fix(ocr): make the OCR render scale configurable instead of hardcoded

docling/datamodel/pipeline_options.py

@@ -216,6 +216,20 @@ class OcrOptions(BaseOptions):
         ),
     ]
 
+    scale: Annotated[
+        float,
+        Field(
+            description=(
+                "Image scale multiplier applied before running OCR. The page is "
+                "rendered at 72 DPI times this factor, so the default 3 yields "
+                "216 DPI. Lower it when the source image is already high "
+                "resolution and upscaling degrades recognition."
+            ),
+            examples=[1.0, 3.0],
+            gt=0.0,
+        ),
+    ] = 3.0
+
     # Deprecated: superseded by `OcrMode.FULL_PAGE`. Kept for backwards compatibility
     # When set to True it forces `mode` to FULL_PAGE
     force_full_page_ocr: Annotated[

docling/models/stages/ocr/easyocr_model.py

@@ -88,7 +88,8 @@ def __init__(
         )
         self.options: EasyOcrOptions
 
-        self.scale = 3  # multiplier for 72 dpi == 216 dpi.
+        # multiplier for 72 dpi; the default 3.0 == 216 dpi.
+        self.scale = self.options.scale
 
         if self.enabled:
             try:

docling/models/stages/ocr/nemotron_ocr_model.py

@@ -120,7 +120,8 @@ def __init__(
             accelerator_options=accelerator_options,
         )
         self.options: NemotronOcrOptions
-        self.scale = 3  # multiplier for 72 dpi == 216 dpi.
+        # multiplier for 72 dpi; the default 3.0 == 216 dpi.
+        self.scale = self.options.scale
 
         if self.enabled:
             self.validate_runtime(accelerator_options=accelerator_options)
@@ -246,7 +247,7 @@ def _prediction_to_cell(
         ocr_rect: BoundingBox,
         image_width: int,
         image_height: int,
-        scale: int,
+        scale: float,
     ) -> TextCell:
         # `nemotron_ocr` returns normalized `left/right` and an inverted
         # pair `lower/upper`, where `lower` is the top Y and `upper` is the

docling/models/stages/ocr/ocr_mac_model.py

@@ -38,7 +38,8 @@ def __init__(
         )
         self.options: OcrMacOptions
 
-        self.scale = 3  # multiplier for 72 dpi == 216 dpi.
+        # multiplier for 72 dpi; the default 3.0 == 216 dpi.
+        self.scale = self.options.scale
 
         if self.enabled:
             if "darwin" != sys.platform:

docling/models/stages/ocr/rapid_ocr_model.py

@@ -325,7 +325,8 @@ def __init__(
         )
         self.options: RapidOcrOptions
 
-        self.scale = 3  # multiplier for 72 dpi == 216 dpi.
+        # multiplier for 72 dpi; the default 3.0 == 216 dpi.
+        self.scale = self.options.scale
 
         if self.enabled:
             try:

docling/models/stages/ocr/tesseract_ocr_cli_model.py

@@ -52,7 +52,8 @@ def __init__(
         )
         self.options: TesseractCliOcrOptions
 
-        self.scale = 3  # multiplier for 72 dpi == 216 dpi.
+        # multiplier for 72 dpi; the default 3.0 == 216 dpi.
+        self.scale = self.options.scale
 
         self._name: Optional[str] = None
         self._version: Optional[str] = None

docling/models/stages/ocr/tesseract_ocr_model.py

@@ -42,7 +42,8 @@ def __init__(
         )
         self.options: TesseractOcrOptions
         self._is_auto: bool = "auto" in self.options.lang
-        self.scale = 3  # multiplier for 72 dpi == 216 dpi.
+        # multiplier for 72 dpi; the default 3.0 == 216 dpi.
+        self.scale = self.options.scale
         self.reader = None
         self.script_readers: dict[str, tesserocr.PyTessBaseAPI] = {}
 

tests/test_options.py

@@ -3,6 +3,7 @@
 from unittest.mock import Mock, patch
 
 import pytest
+from pydantic import ValidationError
 
 from docling.backend.docling_parse_backend import (
     DoclingParseDocumentBackend,
@@ -21,17 +22,21 @@
     ApiKserveV2ImageClassificationEngineOptions,
 )
 from docling.datamodel.pipeline_options import (
+    EasyOcrOptions,
     NemotronOcrOptions,
     PdfPipelineOptions,
     TableFormerMode,
+    TesseractCliOcrOptions,
 )
 from docling.document_converter import (
     ConversionError,
     DocumentConverter,
     PdfFormatOption,
 )
 from docling.models.factories import get_ocr_factory
+from docling.models.stages.ocr.easyocr_model import EasyOcrModel
 from docling.models.stages.ocr.nemotron_ocr_model import NemotronOcrModel
+from docling.models.stages.ocr.tesseract_ocr_cli_model import TesseractOcrCliModel
 from docling.pipeline.legacy_standard_pdf_pipeline import LegacyStandardPdfPipeline
 
 
@@ -455,6 +460,33 @@ def test_page_error_carries_page_no(test_doc_path):
         assert not err.error_message.startswith("Page ")
 
 
+def test_ocr_scale_is_configurable():
+    """The OCR render scale comes from the options instead of a hardcoded 3."""
+    assert TesseractCliOcrOptions().scale == 3.0
+    assert EasyOcrOptions().scale == 3.0
+
+    accelerator_options = AcceleratorOptions()
+
+    tesseract_model = TesseractOcrCliModel(
+        enabled=False,
+        artifacts_path=None,
+        options=TesseractCliOcrOptions(scale=1.0),
+        accelerator_options=accelerator_options,
+    )
+    assert tesseract_model.scale == 1.0
+
+    easyocr_model = EasyOcrModel(
+        enabled=False,
+        artifacts_path=None,
+        options=EasyOcrOptions(scale=1.5),
+        accelerator_options=accelerator_options,
+    )
+    assert easyocr_model.scale == 1.5
+
+    with pytest.raises(ValidationError):
+        TesseractCliOcrOptions(scale=0)
+
+
 def test_nemotron_ocr_backend_registration():
     factory = get_ocr_factory(allow_external_plugins=False)