xfmr-zem 0.2.5__py3-none-any.whl → 0.2.7__py3-none-any.whl

xfmr_zem/servers/ocr/parameters.yml

@@ -2,3 +2,6 @@
 extract_text:
   engine: "tesseract"
   model_id: null
+  scanned_threshold: 50
+  zoom: 2.0
+  temp_dir: "/tmp"

xfmr_zem/servers/ocr/server.py

@@ -1,43 +1,128 @@
+import os
 import pandas as pd
 from xfmr_zem.server import ZemServer
 from xfmr_zem.servers.ocr.engines import OCREngineFactory
 from loguru import logger
+from PIL import Image
+import io
 
 # Initialize ZemServer for OCR
 mcp = ZemServer("ocr")
 
+def extract_pdf_pages(
+    file_path: str, 
+    engine: str, 
+    ocr_engine, 
+    scanned_threshold: int = 50, 
+    zoom: float = 2.0, 
+    temp_dir: str = "/tmp"
+):
+    """Helper to process PDF pages with optional OCR for scanned content."""
+    import fitz  # PyMuPDF
+    
+    results = []
+    doc = fitz.open(file_path)
+    
+    # Ensure temp_dir exists
+    os.makedirs(temp_dir, exist_ok=True)
+    
+    for page_num in range(len(doc)):
+        page = doc[page_num]
+        text = page.get_text().strip()
+        
+        # Determine if we need to OCR (Strategy: text is too short or empty)
+        is_scanned = len(text) < scanned_threshold
+        
+        if is_scanned:
+            logger.info(f"Page {page_num + 1} appears scanned (text length: {len(text)}). Running OCR with {engine}...")
+            # Render page to image for OCR
+            pix = page.get_pixmap(matrix=fitz.Matrix(zoom, zoom))
+            img_data = pix.tobytes("png")
+            img = Image.open(io.BytesIO(img_data))
+            
+            # Temporary save for engine compatibility (engines expect path)
+            temp_path = os.path.join(temp_dir, f"ocr_page_{os.getpid()}_{page_num}.png")
+            img.save(temp_path)
+            
+            try:
+                ocr_result = ocr_engine.process(temp_path)
+                final_text = ocr_result["text"]
+                source = f"{engine}_ocr"
+            finally:
+                if os.path.exists(temp_path):
+                    os.remove(temp_path)
+        else:
+            final_text = text
+            source = "digital_pdf"
+            
+        results.append({
+            "text": final_text,
+            "page": page_num + 1,
+            "engine": source,
+            "metadata": {"file": file_path, "is_scanned": is_scanned}
+        })
+        
+    doc.close()
+    return results
+
 @mcp.tool()
-async def extract_text(file_path: str, engine: str = None, model_id: str = None) -> pd.DataFrame:
+async def extract_text(
+    file_path: str, 
+    engine: str = "tesseract", 
+    model_id: str = None,
+    scanned_threshold: int = 50,
+    zoom: float = 2.0,
+    temp_dir: str = "/tmp"
+) -> pd.DataFrame:
     """
-    Extracts text from an image using the specified OCR engine.
+    Extracts text from an image or PDF using the specified OCR engine.
+    For PDFs, it will automatically handle scanned pages using the OCR engine.
     
     Args:
-        file_path: Path to the image file.
+        file_path: Path to the image or PDF file.
         engine: The OCR engine to use ("tesseract", "paddle", "huggingface", "viet"). Defaults to "tesseract".
-        model_id: Optional model ID for the 'huggingface' engine (e.g., "Qwen/Qwen2-VL-2B-Instruct").
+        model_id: Optional model ID for the 'huggingface' engine.
+        scanned_threshold: Min characters required to skip OCR on PDF page. Defaults to 50.
+        zoom: Rendering zoom factor for scanned PDF pages. Defaults to 2.0.
+        temp_dir: Directory for temporary page images. Defaults to "/tmp".
     """
-    logger.info(f"OCR Extraction: {file_path} using {engine} (model: {model_id})")
+    logger.info(f"OCR Extraction: {file_path} using {engine} (scanned_threshold={scanned_threshold}, zoom={zoom})")
     
+    if not os.path.exists(file_path):
+        raise FileNotFoundError(f"File not found: {file_path}")
+
     try:
-        # Get engine from factory (SOLID Strategy Pattern)
+        # Get engine from factory
         ocr_engine = OCREngineFactory.get_engine(engine, model_id=model_id)
         
-        # Process image
-        result = ocr_engine.process(file_path)
-        
-        # Structure as a single-row DataFrame for Zem compatibility
-        # We wrap in a list to ensure pandas creates a row
-        df = pd.DataFrame([{
-            "text": result["text"],
-            "engine": result["engine"],
-            "metadata": result["metadata"]
-        }])
+        # Handle PDF vs Image
+        if file_path.lower().endswith(".pdf"):
+            logger.info(f"Processing PDF file: {file_path}")
+            data = extract_pdf_pages(
+                file_path, 
+                engine, 
+                ocr_engine, 
+                scanned_threshold=scanned_threshold, 
+                zoom=zoom, 
+                temp_dir=temp_dir
+            )
+            df = pd.DataFrame(data)
+        else:
+            # Process image
+            result = ocr_engine.process(file_path)
+            df = pd.DataFrame([{
+                "text": result["text"],
+                "engine": result["engine"],
+                "metadata": result["metadata"]
+            }])
         
