diff --git a/test_unstructured/partition/test_google_vision_requests.py b/test_unstructured/partition/test_google_vision_requests.py new file mode 100644 index 0000000000..f1db93ea9d --- /dev/null +++ b/test_unstructured/partition/test_google_vision_requests.py @@ -0,0 +1,40 @@ +from unittest.mock import Mock + +import pytest +from google.cloud.vision import AnnotateImageResponse, TextAnnotation +from PIL import Image + +from unstructured.partition.utils.ocr_models.google_vision_ocr import OCRAgentGoogleVision + + +def _agent(response): + agent = object.__new__(OCRAgentGoogleVision) + agent.language = "en" + agent.client = Mock() + agent.client.document_text_detection.return_value = response + return agent + + +def test_layout_elements_use_one_annotation_response_for_text_and_regions(): + response = AnnotateImageResponse(full_text_annotation=TextAnnotation(text="")) + agent = _agent(response) + + layout = agent.get_layout_elements_from_image(Image.new("RGB", (4, 4))) + + assert len(layout) == 0 + assert agent.client.document_text_detection.call_count == 1 + request = agent.client.document_text_detection.call_args.kwargs + assert list(request["image_context"].language_hints) == ["en"] + assert request["image"].content.startswith(b"\x89PNG") + + +@pytest.mark.parametrize( + "method", ["get_text_from_image", "get_layout_from_image", "get_layout_elements_from_image"] +) +def test_service_errors_are_not_silently_treated_as_empty_documents(method): + agent = _agent(AnnotateImageResponse(error={"code": 7, "message": "Permission denied"})) + + with pytest.raises(RuntimeError, match=r"Google Vision OCR failed \(7\): Permission denied"): + getattr(agent, method)(Image.new("RGB", (4, 4))) + + assert agent.client.document_text_detection.call_count == 1 diff --git a/unstructured/partition/utils/ocr_models/google_vision_ocr.py b/unstructured/partition/utils/ocr_models/google_vision_ocr.py index 852f973ba4..2fadf2c94c 100644 --- a/unstructured/partition/utils/ocr_models/google_vision_ocr.py +++ b/unstructured/partition/utils/ocr_models/google_vision_ocr.py @@ -34,26 +34,27 @@ def is_text_sorted(self) -> bool: return True def get_text_from_image(self, image: PILImage.Image) -> str: + return self._detect_document(image).text + + def _detect_document(self, image: PILImage.Image) -> TextAnnotation: + """Request OCR once and reject service errors before interpreting the annotation.""" image_context = ImageContext(language_hints=[self.language]) if self.language else None with BytesIO() as buffer: image.save(buffer, format="PNG") response = self.client.document_text_detection( image=Image(content=buffer.getvalue()), image_context=image_context ) + if response.error.code: + raise RuntimeError( + f"Google Vision OCR failed ({response.error.code}): {response.error.message}" + ) document = response.full_text_annotation assert isinstance(document, TextAnnotation) - return document.text + return document def get_layout_from_image(self, image: PILImage.Image) -> TextRegions: trace_logger.detail("Processing entire page OCR with Google Vision API...") - image_context = ImageContext(language_hints=[self.language]) if self.language else None - with BytesIO() as buffer: - image.save(buffer, format="PNG") - response = self.client.document_text_detection( - image=Image(content=buffer.getvalue()), image_context=image_context - ) - document = response.full_text_annotation - assert isinstance(document, TextAnnotation) + document = self._detect_document(image) regions = self._parse_regions(document) return regions @@ -62,15 +63,11 @@ def get_layout_elements_from_image(self, image: PILImage.Image) -> LayoutElement build_layout_elements_from_ocr_regions, ) - ocr_regions = self.get_layout_from_image( - image, - ) - ocr_text = self.get_text_from_image( - image, - ) + document = self._detect_document(image) + ocr_regions = self._parse_regions(document) return build_layout_elements_from_ocr_regions( ocr_regions=ocr_regions, - ocr_text=ocr_text, + ocr_text=document.text, group_by_ocr_text=False, )