chore: Refactor docling-eval to improve the overall codebase structure

docling_eval/benchmarks/constants.py

@@ -3,7 +3,7 @@
 
 class BenchMarkColumns(str, Enum):
     CONVERTER_TYPE = "converter_type"
-    DOCLING_VERSION = "docling_version"
+    CONVERTER_VERSION = "converter_version"
     DOCLING_PIPELINE = "docling_pipeline"
 
     STATUS = "status"

docling_eval/benchmarks/cvat_annotation/create.py

@@ -1,66 +1,50 @@
 import argparse
-import copy
 import glob
 import json
 import logging
 import os
 from pathlib import Path
-from typing import Dict, Generator, Iterator, List, Optional, Tuple, cast
+from typing import Dict, Iterator, List, Optional, Tuple, cast
 
 import xmltodict  # type: ignore[import]
-from datasets import Dataset, load_dataset
 from docling_core.types.doc.base import BoundingBox, CoordOrigin, Size
 from docling_core.types.doc.document import (
-    DocItem,
     DoclingDocument,
     FloatingItem,
     GraphData,
     ImageRef,
     PageItem,
-    PictureItem,
     ProvenanceItem,
     TableData,
     TableItem,
 )
-from docling_core.types.doc.labels import (
-    DocItemLabel,
-    GroupLabel,
-    PictureClassificationLabel,
-    TableCellLabel,
-)
+from docling_core.types.doc.labels import DocItemLabel
 from docling_parse.pdf_parsers import pdf_parser_v2  # type: ignore[import]
 from PIL import Image  # as PILImage
 from tqdm import tqdm  # type: ignore
 
-from docling_eval.benchmarks.constants import BenchMarkColumns, EvaluationModality
+from docling_eval.benchmarks.constants import (
+    BenchMarkColumns,
+    ConverterTypes,
+    EvaluationModality,
+)
 from docling_eval.benchmarks.cvat_annotation.utils import (
-    AnnotatedDoc,
     AnnotatedImage,
-    AnnotationBBox,
-    AnnotationLine,
     AnnotationOverview,
     BenchMarkDirs,
-    DocLinkLabel,
-    TableComponentLabel,
-    rgb_to_hex,
 )
 from docling_eval.benchmarks.utils import (
-    draw_clusters_with_reading_order,
-    get_binhash,
-    save_comparison_html_with_clusters,
-    save_inspection_html,
-    write_datasets_info,
-)
-from docling_eval.docling.conversion import create_docling_converter
-from docling_eval.docling.utils import (
     crop_bounding_box,
     docling_version,
     extract_images,
     from_pil_to_base64uri,
     get_binary,
-    insert_images,
+    get_binhash,
     save_shard_to_disk,
+    write_datasets_info,
 )
+from docling_eval.converters.conversion import create_pdf_docling_converter
+from docling_eval.visualisation.visualisations import save_comparison_html_with_clusters
 
 # from pydantic import
 
@@ -602,14 +586,14 @@ def create_true_document(basename: str, annot: dict, desc: AnnotatedImage):
         img_height = page_image.height
 
         if pred_doc.pages[page_no] is None:
-            logging.error(f"Page item is None, skipping ...")
+            logging.error("Page item is None, skipping ...")
             continue
 
         pred_page_item = pred_doc.pages[page_no]
 
         pred_page_imageref = pred_page_item.image
         if pred_page_imageref is None:
-            logging.error(f"Page ImageRef is None, skipping ...")
+            logging.error("Page ImageRef is None, skipping ...")
             continue
 
         assert pred_page_imageref.size.width == img_width
@@ -644,14 +628,14 @@ def create_true_document(basename: str, annot: dict, desc: AnnotatedImage):
         page_no = 1
 
         if (page_no not in true_doc.pages) or (true_doc.pages[page_no] is None):
-            logging.error(f"Page item is None, skipping ...")
+            logging.error("Page item is None, skipping ...")
             continue
 
         true_page_item = true_doc.pages[page_no]
 
         true_page_imageref = true_page_item.image
         if true_page_imageref is None:
-            logging.error(f"Page ImageRef is None, skipping ...")
+            logging.error("Page ImageRef is None, skipping ...")
             continue
 
         true_page_pilimage = true_page_imageref.pil_image
@@ -943,7 +927,7 @@ def create_layout_dataset_from_annotations(
 
     # Create Converter
     image_scale = 2.0
-    doc_converter = create_docling_converter(page_image_scale=image_scale)
+    doc_converter = create_pdf_docling_converter(page_image_scale=image_scale)
 
     records = []
     for basename, desc, true_doc in tqdm(
@@ -1000,7 +984,8 @@ def create_layout_dataset_from_annotations(
             )
 
         record = {
-            BenchMarkColumns.DOCLING_VERSION: docling_version(),
+            BenchMarkColumns.CONVERTER_TYPE: ConverterTypes.DOCLING,
+            BenchMarkColumns.CONVERTER_VERSION: docling_version(),
             BenchMarkColumns.STATUS: str(conv_results.status),
             BenchMarkColumns.DOC_ID: str(basename),
             BenchMarkColumns.DOC_PATH: str(basename),

docling_eval/benchmarks/cvat_annotation/create_dataset_from_pdfs.py

@@ -1,33 +1,21 @@
 import argparse
-import copy
 import glob
 import json
-import logging
 import os
 from pathlib import Path
-from typing import Dict, List, Tuple
 
