392e0a4"}, - {file = "pandas-2.2.3-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:5de54125a92bb4d1c051c0659e6fcb75256bf799a732a87184e5ea503965bce3"}, {file = "pandas-2.2.3-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:fffb8ae78d8af97f849404f21411c95062db1496aeb3e56f146f0355c9989319"}, - {file = "pandas-2.2.3-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:6dfcb5ee8d4d50c06a51c2fffa6cff6272098ad6540aed1a76d15fb9318194d8"}, {file = "pandas-2.2.3-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:062309c1b9ea12a50e8ce661145c6aab431b1e99530d3cd60640e255778bd43a"}, {file = "pandas-2.2.3-cp312-cp312-win_amd64.whl", hash = "sha256:59ef3764d0fe818125a5097d2ae867ca3fa64df032331b7e0917cf5d7bf66b13"}, {file = "pandas-2.2.3-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:f00d1345d84d8c86a63e476bb4955e46458b304b9575dcf71102b5c705320015"}, {file = "pandas-2.2.3-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:3508d914817e153ad359d7e069d752cdd736a247c322d932eb89e6bc84217f28"}, - {file = "pandas-2.2.3-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:22a9d949bfc9a502d320aa04e5d02feab689d61da4e7764b62c30b991c42c5f0"}, {file = "pandas-2.2.3-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:f3a255b2c19987fbbe62a9dfd6cff7ff2aa9ccab3fc75218fd4b7530f01efa24"}, - {file = "pandas-2.2.3-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:800250ecdadb6d9c78eae4990da62743b857b470883fa27f652db8bdde7f6659"}, {file = "pandas-2.2.3-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:6374c452ff3ec675a8f46fd9ab25c4ad0ba590b71cf0656f8b6daa5202bca3fb"}, {file = "pandas-2.2.3-cp313-cp313-win_amd64.whl", hash = "sha256:61c5ad4043f791b61dd4752191d9f07f0ae412515d59ba8f005832a532f8736d"}, {file = "pandas-2.2.3-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:3b71f27954685ee685317063bf13c7709a7ba74fc996b84fc6821c59b0f06468"}, {file = "pandas-2.2.3-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:38cf8125c40dae9d5acc10fa66af8ea6fdf760b2714ee482ca691fc66e6fcb18"}, - {file = "pandas-2.2.3-cp313-cp313t-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:ba96630bc17c875161df3818780af30e43be9b166ce51c9a18c1feae342906c2"}, {file = "pandas-2.2.3-cp313-cp313t-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:1db71525a1538b30142094edb9adc10be3f3e176748cd7acc2240c2f2e5aa3a4"}, - {file = "pandas-2.2.3-cp313-cp313t-musllinux_1_2_aarch64.whl", hash = "sha256:15c0e1e02e93116177d29ff83e8b1619c93ddc9c49083f237d4312337a61165d"}, {file = "pandas-2.2.3-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:ad5b65698ab28ed8d7f18790a0dc58005c7629f227be9ecc1072aa74c0c1d43a"}, {file = "pandas-2.2.3-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:bc6b93f9b966093cb0fd62ff1a7e4c09e6d546ad7c1de191767baffc57628f39"}, {file = "pandas-2.2.3-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:5dbca4c1acd72e8eeef4753eeca07de9b1db4f398669d5994086f788a5d7cc30"}, - {file = "pandas-2.2.3-cp39-cp39-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:8cd6d7cc958a3910f934ea8dbdf17b2364827bb4dafc38ce6eef6bb3d65ff09c"}, {file = "pandas-2.2.3-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:99df71520d25fade9db7c1076ac94eb994f4d2673ef2aa2e86ee039b6746d20c"}, - {file = "pandas-2.2.3-cp39-cp39-musllinux_1_2_aarch64.whl", hash = "sha256:31d0ced62d4ea3e231a9f228366919a5ea0b07440d9d4dac345376fd8e1477ea"}, {file = "pandas-2.2.3-cp39-cp39-musllinux_1_2_x86_64.whl", hash = "sha256:7eee9e7cea6adf3e3d24e304ac6b8300646e2a5d1cd3a3c2abed9101b0846761"}, {file = "pandas-2.2.3-cp39-cp39-win_amd64.whl", hash = "sha256:4850ba03528b6dd51d6c5d273c46f183f39a9baf3f0143e566b89450965b105e"}, {file = "pandas-2.2.3.tar.gz", hash = "sha256:4f18ba62b61d7e192368b84517265a99b4d7ee8912f8708660fb4a366cc82667"}, @@ -4380,8 +4369,8 @@ files = [ [package.dependencies] numpy = [ {version = ">=1.22.4", markers = "python_version < \"3.11\""}, - {version = ">=1.23.2", markers = "python_version == \"3.11\""}, {version = ">=1.26.0", markers = "python_version >= \"3.12\""}, + {version = ">=1.23.2", markers = "python_version == \"3.11\""}, ] python-dateutil = ">=2.8.2" pytz = ">=2020.1" @@ -7100,11 +7089,6 @@ optional = false python-versions = ">=3.9" groups = ["main"] files = [ - {file = "torchvision-0.21.0-1-cp310-cp310-manylinux_2_28_aarch64.whl", hash = "sha256:5568c5a1ff1b2ec33127b629403adb530fab81378d9018ca4ed6508293f76e2b"}, - {file = "torchvision-0.21.0-1-cp311-cp311-manylinux_2_28_aarch64.whl", hash = "sha256:ff96666b94a55e802ea6796cabe788541719e6f4905fc59c380fed3517b6a64d"}, - {file = "torchvision-0.21.0-1-cp312-cp312-manylinux_2_28_aarch64.whl", hash = "sha256:ffa2a16499508fe6798323e455f312c7c55f2a88901c9a7c0fb1efa86cf7e327"}, - {file = "torchvision-0.21.0-1-cp313-cp313-manylinux_2_28_aarch64.whl", hash = "sha256:7e9e9afa150e40cd2a8f0701c43cb82a8d724f512896455c0918b987f94b84a4"}, - {file = "torchvision-0.21.0-1-cp39-cp39-manylinux_2_28_aarch64.whl", hash = "sha256:669575b290ec27304569e188a960d12b907d5173f9cd65e86621d34c4e5b6c30"}, {file = "torchvision-0.21.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:044ea420b8c6c3162a234cada8e2025b9076fa82504758cd11ec5d0f8cd9fa37"}, {file = "torchvision-0.21.0-cp310-cp310-manylinux1_x86_64.whl", hash = "sha256:b0c0b264b89ab572888244f2e0bad5b7eaf5b696068fc0b93e96f7c3c198953f"}, {file = "torchvision-0.21.0-cp310-cp310-manylinux_2_28_aarch64.whl", hash = "sha256:54815e0a56dde95cc6ec952577f67e0dc151eadd928e8d9f6a7f821d69a4a734"}, @@ -7996,4 +7980,4 @@ hyperscalers = ["azure-ai-formrecognizer", "azure-common", "azure-core", "boto3" [metadata] lock-version = "2.1" python-versions = "^3.10" -content-hash = "6e4e1c7151fc551b4f3f2bd94f3c6906dd554384484da38b0e5bcc5603c59fdd" +content-hash = "ad48608fca439c925fd79021a7323b74448f506595a7c79f652a07e9538dbd13" diff --git a/pyproject.toml b/pyproject.toml index 60afaa23..fc672ad8 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -25,7 +25,7 @@ packages = [{include = "docling_eval"}] # actual dependencies: ###################### python = "^3.10" -docling-core = "^2.27.0" +docling-core = "^2.28.0" pydantic = "^2.0.0" lxml = "^5.3.0" datasets = "^3.2.0"