-        logger.info(f"Successfully extracted text using {engine}")
+        logger.info(f"Successfully extracted text from {file_path}")
         return df.to_dict(orient="records")
         
     except Exception as e:
         logger.error(f"OCR Error with {engine}: {e}")
+        import traceback
+        logger.error(traceback.format_exc())
         raise RuntimeError(f"OCR failed: {str(e)}")
 
 if __name__ == "__main__":

{xfmr_zem-0.2.5.dist-info → xfmr_zem-0.2.7.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: xfmr-zem
-Version: 0.2.5
+Version: 0.2.7
 Summary: Zem: Unified Data Pipeline Framework (ZenML + NeMo Curator + DataJuicer) for multi-domain processing
 Project-URL: Homepage, https://github.com/OAI-Labs/xfmr-zem
 Project-URL: Repository, https://github.com/OAI-Labs/xfmr-zem
@@ -51,6 +51,7 @@ Requires-Dist: paddlepaddle>=2.6.0; extra == 'ocr'
 Requires-Dist: pdfplumber>=0.11.0; extra == 'ocr'
 Requires-Dist: pillow>=10.0.0; extra == 'ocr'
 Requires-Dist: pyclipper; extra == 'ocr'
+Requires-Dist: pymupdf>=1.23.0; extra == 'ocr'
 Requires-Dist: pytesseract>=0.3.10; extra == 'ocr'
 Requires-Dist: ruamel-yaml>=0.17.0; extra == 'ocr'
 Requires-Dist: shapely; extra == 'ocr'

{xfmr_zem-0.2.5.dist-info → xfmr_zem-0.2.7.dist-info}/RECORD

@@ -17,8 +17,8 @@ xfmr_zem/servers/nemo_curator/parameters.yml,sha256=EGEzo0heI-ajkwFFy3xxq_YD7cXU
 xfmr_zem/servers/nemo_curator/server.py,sha256=zcHoSwxxoK_rMaDIAbEy1s8qfdp68Ue4B-XBcjGxQak,3848
 xfmr_zem/servers/ocr/engines.py,sha256=zScn4Qjxbpl2nB8UXEf3kd9l8z84TEwGs6bV5ka8Lks,10295
 xfmr_zem/servers/ocr/install_models.py,sha256=t02zpoy8djVhITOLEaRJ2mjiMrFfA9H6fpeHD3hXuio,2135
-xfmr_zem/servers/ocr/parameters.yml,sha256=04v59-6QXwN6XEpnHLc5pz6iTgNBDhloHtCCjHr8YRA,89
-xfmr_zem/servers/ocr/server.py,sha256=Yef1CYJR5RDH38jffgbcpGE-1VZLaU4w1wi572oPZcY,1571
+xfmr_zem/servers/ocr/parameters.yml,sha256=UTMwtTu0Eeit0tFkYcZOxpuzD78UBlpONXZIx6STYwc,144
+xfmr_zem/servers/ocr/server.py,sha256=eJtQnMVBFX6PLZMxZITNlNEXGarjsvkz003-uT1iIo0,4369
 xfmr_zem/servers/ocr/deepdoc_vietocr/__init__.py,sha256=XJE7RnOu5oo5p902HPWPDBd7FhVQXetmnr2-kWEG0nI,2419
 xfmr_zem/servers/ocr/deepdoc_vietocr/implementations.py,sha256=79fYr76fx8yZda3HaFcK1d5G-4sDVf1JFHNW_OBQAk8,47348
 xfmr_zem/servers/ocr/deepdoc_vietocr/layout_recognizer.py,sha256=7BeLHzf9FQUkkHMb5jDpggruJmfXVMU78MF_EeZ9PG4,10462
@@ -51,8 +51,8 @@ xfmr_zem/servers/sinks/parameters.yml,sha256=9HAnv84Utw2qWsVZH8uOjVE62lnAKBkzv4P
 xfmr_zem/servers/sinks/server.py,sha256=jI_r4sq_U_avNwF1PiE0alpaDrYpzOI-qPeLU7hgHP0,1589
 xfmr_zem/servers/unstructured/parameters.yml,sha256=N31cmc56GTr3rkVhbni4yOpbnHISReN8f-KnRZTDbBc,118
 xfmr_zem/servers/unstructured/server.py,sha256=0XmXWMAUNEJboX-J4bn_8EBUfMHIqu_ylNC_s9YOZdk,1996
-xfmr_zem-0.2.5.dist-info/METADATA,sha256=QxGjfN7Y4zZOGmcDwohYh9HcFj2JDw7XmKyC4400z6M,6332
-xfmr_zem-0.2.5.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
-xfmr_zem-0.2.5.dist-info/entry_points.txt,sha256=uxs-IXFxpSakHivpFN3mEr13cz-z-0vkeSF_4dEBMa4,65
-xfmr_zem-0.2.5.dist-info/licenses/LICENSE,sha256=kf_ILr0zLkSy5-EBu0VF2PGaOykYo83z3UijI-bZeAE,11342
-xfmr_zem-0.2.5.dist-info/RECORD,,
+xfmr_zem-0.2.7.dist-info/METADATA,sha256=Iv77eb-eHw6rdJhG1LfoNY4Hf9I7oFlIsx1K3K7_sH0,6379
+xfmr_zem-0.2.7.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
+xfmr_zem-0.2.7.dist-info/entry_points.txt,sha256=uxs-IXFxpSakHivpFN3mEr13cz-z-0vkeSF_4dEBMa4,65
+xfmr_zem-0.2.7.dist-info/licenses/LICENSE,sha256=kf_ILr0zLkSy5-EBu0VF2PGaOykYo83z3UijI-bZeAE,11342
+xfmr_zem-0.2.7.dist-info/RECORD,,