polytext 0.2.8b4__tar.gz → 0.2.8b5__tar.gz

{polytext-0.2.8b4 → polytext-0.2.8b5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: polytext
-Version: 0.2.8b4
+Version: 0.2.8b5
 Summary: Python utilities to simplify document files management
 Home-page: https://github.com/docsity/polytext
 Author: Matteo Senardi

{polytext-0.2.8b4 → polytext-0.2.8b5}/polytext/converter/document_ocr_to_text.py

@@ -115,6 +115,7 @@ def get_document_ocr(
     max_output_tokens: int | None = None,
     include_image_descriptions: bool = False,
     allow_partial_ocr_failures: bool = False,
+    ocr_render_dpi: int | None = None,
 ):
     """
     Convenience function to extract text from an image file using OCR, optionally formatted as Markdown.
@@ -141,6 +142,8 @@ def get_document_ocr(
         allow_partial_ocr_failures (bool, optional): If True, pages that still
             fail OCR after all retries are recorded inline instead of aborting
             the whole document extraction. Defaults to False.
+        ocr_render_dpi (int | None, optional): Render PDF pages at this DPI before
+            OCR. When omitted, preserves PyMuPDF's default rendering behavior.
 
     Returns:
         dict: Dictionary containing the OCR results and metadata.
@@ -155,6 +158,7 @@ def get_document_ocr(
         max_output_tokens=max_output_tokens,
         include_image_descriptions=include_image_descriptions,
         allow_partial_ocr_failures=allow_partial_ocr_failures,
+        ocr_render_dpi=ocr_render_dpi,
     )
     return converter.get_document_ocr(document_for_ocr)
 
@@ -164,7 +168,8 @@ class DocumentOCRToTextConverter:
                  page_range=None, timeout_minutes: int = None, fallback_stage: int = 0,
                  max_output_tokens: int | None = None, include_image_descriptions: bool = False,
                  prompt_variant: str = OCR_PROMPT_VARIANT_DEFAULT,
-                 allow_partial_ocr_failures: bool = False):
+                 allow_partial_ocr_failures: bool = False,
+                 ocr_render_dpi: int | None = None):
         """
         Initialize the DocumentOCRToTextConverter class with specified OCR model and formatting options.
 
@@ -192,6 +197,8 @@ class DocumentOCRToTextConverter:
             allow_partial_ocr_failures (bool, optional): If True, pages that still
                 fail OCR after all retries are recorded inline instead of aborting
                 the whole document extraction. Defaults to False.
+            ocr_render_dpi (int | None, optional): Render PDF pages at this DPI
+                before OCR. Defaults to PyMuPDF's native rendering behavior.
 
         Raises:
             OSError: If temp directory creation fails
@@ -207,6 +214,7 @@ class DocumentOCRToTextConverter:
         self.include_image_descriptions = include_image_descriptions
         self.prompt_variant = prompt_variant
         self.allow_partial_ocr_failures = allow_partial_ocr_failures
+        self.ocr_render_dpi = ocr_render_dpi
         requested_output_tokens = OCR_MAX_OUTPUT_TOKENS if max_output_tokens is None else max_output_tokens
         self.max_output_tokens = max(requested_output_tokens, OCR_MIN_OUTPUT_TOKENS)
         self.fallback_stage = fallback_stage
@@ -297,6 +305,7 @@ class DocumentOCRToTextConverter:
             include_image_descriptions=self.include_image_descriptions,
             prompt_variant=resolved_prompt_variant,
             allow_partial_ocr_failures=self.allow_partial_ocr_failures,
+            ocr_render_dpi=self.ocr_render_dpi,
         )
         result = fallback_converter.get_ocr(
             file_for_ocr=file_for_ocr,
@@ -553,7 +562,10 @@ class DocumentOCRToTextConverter:
 
             try:
                 # Convert page to image
-                pix = page.get_pixmap()
+                if self.ocr_render_dpi is None:
+                    pix = page.get_pixmap()
+                else:
+                    pix = page.get_pixmap(dpi=self.ocr_render_dpi, alpha=False)
                 pix.save(temp_image_path)
 
