diff options
Diffstat (limited to 'src/server/chunker/pdf_chunker.py')
| -rw-r--r-- | src/server/chunker/pdf_chunker.py | 41 |
1 files changed, 22 insertions, 19 deletions
diff --git a/src/server/chunker/pdf_chunker.py b/src/server/chunker/pdf_chunker.py index 7a3244fbc..130987343 100644 --- a/src/server/chunker/pdf_chunker.py +++ b/src/server/chunker/pdf_chunker.py @@ -64,13 +64,15 @@ class ElementExtractor: A class that uses a YOLO model to extract tables and images from a PDF page. """ - def __init__(self, output_folder: str): + def __init__(self, output_folder: str, doc_id: str): """ Initializes the ElementExtractor with the output folder for saving images and the YOLO model. :param output_folder: Path to the folder where extracted elements will be saved. """ - self.output_folder = output_folder + self.doc_id = doc_id + self.output_folder = os.path.join(output_folder, doc_id) + os.makedirs(self.output_folder, exist_ok=True) self.model = YOLO('keremberke/yolov8m-table-extraction') # Load YOLO model for table extraction self.model.overrides['conf'] = 0.25 # Set confidence threshold for detection self.model.overrides['iou'] = 0.45 # Set Intersection over Union (IoU) threshold @@ -114,20 +116,18 @@ class ElementExtractor: # Save the full page with the red outline table_filename = f"table_page{page_num + 1}_{idx + 1}.png" + file_path_for_client = f"{self.doc_id}/{table_filename}" table_path = os.path.join(self.output_folder, table_filename) page_with_outline.save(table_path) - # Convert the full-page image with red outline to base64 - base64_data = self.image_to_base64(page_with_outline) - tables.append({ 'metadata': { "type": "table", "location": [x1 / img.width, y1 / img.height, x2 / img.width, y2 / img.height], - "file_path": table_path, + "file_path": file_path_for_client, "start_page": page_num, "end_page": page_num, - "base64_data": base64_data, + "base64_data": self.image_to_base64(page_with_outline) } }) @@ -173,21 +173,19 @@ class ElementExtractor: # Save the full page with the red outline image_filename = f"image_page{page_num + 1}_{img_index + 1}.png" + file_path_for_client = f"{self.doc_id}/{image_filename}" image_path = os.path.join(self.output_folder, image_filename) page_with_outline.save(image_path) - # Convert the full-page image with red outline to base64 - base64_data = self.image_to_base64(page_with_outline) - images.append({ 'metadata': { "type": "image", "location": [x1 / page.rect.width, y1 / page.rect.height, x2 / page.rect.width, y2 / page.rect.height], - "file_path": image_path, + "file_path": file_path_for_client, "start_page": page_num, "end_page": page_num, - "base64_data": base64_data, + "base64_data": self.image_to_base64(image) } }) @@ -269,7 +267,7 @@ class PDFChunker: The main class responsible for chunking PDF files into text and visual elements (tables/images). """ - def __init__(self, output_folder: str = "output", image_batch_size: int = 5) -> None: + def __init__(self, output_folder: str = "output", doc_id: str = '', image_batch_size: int = 5) -> None: """ Initializes the PDFChunker with an output folder and an element extractor for visual elements. @@ -279,7 +277,8 @@ class PDFChunker: self.client = Anthropic(api_key=os.getenv("ANTHROPIC_API_KEY")) # Initialize the Anthropic API client self.output_folder = output_folder self.image_batch_size = image_batch_size # Batch size for image processing - self.element_extractor = ElementExtractor(output_folder) # Initialize the element extractor + self.doc_id = doc_id # Add doc_id + self.element_extractor = ElementExtractor(output_folder, doc_id) async def chunk_pdf(self, file_data: bytes, file_name: str, doc_id: str, job_id: str) -> List[Dict[str, Any]]: """ @@ -364,6 +363,7 @@ class PDFChunker: for j, elem in enumerate(batch, start=1): if j in summaries: elem['metadata']['text'] = re.sub(r'^(Image|Table):\s*', '', summaries[j]) + elem['metadata']['base64_data'] = '' processed_elements.append(elem) progress = ((i // image_batch_size) + 1) / total_batches * 100 # Calculate progress @@ -629,10 +629,11 @@ class PDFChunker: return summaries - except Exception: - #print(f"Error in batch_summarize_images: {str(e)}") - #print("Returning placeholder summaries") - return {number: "Error: No summary available" for number in images} + except Exception as e: + # Print errors to stderr so they don't interfere with JSON output + print(json.dumps({"error": str(e)}), file=sys.stderr) + sys.stderr.flush() + class DocumentType(Enum): """ @@ -688,7 +689,7 @@ class Document: """ Process the document: extract chunks, embed them, and generate a summary. """ - pdf_chunker = PDFChunker(output_folder=self.output_folder) + pdf_chunker = PDFChunker(output_folder=self.output_folder, doc_id=self.doc_id) # Initialize PDFChunker self.chunks = asyncio.run(pdf_chunker.chunk_pdf(self.file_data, self.file_name, self.doc_id, self.job_id)) # Extract chunks self.num_pages = self._get_pdf_pages() # Get the number of pages in the document @@ -823,6 +824,8 @@ def main(): output_folder = sys.argv[4] # Get the output folder from arguments try: + os.makedirs(output_folder, exist_ok=True) + # Decode the base64 file data file_bytes = base64.b64decode(file_data) |