 from datasets import Features
 from datasets import Image as Features_Image
 from datasets import Sequence, Value
 
 from docling_eval.benchmarks.constants import BenchMarkColumns
 from docling_eval.benchmarks.utils import (
-    add_pages_to_true_doc,
-    convert_html_table_into_docling_tabledata,
-    save_comparison_html,
-    save_comparison_html_with_clusters,
-    write_datasets_info,
-)
-from docling_eval.docling.conversion import create_docling_converter
-from docling_eval.docling.utils import (
-    crop_bounding_box,
     docling_version,
     extract_images,
-    from_pil_to_base64uri,
     get_binary,
     save_shard_to_disk,
 )
+from docling_eval.converters.conversion import create_pdf_docling_converter
 
 
 def parse_args():
@@ -61,7 +49,7 @@ def _write_datasets_info(
 ):
     features = Features(
         {
-            BenchMarkColumns.DOCLING_VERSION: Value("string"),
+            BenchMarkColumns.CONVERTER_VERSION: Value("string"),
             BenchMarkColumns.STATUS: Value("string"),
             BenchMarkColumns.DOC_ID: Value("string"),
             # BenchMarkColumns.DOC_PATH: Value("string"),
@@ -113,7 +101,7 @@ def main():
             os.makedirs(_)
 
     # Create Converter
-    doc_converter = create_docling_converter(
+    doc_converter = create_pdf_docling_converter(
         page_image_scale=image_scale, artifacts_path=artifacts_path
     )
 
@@ -134,7 +122,7 @@ def main():
         )
 
         record = {
-            BenchMarkColumns.DOCLING_VERSION: docling_version(),
+            BenchMarkColumns.CONVERTER_VERSION: docling_version(),
             BenchMarkColumns.STATUS: str(conv_results.status),
             BenchMarkColumns.DOC_ID: str(os.path.basename(pdf_path)),
             BenchMarkColumns.PREDICTION: json.dumps(pred_doc.export_to_dict()),

docling_eval/benchmarks/cvat_annotation/eval.py

@@ -1,13 +1,4 @@
-import argparse
 import logging
-import os
-from pathlib import Path
-
-from huggingface_hub import snapshot_download
-from tabulate import tabulate  # type: ignore
-
-from docling_eval.benchmarks.constants import BenchMarkNames, EvaluationModality
-from docling_eval.cli.main import evaluate, visualise
 
 # Configure logging
 logging.basicConfig(

docling_eval/benchmarks/cvat_annotation/preannotate.py

@@ -1,43 +1,29 @@
 import argparse
-import copy
 import glob
 import json
 import logging
 import os
 from pathlib import Path
-from typing import Dict, List, Tuple
-
-from datasets import Dataset, load_dataset
-from docling_core.types.doc.base import BoundingBox, CoordOrigin, ImageRefMode, Size
-from docling_core.types.doc.document import (
-    DocItem,
-    DoclingDocument,
-    PictureItem,
-    TableItem,
-)
-from docling_core.types.doc.labels import (
-    DocItemLabel,
-    GroupLabel,
-    PictureClassificationLabel,
-    TableCellLabel,
-)
-from pydantic import BaseModel
+from typing import List
+
+from datasets import load_dataset
+from docling_core.types.doc.base import BoundingBox, ImageRefMode
+from docling_core.types.doc.document import DocItem, DoclingDocument
+from docling_core.types.doc.labels import DocItemLabel, PictureClassificationLabel
 from tqdm import tqdm  # type: ignore
 
 from docling_eval.benchmarks.constants import BenchMarkColumns
 from docling_eval.benchmarks.cvat_annotation.utils import (
     AnnotatedDoc,
     AnnotatedImage,
     AnnotationBBox,
-    AnnotationLine,
     AnnotationOverview,
     BenchMarkDirs,
     DocLinkLabel,
     TableComponentLabel,
     rgb_to_hex,
 )
-from docling_eval.benchmarks.utils import get_binhash
-from docling_eval.docling.utils import insert_images
+from docling_eval.benchmarks.utils import get_binhash, insert_images
 
 # Configure logging
 logging.basicConfig(
@@ -252,11 +238,11 @@ def create_cvat_preannotation_file_for_single_page(
                     annotated_image.page_nos = [page_no]
                     overview.img_annotations[filename] = annotated_image
                 else:
-                    logging.warning(f"missing pillow image of the page, skipping ...")
+                    logging.warning("missing pillow image of the page, skipping ...")
                     continue
 
             else:
-                logging.warning(f"missing image-ref of the page, skipping ...")
+                logging.warning("missing image-ref of the page, skipping ...")
                 continue
 
             page_bboxes: List[AnnotationBBox] = []

docling_eval/benchmarks/cvat_annotation/utils.py

@@ -6,13 +6,8 @@
 from pathlib import Path
 from typing import Dict, List, Tuple
 
-from docling_core.types.doc.base import BoundingBox, CoordOrigin, ImageRefMode, Size
-from docling_core.types.doc.labels import (
-    DocItemLabel,
-    GroupLabel,
-    PictureClassificationLabel,
-    TableCellLabel,
-)
+from docling_core.types.doc.base import BoundingBox
+from docling_core.types.doc.labels import DocItemLabel
 from pydantic import BaseModel