                 # Perform OCR on the page

{polytext-0.2.8b4 → polytext-0.2.8b5}/polytext/loader/base.py

@@ -92,6 +92,7 @@ class BaseLoader:
                  ocr_model: str = "gpt-5-mini", timeout_minutes: int | None = None,
                  include_image_descriptions: bool | None = None,
                  force_ocr: bool = False,
+                 ocr_render_dpi: int | None = None,
                  **kwargs):
         """
         Initialize the BaseLoader with cloud storage and LLM configurations.
@@ -112,6 +113,8 @@ class BaseLoader:
                 If None, defaults from OCR_INCLUDE_IMAGE_DESCRIPTIONS. Defaults to None.
             force_ocr (bool, optional): If True, supported document files are routed to
                 OCRLoader instead of the standard DocumentLoader. Defaults to False.
+            ocr_render_dpi (int | None, optional): Render document pages at this DPI
+                before OCR. Defaults to the existing renderer behavior.
              **kwargs: Additional keyword arguments to pass to the underlying loader or extraction logic.
                 - target_size (int, optional): Target file size in bytes. Defaults to 1MB
                 - source (str): Source of the document. Must be either "cloud" or "local"
@@ -136,6 +139,7 @@ class BaseLoader:
             else include_image_descriptions
         )
         self.force_ocr = force_ocr
+        self.ocr_render_dpi = ocr_render_dpi
         self.kwargs = kwargs
         self.target_size = kwargs.get("target_size", 1)
         self.source = kwargs.get("source", "cloud")
@@ -422,7 +426,7 @@ class BaseLoader:
             file_extension = file_extension.lower()
 
         if is_document_fallback:
-            return DocumentOCRLoader(llm_api_key=llm_api_key, markdown_output=self.markdown_output, temp_dir=self.temp_dir, timeout_minutes=self.timeout_minutes, ocr_provider=self.provider, ocr_model=self.ocr_model, include_image_descriptions=self.include_image_descriptions, **kwargs)
+            return DocumentOCRLoader(llm_api_key=llm_api_key, markdown_output=self.markdown_output, temp_dir=self.temp_dir, timeout_minutes=self.timeout_minutes, ocr_provider=self.provider, ocr_model=self.ocr_model, include_image_descriptions=self.include_image_descriptions, ocr_render_dpi=self.ocr_render_dpi, **kwargs)
 
         if file_extension in [".xml", ".xbrl"]:
             return XmlXbrlLoader(temp_dir=self.temp_dir, markdown_output=self.markdown_output, **kwargs)
@@ -458,6 +462,7 @@ class BaseLoader:
                         ocr_provider=self.provider,
                         ocr_model=self.ocr_model,
                         include_image_descriptions=self.include_image_descriptions,
+                        ocr_render_dpi=self.ocr_render_dpi,
                         **document_kwargs,
                     )
 

{polytext-0.2.8b4 → polytext-0.2.8b5}/polytext/loader/document_ocr.py

@@ -47,6 +47,7 @@ class DocumentOCRLoader:
                  ocr_model: str | None = None,
                  include_image_descriptions: bool = False,
                  allow_partial_ocr_failures: bool = False,
+                 ocr_render_dpi: int | None = None,
                  **kwargs
                  ):
         """
@@ -86,6 +87,8 @@ class DocumentOCRLoader:
             allow_partial_ocr_failures (bool, optional): If True, pages that still
                 fail OCR after all retries are recorded inline instead of aborting
                 the whole document extraction. Defaults to False.
+            ocr_render_dpi (int | None, optional): Render PDF pages at this DPI
+                before OCR. Defaults to the existing PyMuPDF rendering behavior.
             **kwargs:
                 max_output_tokens (int, optional): Maximum Gemini output tokens for
                     Google document OCR generation.
@@ -110,6 +113,7 @@ class DocumentOCRLoader:
         self.ocr_model = ocr_model
         self.include_image_descriptions = include_image_descriptions
         self.allow_partial_ocr_failures = allow_partial_ocr_failures
+        self.ocr_render_dpi = ocr_render_dpi
         self.max_output_tokens = kwargs.get("max_output_tokens")
 
         # Set up custom temp directory
@@ -277,6 +281,7 @@ class DocumentOCRLoader:
                 max_output_tokens=self.max_output_tokens,
                 include_image_descriptions=self.include_image_descriptions,
                 allow_partial_ocr_failures=self.allow_partial_ocr_failures,
+                ocr_render_dpi=self.ocr_render_dpi,
             )
 
         result_dict["type"] = self.type

{polytext-0.2.8b4 → polytext-0.2.8b5}/polytext.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: polytext
-Version: 0.2.8b4
+Version: 0.2.8b5
 Summary: Python utilities to simplify document files management
 Home-page: https://github.com/docsity/polytext
 Author: Matteo Senardi

{polytext-0.2.8b4 → polytext-0.2.8b5}/setup.py

@@ -51,7 +51,7 @@ def get_requirements(*requirements_file):
 
 setup(
     name='polytext',
-    version='0.2.8b4',
+    version='0.2.8b5',
     url='https://github.com/docsity/polytext',
     # download_url='https://github.com/pualien/py-polytext/archive/0.1.23.tar.gz',
     license='MIT',

{polytext-0.2.8b4 → polytext-0.2.8b5}/tests/test_ocr_fallbacks.py

@@ -88,8 +88,10 @@ class _FakePixmap:
 class _FakePage:
     def __init__(self, payload: bytes = b"fake-page-image"):
         self.payload = payload
+        self.pixmap_calls = []
 
-    def get_pixmap(self):
+    def get_pixmap(self, **kwargs):
+        self.pixmap_calls.append(kwargs)
         return _FakePixmap(payload=self.payload)
 
 
@@ -112,6 +114,44 @@ def _immediate_as_completed(futures):
 
 
 class TestOcrFallbacks(unittest.TestCase):
+    @patch("polytext.converter.document_ocr_to_text.genai.Client")
+    @patch("concurrent.futures.as_completed", side_effect=_immediate_as_completed)
+    @patch("concurrent.futures.ThreadPoolExecutor", _ImmediateExecutor)
+    @patch("fitz.open")
+    def test_document_ocr_renders_pages_at_requested_dpi(
+        self,
+        mock_fitz_open,
+        _mock_as_completed,
+        mock_client_cls,
+    ):
+        page = _FakePage()
+        mock_fitz_open.return_value = _FakePdf([page])
+        mock_client_cls.return_value = _FakeClient()
+
+        converter = DocumentOCRToTextConverter(ocr_render_dpi=200)
+        converter.get_document_ocr("dummy.pdf")
+
+        self.assertEqual(page.pixmap_calls, [{"dpi": 200, "alpha": False}])
+
+    @patch("polytext.converter.document_ocr_to_text.genai.Client")
+    @patch("concurrent.futures.as_completed", side_effect=_immediate_as_completed)
+    @patch("concurrent.futures.ThreadPoolExecutor", _ImmediateExecutor)
+    @patch("fitz.open")
+    def test_document_ocr_preserves_default_pixmap_rendering(
+        self,
+        mock_fitz_open,
+        _mock_as_completed,
+        mock_client_cls,
+    ):
+        page = _FakePage()
+        mock_fitz_open.return_value = _FakePdf([page])
+        mock_client_cls.return_value = _FakeClient()
+
+        converter = DocumentOCRToTextConverter()
+        converter.get_document_ocr("dummy.pdf")
+
+        self.assertEqual(page.pixmap_calls, [{}])
+
     def test_default_ocr_max_output_tokens_is_8192(self):
         converter = OCRToTextConverter()
         self.assertEqual(converter.max_output_tokens, 8192)

{polytext-0.2.8b4 → polytext-0.2.8b5}/tests/test_ocr_image_descriptions.py

@@ -138,6 +138,23 @@ class TestOCRImageDescriptions(unittest.TestCase):
         self.assertIsInstance(document_ocr_loader, DocumentOCRLoader)
         self.assertFalse(document_ocr_loader.allow_partial_ocr_failures)
 
+    def test_forced_document_ocr_passes_custom_render_dpi(self):
+        loader = BaseLoader(
+            source="local",
+            force_ocr=True,
+            ocr_render_dpi=200,
+        )
+
+        document_ocr_loader = loader.init_loader_class(
+            input="/tmp/example.pdf",
+            storage_client={},
+            llm_api_key=None,
+            source="local",
+        )
+
+        self.assertIsInstance(document_ocr_loader, DocumentOCRLoader)
+        self.assertEqual(document_ocr_loader.ocr_render_dpi, 200)
+
     def test_google_ocr_converter_builds_augmented_markdown_prompt(self):
         converter = OCRToTextConverter(include_image_descriptions=True)